├── .gitignore ├── LICENSE ├── Makefile ├── Makefile.common ├── PKGBUILD ├── README.md ├── blkio.c ├── blkio.h ├── ctypes.h ├── kv.c ├── kv.h ├── lib.c ├── lib.h ├── remixdb.h ├── remixdb.strip ├── sotest.c ├── sst.c ├── sst.h ├── wh.c ├── wh.h ├── xdb.c ├── xdb.h ├── xdb.py ├── xdbcow.sh ├── xdbdemo.c ├── xdbexit.c └── xdbtest.c /.gitignore: -------------------------------------------------------------------------------- 1 | *.out 2 | *.so 3 | *.o 4 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /Makefile: -------------------------------------------------------------------------------- 1 | # Makefile 2 | # rules (always with .out) 3 | # SRC-X.out += abc # extra source: abc.c 4 | # MOD-X.out += abc # extra module: abc.c abc.h 5 | # ASM-X.out += abc # extra assembly: abc.S 6 | # DEP-X.out += abc # extra dependency: abc 7 | # FLG-X.out += -finline # extra flags 8 | # LIB-X.out += abc # extra -labc options 9 | 10 | # X.out : xyz.h xyz.c # for extra dependences that are to be compiled/linked. 11 | 12 | # X => X.out 13 | TARGETS += xdbdemo xdbtest xdbexit 14 | # X => X.c only 15 | SOURCES += 16 | SOURCES += $(EXTRASRC) 17 | # X => X.S only 18 | ASSMBLY += 19 | ASSMBLY += $(EXTRAASM) 20 | # X => X.c X.h 21 | MODULES += lib kv wh blkio sst xdb 22 | MODULES += $(EXTRAMOD) 23 | # X => X.h 24 | HEADERS += ctypes 25 | HEADERS += $(EXTRAHDR) 26 | 27 | # EXTERNSRC/EXTERNDEP do not belong to this repo. 28 | # extern-src will be linked 29 | EXTERNSRC += 30 | # extern-dep will not be linked 31 | EXTERNDEP += 32 | 33 | FLG += 34 | LIB += m 35 | 36 | 37 | #### all 38 | .PHONY : all 39 | all : bin libremixdb.so sotest.out 40 | 41 | libremixdb.so : Makefile Makefile.common lib.h kv.h wh.h blkio.h sst.h xdb.h lib.c kv.c wh.c blkio.c sst.c xdb.c 42 | $(eval ALLFLG := $(CSTD) $(EXTRA) $(FLG) -shared -fPIC) 43 | $(eval ALLLIB := $(addprefix -l,$(LIB) $(LIB-$@))) 44 | $(CCC) $(ALLFLG) -o $@ lib.c kv.c wh.c blkio.c sst.c xdb.c $(ALLLIB) 45 | strip --strip-all --discard-all @remixdb.strip $@ 46 | 47 | sotest.out : sotest.c Makefile Makefile.common libremixdb.so remixdb.h 48 | $(eval ALLFLG := $(CSTD) $(EXTRA) $(FLG)) 49 | $(CCC) $(ALLFLG) -o $@ $< -L . -lremixdb 50 | @echo "$(shell $(TPUT) setaf 4)Now run $ LD_LIBRARY_PATH=. ./sotest.out$(shell $(TPUT) sgr0)" 51 | 52 | .PHONY : install 53 | install : libremixdb.so remixdb.h 54 | install -D --mode=0755 libremixdb.so $(PREFIX)/lib/libremixdb.so 55 | install -D --mode=0644 remixdb.h $(PREFIX)/usr/include/remixdb.h 56 | 57 | # append common rules (have to do it here) 58 | include Makefile.common 59 | -------------------------------------------------------------------------------- /Makefile.common: -------------------------------------------------------------------------------- 1 | #usage: include Makefile.common at the end of your Makefile 2 | 3 | # no builtin rules/vars (CC, CXX, etc. are still defined but will be empty) 4 | MAKEFLAGS += -r -R 5 | 6 | HDR = $(addsuffix .h,$(MODULES) $(HEADERS)) 7 | SRC = $(addsuffix .c,$(MODULES) $(SOURCES)) 8 | ASM = $(addsuffix .S,$(ASSMBLY)) 9 | OBJ = $(addsuffix .o,$(MODULES) $(SOURCES) $(ASSEMBLY)) 10 | DEP = Makefile.common Makefile $(HDR) $(EXTERNDEP) $(EXTERNSRC) 11 | BIN = $(addsuffix .out,$(TARGETS)) 12 | DIS = $(addsuffix .dis,$(TARGETS)) 13 | 14 | # clang: 15 | # EXTRA="-Rpass=loop-vectorize" # IDs loops that were successfully V-ed 16 | # EXTRA="-Rpass-missed=loop-vectorize" # IDs loops that failed V 17 | # EXTRA="-Rpass-analysis=loop-vectorize" # IDs the statements that caused V to fail 18 | # EXTRA="-Rpass=\ *" # remarks for all passes 19 | # other passes: https://llvm.org/docs/Passes.html 20 | 21 | O ?= rg 22 | 23 | # predefined OPT: make O={rg,r,0g,3g,p,0s,3s,cov,mc,hc,wn,stk} 24 | ifeq ($O,rg) # make O=rg 25 | OPT ?= -DNDEBUG -g3 -O3 -flto -fno-stack-protector 26 | else ifeq ($O,r) # make O=r (for release) 27 | OPT ?= -DNDEBUG -O3 -flto -fno-stack-protector 28 | else ifeq ($O,ns) # make O=ns (no signal handlers) 29 | OPT ?= -DNDEBUG -O3 -flto -fno-stack-protector -DNOSIGNAL 30 | else ifeq ($O,0g) # make O=0g 31 | OPT ?= -g3 -O0 -fno-inline 32 | else ifeq ($O,2g) # make O=2g 33 | OPT ?= -g3 -O2 34 | else ifeq ($O,3g) # make O=3g 35 | OPT ?= -g3 -O3 -flto -fno-inline 36 | else ifeq ($O,p) # make O=p (profiling: rg+noinline) 37 | OPT ?= -DNDEBUG -g3 -O3 -flto -fno-stack-protector -fno-inline 38 | else ifeq ($O,0s) # make O=0s (address sanitizer) 39 | OPT ?= -g3 -O0 -fno-inline -fsanitize=address -fno-omit-frame-pointer -DHEAPCHECKING 40 | else ifeq ($O,3s) # make O=3s (address sanitizer) 41 | OPT ?= -g3 -O3 -fno-inline -fsanitize=address -fno-omit-frame-pointer -DHEAPCHECKING 42 | else ifeq ($O,t) # make O=0t (thread sanitizer) 43 | OPT ?= -g3 -O1 -fno-inline -fsanitize=thread -fno-stack-protector 44 | else ifeq ($O,cov) # make O=cov (for gcov) 45 | OPT ?= -g3 -DNDEBUG -O0 --coverage 46 | CCC = gcc 47 | else ifeq ($O,mc) # make O=mc (for valgrind memcheck) 48 | OPT ?= -g3 -O1 -fno-inline -DHEAPCHECKING 49 | ARCH ?= broadwell 50 | else ifeq ($O,hc) # make O=hc (for gperftools heapcheck) 51 | OPT ?= -g3 -O1 -fno-inline 52 | LIB += tcmalloc 53 | else ifeq ($O,wn) # more warning 54 | OPT ?= -g3 -O3 -Wvla -Wformat=2 -Wconversion -Wstrict-prototypes -Wmissing-prototypes 55 | else ifeq ($O,stk) # check stack usage with gcc 56 | OPT ?= -g3 -O3 -DNDEBUG -fstack-usage 57 | CCC = gcc 58 | endif 59 | 60 | # malloc: g:glibc, t:tcmalloc, j:jemalloc 61 | M ?= g 62 | 63 | ifeq ($M,t) 64 | LIB += tcmalloc 65 | FLG += -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free 66 | else ifeq ($M,j) 67 | LIB += jemalloc 68 | endif 69 | 70 | UNAME_S := $(shell uname -s) 71 | ifeq ($(UNAME_S),Linux) 72 | CHECK_S := -D__linux__ 73 | LIB += rt 74 | else ifeq ($(UNAME_S),FreeBSD) 75 | CHECK_S := -D__FreeBSD__ 76 | FLG += -I/usr/local/include -L/usr/local/lib 77 | LIB += rt 78 | LIB += execinfo 79 | TPUT := /usr/local/bin/tput 80 | else ifeq ($(UNAME_S),Darwin) 81 | CHECK_S := -D__APPLE__ -D__MACH__ 82 | # do nothing 83 | else 84 | $(error "Supported Platforms: Linux, FreeBSD, Darwin") 85 | endif 86 | TPUT ?= tput 87 | 88 | CCC ?= clang 89 | CSTD = -std=gnu18 90 | XCC ?= clang++ 91 | XSTD = -std=gnu++17 92 | 93 | UNAME_M := $(shell uname -m) 94 | ifeq ($(UNAME_M),aarch64) # "native" does not work for clang@aarch64 95 | CHECK_M := -D__aarch64__ 96 | ARCH ?= armv8-a+crc 97 | else ifeq ($(UNAME_M),arm64) # "native" does not work for clang@aarch64 98 | CHECK_M := -D__aarch64__ 99 | ARCH ?= armv8-a+crc 100 | else ifeq ($(UNAME_M),x86_64) 101 | CHECK_M := -D__x86_64__ 102 | ARCH ?= native 103 | else ifeq ($(UNAME_M),amd64) # freebsd 104 | CHECK_M := -D__x86_64__ 105 | ARCH ?= native 106 | else 107 | $(error "Supported Platforms: aarch64, x86_64") 108 | endif 109 | 110 | TUNE ?= native 111 | 112 | NBI += memcpy memmove memcmp 113 | 114 | # minimal requirement on x86_64: -march=nehalem 115 | # minimal requirement on aarch64: -march=armv8-a+crc 116 | FLG += -march=$(ARCH) -mtune=$(TUNE) 117 | FLG += -pthread -Wall -Wextra -Wshadow #-Weverything 118 | FLG += $(addprefix -fno-builtin-,$(NBI)) 119 | FLG += $(OPT) 120 | 121 | ifneq ($(shell $(CCC) --version 2>/dev/null | grep clang),) 122 | FLG += -ferror-limit=3 123 | CCCTYPE := clang 124 | else ifneq ($(shell $(CCC) --version 2>/dev/null | grep gcc),) 125 | FLG += -fmax-errors=3 126 | FLG += -Wno-unknown-pragmas 127 | CCCTYPE := gcc 128 | else 129 | $(error "Supported Compilers: clang, gcc") 130 | endif 131 | 132 | ifeq ($(CCCTYPE),clang) 133 | CCINST = /usr/lib/clang/$(shell $(CCC) --version 2>/dev/null | awk '/^clang/ { print $$3 }') 134 | CCINC = $(CCINST)/include 135 | else ifeq ($(CCCTYPE),gcc) 136 | CCINST = /usr/lib/gcc/$(shell $(CCC) -dumpmachine)/$(shell $(CCC) -dumpversion) 137 | CCINC = $(CCINST)/include $(CCINST)/include-fixed 138 | endif 139 | CCINC = /usr/include /usr/local/include 140 | 141 | ifneq ($(shell find $(CCINC) -name backtrace-supported.h 2>/dev/null),) 142 | LIB += backtrace 143 | FLG += -DBACKTRACE 144 | endif 145 | 146 | ifneq ($(shell find $(CCINC) -name liburing.h 2>/dev/null),) 147 | LIB += uring 148 | FLG += -DLIBURING 149 | endif 150 | 151 | 152 | uniq = $(if $1,$(firstword $1) $(call uniq,$(filter-out $(firstword $1),$1))) 153 | magentatxt := $(shell $(TPUT) setaf 5) 154 | greentxt := $(shell $(TPUT) setaf 2) 155 | bluetxt := $(shell $(TPUT) setaf 4) 156 | normaltxt := $(shell $(TPUT) sgr0) 157 | 158 | .PHONY : bin dis def clean cleanx check tags 159 | 160 | bin : $(BIN) 161 | dis : $(DIS) bin 162 | .DEFAULT_GOAL = bin 163 | .SECONDEXPANSION: 164 | 165 | ifeq ($(J),o) 166 | # DANGER. Don't use unless it works! 167 | # build from .o files but target-specific flags are missing in %.o : %.x 168 | %.out : %.o $(OBJ) $$(addsuffix .o,$$(SRC-$$@) $$(MOD-$$@) $$(ASM-$$@)) 169 | $(eval ALLFLG := $(CSTD) $(EXTRA) $(FLG) $(FLG-$*) $(FLG-$*.o) $(FLG-$@) -rdynamic) 170 | $(eval ALLLIB := $(addprefix -l,$(LIB) $(LIB-$@))) 171 | $(CCC) $(ALLFLG) -o $@ $^ $(ALLLIB) 172 | # 173 | else # default: all-in-one command 174 | %.out : %.c $(SRC) $(ASM) $(DEP) $$(DEP-$$@) $$(addsuffix .c,$$(SRC-$$@) $$(MOD-$$@)) $$(addsuffix .h,$$(HDR-$$@) $$(MOD-$$@)) $$(addsuffix .S,$$(ASM-$$@)) 175 | $(eval ALLSRC := $(SRC) $(addsuffix .c,$(SRC-$@) $(MOD-$@)) $(ASM) $(addsuffix .S,$(ASM-$@))) 176 | $(eval UNIQSRC := $(call uniq,$(ALLSRC))) 177 | $(eval ALLFLG := $(CSTD) $(EXTRA) $(FLG) $(FLG-$@) -rdynamic) 178 | $(eval ALLLIB := $(addprefix -l,$(LIB) $(LIB-$@))) 179 | @printf '$(bluetxt)$@$(magentatxt) <= $(greentxt)$< $(UNIQSRC)$(normaltxt)\n' 180 | $(CCC) $(ALLFLG) -o $@ $< $(UNIQSRC) $(ALLLIB) 181 | # 182 | endif 183 | 184 | 185 | %.dis : %.out 186 | objdump -SlwtC $< 1>$@ 2>/dev/null 187 | 188 | %.o : %.cc $(DEP) $$(DEP-$$@) $$(addsuffix .h,$$(HDR-$$@) $$(MOD-$$@)) 189 | $(XCC) $(XSTD) $(EXTRA) $(FLG) $(FLG-$*) $(FLG-$@) -o $@ -c $< 190 | 191 | %.o : %.c $(DEP) $$(DEP-$$@) $$(addsuffix .h,$$(HDR-$$@) $$(MOD-$$@)) 192 | $(CCC) $(CSTD) $(EXTRA) $(FLG) $(FLG-$*) $(FLG-$@) -o $@ -c $< 193 | 194 | %.o : %.S $(DEP) $$(DEP-$$@) $$(addsuffix .h,$$(HDR-$$@) $$(MOD-$$@)) 195 | $(CCC) $(CSTD) $(EXTRA) $(FLG) $(FLG-$*) $(FLG-$@) -o $@ -c $< 196 | 197 | %.s : %.c $(DEP) $$(DEP-$$@) $$(addsuffix .h,$$(HDR-$$@) $$(MOD-$$@)) 198 | $(CCC) $(CSTD) $(EXTRA) $(FLG) $(FLG-$*) $(FLG-$*.o) -S -o $@ -c $< 199 | 200 | def : 201 | $(CCC) $(FLG) -dM -E - 2 | 3 | pkgname=remixdb 4 | pkgver=0.1 5 | pkgrel=1 6 | pkgdesc="RemixDB Embedded Key-Value Store" 7 | arch=('x86_64') 8 | url="https://github.com/wuxb45/remixdb" 9 | license=('GPL3.0') 10 | depends=('glibc' 'liburing') 11 | provides=('libremixdb.so') 12 | source=("git+https://github.com/wuxb45/remixdb") 13 | sha512sums=('SKIP') 14 | validpgpkeys=('SKIP') 15 | 16 | build() { 17 | cd "$pkgname" 18 | make O=r libremixdb.so 19 | } 20 | 21 | package() { 22 | cd "$pkgname" 23 | make PREFIX="$pkgdir" install 24 | } 25 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # REMIX and RemixDB 2 | 3 | The REMIX data structure was introduced in paper ["REMIX: Efficient Range Query for LSM-trees", FAST'21](https://www.usenix.org/conference/fast21/presentation/zhong). 4 | 5 | This repository maintains a reference implementation of the REMIX index data structure, 6 | as well as a thread-safe embedded key-value store implementation, namely RemixDB. 7 | It compiles on recent Linux/FreeBSD/MacOS and supports x86\_64 and AArch64 CPUs. 8 | 9 | This code repository is being actively maintained and contains optimizations beyong the original RemixDB implementation. 10 | 11 | # News 12 | 13 | * An experimental Python API is available in `xdb.py`. See the end of `xdb.py` for examples. 14 | 15 | * RemixDB now provides `xdb_merge` for atomic read-modify-write operations. 16 | 17 | * Two optimizations have been added to boost compaction and point query performance (see below). 18 | 19 | # Optimizations 20 | 21 | ## Optimization 1: Minimizing REMIX (Re-)Building Cost 22 | 23 | This implementation employs an optimization to minimize the REMIX building cost. 24 | This optimization improves the throughput by 2x (0.96MOPS vs. 0.50MOPS) in a random-write experiment, compared to the implementation described in the REMIX paper. 25 | Configuration: klen=16; vlen=120; 2.02 billion KVs; 256GB valid KV data; single-threaded loading in random order; no compression. 26 | 27 | When creating a new table file, RemixDB can create a copy of all the keys in the table. 28 | Specificially, it encodes all the keys (without values) in sorted order using prefix compression, which creates a *Compressed Keys Block* (*CKB*). 29 | The CKB is stored at the end of the table file. 30 | This feature can be freely turned on and off. There is no compatibility issue when tables with and without the CKB are used together. 31 | 32 | When creating a new REMIX, the building process will check if every input table contains a CKB. 33 | If true, the process will build the new REMIX using these CKBs. It also leverages the existing REMIX to avoid unncecssary key comparisons. 34 | In this way, the new REMIX will be created by reading the old REMIX and the CKBs, without accessing the key-value data blocks of the table files. 35 | 36 | In a running system the old REMIX structures are usually cache-resident. 37 | The CKBs are only used for REMIX building, which are read into memory in batch, and discarded once the building is finished. 38 | 39 | A CKB is often much smaller than the original key-value data block, unless the system manages huge keys with small values. 40 | Suppose the average CKB size is 10% of the average key-value data block size, 41 | this optimization trades 10% more write I/O and storage space usage for a 90% reduction of read I/O during REMIX building. 42 | 43 | `remixdb_open` opens/creates a remixdb with the optimization turned on. Each newly created sstable will have the CKB. 44 | You should use `remixdb_open` unless it's absolutely necessary to save a little bit disk space. 45 | `remixdb_open_compact` opens a remixdb with the optimization turned off. Each newly created sstable will not contain a CKB. 46 | A store created by one of these functions can be safely opened by the other function. 47 | 48 | ## Optimization 2: Improving Point Query with Hash Tags 49 | 50 | A point query in the original RemixDB performs binary search in a segment, which takes about five key comparisons and can cost multiple I/Os. 51 | The current implementation provides a new option, named `tags` (the last argument of `remixdb_open`). 52 | 53 | With this option enabled, every new REMIX will store an array of 8-bit hash tags. Each tag corresponds to a key managed by the REMIX. 54 | A point query (GET/PROBE) will first locate the target segment as usual. 55 | Then it will check the tags to find candidate keys for full-key matching without using binary search in the segment. 56 | With 8-bit tags and at most 32 keys in a segment, a point query takes about 1.06 key comparisons if the key is found, 57 | and about 0.12 key comparisons if the key does not exist. 58 | 59 | TODO: the tags can also be used to speed up iterator seeks with existing keys. 60 | 61 | # Limitations of the Current Implementation 62 | 63 | * *KV size*: The maximum key+value size is capped at 65500 bytes. 64 | This roughly corresponds to the 64KB block size limit. 65 | TODO: store huge KV pairs in a separate file and store the file address of the KV pair in RemixDB. 66 | 67 | # Configuration and Tuning 68 | 69 | ## CPU affinity 70 | RemixDB employs background threads to perform asynchronous compaction. 71 | When possible (on Linux or FreeBSD), these threads are pinned on specific cores for efficiency. 72 | To avoid interferences with the foreground threads, it is necessary to separate the cores used by different threads. 73 | By default, RemixDB pins 4 compaction threads on the last four cores of the current process's affinity list. 74 | For example, on a machine with two 10-core processors, cores 0,2,4,...,16,18 belong to numa node 0, 75 | and the rest cores belong to numa node 1. 76 | The default behavior is to use the cores from 16 to 19, which is a suboptimal setup. 77 | To avoid the performance penalty, one should use `numactl` to specify the cpu affinity. 78 | 79 | ``` 80 | $ numactl -C 0,2,4,6,8 ./xdbdemo.out # compaction threads on 2,4,6,8 81 | 82 | $ numactl -C 0,2,4,6,8,10,12,14 ./xdbtest.out 256 256 18 18 100 # user threads on 0,2,4,6; compaction threads on 8,10,12,14 83 | ``` 84 | 85 | The worker threads affinity can also be explicitly specified using `xdb_open`. 86 | 87 | ## Maximum number of open files 88 | The current implementation keeps every table file open at run time. 89 | This requires a large `nofile` in `/etc/security/limits.conf`. 90 | For example, add `* - nofile 100000` to `limits.conf`, reboot/relogin, and double-check with `ulimit -n`. 91 | 92 | ## Maximum Table File Size 93 | `MSSTZ_NBLKS` (sst.c) controls the maximum number of 4KB blocks in an SST file. The default number is 20400. 94 | The maximum value is 65520 (256MB data blocks, plus metadata). 95 | 96 | ## Hugepages 97 | 98 | Configuring huge pages can effectively improve RemixDB's performance. 99 | Usually a few hundred 2MB hugepages would be sufficient for memory allocation in MemTables. 100 | The block cache automatically detects and uses 1GB huge pages when available (otherwise, fall back to 2MB pages, and then 4KB pages). 101 | 4x 1GB huge pages should be configured if you set cache size to 4GB. 102 | 103 | # Getting Started 104 | 105 | RemixDB by default uses `liburing` (`io_uring`) and thus requires a Linux kernel >= 5.1. 106 | It also works with POSIX AIO on all the supported platforms but the performance can be negatively affected. 107 | 108 | `clang` is the default compiler. It usually produces faster code than GCC. To use GCC: 109 | 110 | $ make CCC=gcc 111 | 112 | `jemalloc` is highly recommended. If jemalloc is available and you prefer to use it, use `M=j` with `make`: 113 | 114 | $ make M=j 115 | 116 | Similarly, `tcmalloc` can be linked with `M=t`. 117 | 118 | The `xdbdemo.c` contains sample code that uses the `remixdb_*` functions. 119 | These functions present a clean programming interface without using special data types or structures. 120 | 121 | ## xdbdemo 122 | To compile and run the demo code: 123 | 124 | $ make M=j xdbdemo.out 125 | $ ./xdbdemo.out 126 | 127 | ## xdbtest 128 | 129 | `xdbtest` is a stress test program that uses the `remixdb_*` functions. 130 | It trys to use all the available cores on the affinity list, which can lead to mediocre performance. 131 | You should use numactl to specify what cores are available for the tester threads. 132 | Suppose you have eight cores (0...7) in total, the best practice is to let the testers to run on the first four cores and assign the last four cores to the compaction threads. The following examples use this configuration. 133 | 134 | Run with a 4GB block cache, 4GB MemTables, and a dataset with 32 million KVs (2^25), performing 1 million updates in each round (2^20): 135 | 136 | $ make M=j xdbtest.out 137 | $ numactl -N 0 ./xdbtest.out /tmp/xdbtest 4096 4096 25 20 100 138 | 139 | To run with smaller memory footprint (a 256MB block cache, 256MB Memtables, and 1 million KVs): 140 | 141 | $ numactl -N 0 ./xdbtest.out /tmp/xdbtest 256 256 20 20 100 142 | 143 | This setup consumes up to 850MB memory (RSS) and 1.8GB space in /tmp/xdbtest. 144 | 145 | A first run of xdbtest.out should always show stale=0. 146 | If you run it again without deleting `/tmp/xdbtest`, 147 | it will show non-zero stale numbers at the beginning but it will quickly drop and eventually reach zero. 148 | 149 | ## xdbexit 150 | 151 | `xdbexit` is a simple program testing crash-recovery. 152 | It inserts some new keys and calls `remixdb_sync()` to make all buffered data persist in the WAL. 153 | Then it immediately calls `exit()` without doing any clean-up. 154 | Run it repeatedly. In each run it should show that all the previously inserted KVs are found. 155 | 156 | Run with a small footprint: 157 | 158 | $ for i in $(seq 1 30); do ./xdbexit.out ./dbdir 256 256; done 159 | 160 | Run with in a regular-sized setup: 161 | 162 | $ for i in $(seq 1 30); do ./xdbexit.out ./dbdir 4096 4096; done 163 | 164 | ## libremixdb.so 165 | 166 | To use remixdb as a shared library, run `make libremixdb.so` and `make install`. 167 | A PKGBUILD (for Archlinux's pacman) is included as an example packaging script. 168 | -------------------------------------------------------------------------------- /blkio.c: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #define _GNU_SOURCE 7 | 8 | // include {{{ 9 | #include "lib.h" 10 | #include "ctypes.h" 11 | #include 12 | #include 13 | #include 14 | #include "blkio.h" 15 | 16 | // }}} include 17 | 18 | // wring {{{ 19 | // wring is NOT thread-safe 20 | struct wring { 21 | void * mem; 22 | void * head; 23 | size_t memsz; 24 | u32 iosz; 25 | int fd; 26 | #if defined(LIBURING) 27 | u32 batch; // submit when pending >= batch 28 | u32 nring; // number of pending + submitted 29 | u32 pending; // number of unsubmitted sqes 30 | bool fixed_file; // use this fd for write (0 for registered file) 31 | bool fixed_mem; // use write_fixed() 32 | bool padding[2]; 33 | struct io_uring uring; 34 | #else 35 | u32 off_mask; 36 | u32 off_submit; 37 | u32 off_finish; // nothing to finish if == submit 38 | u32 reserved; 39 | struct { 40 | struct aiocb aiocb; 41 | void * data; 42 | } aring[0]; 43 | #endif // LIBURING 44 | }; 45 | 46 | struct wring * 47 | wring_create(const int fd, const u32 iosz, const u32 depth0) 48 | { 49 | debug_assert(depth0); 50 | #if defined(LIBURING) 51 | const u32 depth = depth0 < 64 ? bits_p2_up_u32(depth0) : 64; 52 | struct wring * const wring = calloc(1, sizeof(*wring)); 53 | #else // POSIX AIO 54 | const u32 depth = depth0 < 8 ? bits_p2_up_u32(depth0) : 8; 55 | struct wring * const wring = calloc(1, sizeof(*wring) + (sizeof(wring->aring[0]) * depth)); 56 | #endif // LIBURING 57 | const u64 iosz_64 = iosz; // memsz can be larger 58 | 59 | if (!wring) 60 | return NULL; 61 | const size_t memsz = bits_round_up(iosz_64 * depth, 21); // a multiple of 2MB 62 | // 2MB buffer (initialized to zero by mmap) 63 | u8 * const mem = pages_alloc_best(memsz, false, &wring->memsz); 64 | if (!mem) { 65 | free(wring); 66 | return NULL; 67 | } 68 | // link all 69 | for (u64 i = 0; i < depth-1; i++) 70 | *(u64 *)(mem + (iosz_64 * i)) = (u64)(mem + (iosz_64 * (i+1))); 71 | wring->mem = mem; 72 | wring->head = mem; 73 | wring->iosz = iosz; 74 | wring->fd = fd; 75 | #if defined(LIBURING) 76 | wring->batch = depth >> 2; // 1/4 77 | struct io_uring_params p = {}; 78 | // uncomment to use sqpoll (must use root or sys_admin) 79 | //p.flags = IORING_SETUP_SQPOLL | IORING_SETUP_SQ_AFF; 80 | //p.sq_thread_cpu = 2; 81 | if (io_uring_queue_init_params(depth << 1, &wring->uring, &p)) { 82 | pages_unmap(wring->mem, wring->memsz); 83 | free(wring); 84 | return NULL; 85 | } 86 | 87 | // register memory and file 88 | struct iovec vecs = {mem, wring->memsz}; 89 | // enable memlock in /etc/security/limits.conf 90 | wring->fixed_mem = io_uring_register_buffers(&wring->uring, &vecs, 1) == 0; 91 | // this usually works 92 | wring->fixed_file = io_uring_register_files(&wring->uring, &fd, 1) == 0; 93 | 94 | #else 95 | wring->off_mask = depth - 1; 96 | #endif // LIBURING 97 | return wring; 98 | } 99 | 100 | void 101 | wring_update_fd(struct wring * const wring, const int fd) 102 | { 103 | wring->fd = fd; 104 | #if defined(LIBURING) 105 | if (wring->fixed_file) { 106 | const int r = io_uring_register_files_update(&wring->uring, 0, &wring->fd, 1); 107 | if (r != 1) 108 | wring->fixed_file = false; 109 | } 110 | #endif // LIBURING 111 | } 112 | 113 | void 114 | wring_destroy(struct wring * const wring) 115 | { 116 | wring_flush(wring); 117 | 118 | #if defined(LIBURING) 119 | io_uring_queue_exit(&wring->uring); 120 | #endif // LIBURING 121 | 122 | pages_unmap(wring->mem, wring->memsz); 123 | free(wring); 124 | } 125 | 126 | static void * 127 | wring_wait(struct wring * const wring) 128 | { 129 | #if defined(LIBURING) 130 | debug_assert(wring->nring); 131 | struct io_uring_cqe * cqe = NULL; 132 | // wait and directly return buffer 133 | int ret = io_uring_wait_cqe(&wring->uring, &cqe); 134 | if (ret) 135 | debug_die(); 136 | if (cqe->res < 0) 137 | debug_die(); 138 | 139 | void * const ptr = io_uring_cqe_get_data(cqe); 140 | io_uring_cqe_seen(&wring->uring, cqe); 141 | wring->nring--; 142 | #else // AIO 143 | debug_assert(wring->off_submit != wring->off_finish); 144 | const u32 i = wring->off_finish & wring->off_mask; 145 | struct aiocb * const cb = &(wring->aring[i].aiocb); 146 | do { 147 | const int r = aio_error(cb); 148 | if (r == 0) 149 | break; 150 | else if (r != EINPROGRESS) 151 | debug_die_perror(); 152 | cpu_pause(); 153 | } while (true); 154 | const ssize_t ret = aio_return(cb); 155 | if (ret != (ssize_t)cb->aio_nbytes) 156 | debug_die(); 157 | void * const ptr = wring->aring[i].data; 158 | wring->off_finish++; 159 | #endif // LIBURING 160 | return ptr; 161 | } 162 | 163 | static void * 164 | wring_wait_buf(struct wring * const wring) 165 | { 166 | do { 167 | void * const buf = wring_wait(wring); 168 | if (buf) 169 | return buf; 170 | } while (true); 171 | } 172 | 173 | void * 174 | wring_acquire(struct wring * const wring) 175 | { 176 | if (wring->head == NULL) 177 | return wring_wait_buf(wring); 178 | // use the free list 179 | void * const ptr = wring->head; 180 | debug_assert(ptr); 181 | wring->head = (void *)(*(u64*)ptr); 182 | return ptr; 183 | } 184 | 185 | static void 186 | wring_finish(struct wring * const wring) 187 | { 188 | void * const ptr = wring_wait(wring); 189 | if (ptr) { // may return NULL for fsync 190 | *(u64*)ptr = (u64)(wring->head); 191 | wring->head = ptr; 192 | } 193 | } 194 | 195 | #if defined(LIBURING) 196 | static void 197 | wring_submit(struct wring * const wring) 198 | { 199 | const int n = io_uring_submit(&wring->uring); 200 | if (unlikely(n < 0)) 201 | debug_die(); 202 | debug_assert(n > 0 && (u32)n <= wring->pending); 203 | wring->pending -= (u32)n; 204 | } 205 | #else 206 | static void 207 | wring_wait_slot(struct wring * const wring) 208 | { 209 | // test if the ring is already full 210 | if ((wring->off_finish + wring->off_mask) == wring->off_submit) 211 | wring_finish(wring); 212 | } 213 | #endif // LIBURING 214 | 215 | // write a 4kB page 216 | void 217 | wring_write_partial(struct wring * const wring, const off_t off, 218 | void * const buf, const size_t buf_off, const u32 size) 219 | { 220 | debug_assert((buf_off + size) <= wring->iosz); 221 | u8 * const wbuf = ((u8 *)buf) + buf_off; 222 | #if defined(LIBURING) 223 | struct io_uring_sqe * const sqe = io_uring_get_sqe(&wring->uring); 224 | debug_assert(sqe); 225 | 226 | const int fd = wring->fixed_file ? 0 : wring->fd; 227 | if (wring->fixed_mem) { 228 | io_uring_prep_write_fixed(sqe, fd, wbuf, size, off, 0); 229 | } else { 230 | io_uring_prep_write(sqe, fd, wbuf, size, off); 231 | } 232 | 233 | if (wring->fixed_file) 234 | io_uring_sqe_set_flags(sqe, IOSQE_FIXED_FILE); 235 | 236 | io_uring_sqe_set_data(sqe, buf); 237 | 238 | wring->pending++; 239 | wring->nring++; 240 | if (wring->pending >= wring->batch) 241 | wring_submit(wring); 242 | #else // AIO 243 | wring_wait_slot(wring); 244 | const u32 i = wring->off_submit & wring->off_mask; 245 | struct aiocb * const cb = &wring->aring[i].aiocb; 246 | cb->aio_fildes = wring->fd; 247 | cb->aio_buf = wbuf; 248 | cb->aio_nbytes = size; 249 | cb->aio_offset = off; 250 | wring->aring[i].data = buf; 251 | const int r = aio_write(cb); 252 | if (r != 0) 253 | debug_die_perror(); 254 | 255 | wring->off_submit++; 256 | #endif // LIBURING 257 | } 258 | 259 | inline void 260 | wring_write(struct wring * const wring, const off_t off, void * const buf) 261 | { 262 | wring_write_partial(wring, off, buf, 0, wring->iosz); 263 | } 264 | 265 | static bool 266 | wring_empty(struct wring * const wring) 267 | { 268 | #if defined(LIBURING) 269 | return wring->nring == 0; 270 | #else 271 | return wring->off_submit == wring->off_finish; 272 | #endif // LIBURING 273 | } 274 | 275 | void 276 | wring_flush(struct wring * const wring) 277 | { 278 | #if defined(LIBURING) 279 | while (wring->pending) 280 | wring_submit(wring); 281 | #endif // LIBURING 282 | while (!wring_empty(wring)) 283 | wring_finish(wring); 284 | } 285 | 286 | void 287 | wring_fsync(struct wring * const wring) 288 | { 289 | #if defined(LIBURING) 290 | struct io_uring_sqe * const sqe = io_uring_get_sqe(&wring->uring); 291 | io_uring_prep_fsync(sqe, wring->fd, IORING_FSYNC_DATASYNC); 292 | io_uring_sqe_set_data(sqe, NULL); 293 | wring->pending++; 294 | wring->nring++; 295 | wring_submit(wring); // just submit 296 | #else 297 | wring_wait_slot(wring); 298 | const u32 i = wring->off_submit & wring->off_mask; 299 | struct aiocb * const cb = &wring->aring[i].aiocb; 300 | cb->aio_fildes = wring->fd; 301 | cb->aio_buf = NULL; 302 | cb->aio_nbytes = 0; 303 | wring->aring[i].data = NULL; 304 | #if defined(__FreeBSD__) 305 | const int flags = O_SYNC; 306 | #else 307 | const int flags = O_DSYNC; 308 | #endif 309 | const int r = aio_fsync(flags, cb); 310 | if (r != 0) 311 | debug_die_perror(); 312 | 313 | wring->off_submit++; 314 | #endif // LIBURING 315 | } 316 | // }}} wring 317 | 318 | // coq {{{ 319 | 320 | // struct {{{ 321 | #define COQ_NR ((128u)) 322 | #define COQ_MASK ((COQ_NR - 1u)) 323 | #define COQ_MAX ((COQ_MASK)) 324 | 325 | static_assert((COQ_NR & COQ_MASK) == 0, "COQ_NR"); 326 | 327 | // the wait queue 328 | struct cowqe { 329 | cowq_func func; 330 | void * priv; 331 | }; 332 | 333 | struct coht { 334 | au32 head; // consume the head 335 | au32 tail; // append to the tail 336 | }; 337 | 338 | struct coq { 339 | struct coht rqht; 340 | struct coht wqht; 341 | 342 | struct co * rq[COQ_NR]; 343 | struct cowqe wq[COQ_NR]; 344 | #if defined(LIBURING) 345 | struct io_uring uring[0]; // optional uring at the end 346 | #endif // LIBURING 347 | }; 348 | // }}} struct 349 | 350 | // helpers {{{ 351 | static inline u32 352 | coht_nr(struct coht * const ht) 353 | { 354 | const u32 nr = ht->tail - ht->head; 355 | debug_assert(nr < COQ_NR); 356 | return nr; 357 | } 358 | 359 | static inline bool 360 | coht_full(struct coht * const ht) 361 | { 362 | return coht_nr(ht) == COQ_MAX; 363 | } 364 | 365 | static inline bool 366 | coht_empty(struct coht * const ht) 367 | { 368 | return coht_nr(ht) == 0; 369 | } 370 | 371 | static inline u32 372 | coht_enqueue(struct coht * const ht) 373 | { 374 | debug_assert(!coht_full(ht)); 375 | 376 | const u32 i = ht->tail & COQ_MASK; 377 | ht->tail++; 378 | cpu_cfence(); 379 | return i; 380 | } 381 | 382 | static inline u32 383 | coht_dequeue(struct coht * const ht) 384 | { 385 | debug_assert(!coht_empty(ht)); 386 | const u32 i = ht->head & COQ_MASK; 387 | ht->head++; 388 | cpu_cfence(); 389 | return i; 390 | } 391 | 392 | static inline u32 393 | corq_nr(struct coq * const q) 394 | { 395 | return coht_nr(&(q->rqht)); 396 | } 397 | 398 | static inline u32 399 | cowq_nr(struct coq * const q) 400 | { 401 | return coht_nr(&(q->wqht)); 402 | } 403 | 404 | static inline bool 405 | corq_full(struct coq * const q) 406 | { 407 | return coht_full(&(q->rqht)); 408 | } 409 | 410 | static inline bool 411 | cowq_full(struct coq * const q) 412 | { 413 | return coht_full(&(q->wqht)); 414 | } 415 | 416 | static inline bool 417 | corq_empty(struct coq * const q) 418 | { 419 | return coht_empty(&(q->rqht)); 420 | } 421 | 422 | static inline bool 423 | cowq_empty(struct coq * const q) 424 | { 425 | return coht_empty(&(q->wqht)); 426 | } 427 | // }}} helpers 428 | 429 | // coq {{{ 430 | struct coq * 431 | coq_create(void) 432 | { 433 | struct coq * const q = calloc(1, sizeof(*q)); 434 | return q; 435 | } 436 | 437 | void 438 | coq_destroy(struct coq * const coq) 439 | { 440 | free(coq); 441 | } 442 | 443 | struct coq * 444 | coq_create_auto(const u32 depth) 445 | { 446 | #if defined(LIBURING) 447 | return coq_uring_create_pair(depth); 448 | #else 449 | (void)depth; 450 | return coq_create(); 451 | #endif // LIBURING 452 | } 453 | 454 | void 455 | coq_destroy_auto(struct coq * const coq) 456 | { 457 | #if defined(LIBURING) 458 | coq_uring_destroy_pair(coq); 459 | #else 460 | coq_destroy(coq); 461 | #endif // LIBURING 462 | } 463 | 464 | // return the position in the queue 465 | u32 466 | corq_enqueue(struct coq * const q, struct co * const co) 467 | { 468 | if (corq_full(q)) 469 | return UINT32_MAX; 470 | 471 | const u32 i = coht_enqueue(&(q->rqht)); 472 | q->rq[i] = co; 473 | return i; 474 | } 475 | 476 | // return the position in the queue 477 | // the func should do clean up for the target coroutine 478 | u32 479 | cowq_enqueue(struct coq * const q, cowq_func func, void * const priv) 480 | { 481 | if (cowq_full(q)) 482 | return UINT32_MAX; 483 | 484 | const u32 i = coht_enqueue(&(q->wqht)); 485 | q->wq[i].func = func; 486 | q->wq[i].priv = priv; 487 | return i; 488 | } 489 | 490 | static void 491 | corq_process(struct coq * const q) 492 | { 493 | if (corq_empty(q)) 494 | return; 495 | 496 | const u32 i = coht_dequeue(&(q->rqht)); 497 | struct co * const co = q->rq[i]; 498 | co_enter(co, 0); 499 | if (!co_valid(co)) 500 | co_destroy(co); 501 | } 502 | 503 | static void 504 | cowq_process(struct coq * const q) 505 | { 506 | do { 507 | const u32 i = coht_dequeue(&q->wqht); 508 | cowq_func func = q->wq[i].func; 509 | if (func) { // skip empty entries 510 | func(q->wq[i].priv); 511 | return; 512 | } 513 | } while (cowq_nr(q)); 514 | } 515 | 516 | void 517 | cowq_remove(struct coq * const q, const u32 i) 518 | { 519 | q->wq[i].func = NULL; 520 | } 521 | 522 | // will not give control to wq workers 523 | void 524 | coq_yield(struct coq * const q) 525 | { 526 | corq_enqueue(q, co_self()); 527 | co_back(0); 528 | } 529 | 530 | static bool 531 | coq_process_idle(void * const priv) 532 | { 533 | struct co * const co = (typeof(co))priv; 534 | co_enter(co, 0); 535 | if (!co_valid(co)) 536 | co_destroy(co); 537 | return true; 538 | } 539 | 540 | // resume after all the wq workers have run 541 | void 542 | coq_idle(struct coq * const q) 543 | { 544 | cowq_enqueue(q, coq_process_idle, co_self()); 545 | co_back(0); 546 | } 547 | 548 | void 549 | coq_run(struct coq * const q) 550 | { 551 | while (!(corq_empty(q) && cowq_empty(q))) { 552 | 553 | // flush the run-queue 554 | while (corq_nr(q)) 555 | corq_process(q); 556 | 557 | // process work-completion as long as the run-queue is empty 558 | while (cowq_nr(q) && corq_empty(q)) 559 | cowq_process(q); 560 | } 561 | } 562 | 563 | static __thread struct coq * coq_curr = NULL; 564 | 565 | inline void 566 | coq_install(struct coq * const q) 567 | { 568 | if (coq_curr) 569 | debug_die(); 570 | coq_curr = q; 571 | } 572 | 573 | inline void 574 | coq_uninstall(void) 575 | { 576 | if (coq_curr == NULL) 577 | debug_die(); 578 | coq_curr = NULL; 579 | } 580 | 581 | inline struct coq * 582 | coq_current(void) 583 | { 584 | return coq_curr; 585 | } 586 | // }}} coq 587 | 588 | // aio {{{ 589 | static bool 590 | cowq_process_aio(void * const priv) 591 | { 592 | struct co * const co = (typeof(co))priv; 593 | debug_assert(co); 594 | co_enter(co, 0); 595 | if (!co_valid(co)) 596 | co_destroy(co); 597 | return true; 598 | } 599 | 600 | static ssize_t 601 | coq_wait_aio(struct coq * const q, struct aiocb * const cb, struct co * const self) 602 | { 603 | do { 604 | cowq_enqueue(q, cowq_process_aio, (void *)self); 605 | co_back(0); 606 | const int r = aio_error(cb); 607 | if (r != EINPROGRESS) 608 | return aio_return(cb); 609 | } while (true); 610 | } 611 | 612 | ssize_t 613 | coq_pread_aio(struct coq * const q, const int fd, void * const buf, const size_t count, const off_t offset) 614 | { 615 | struct co * const self = co_self(); 616 | if (!self) 617 | return pread(fd, buf, count, offset); 618 | 619 | struct aiocb cb = { .aio_fildes = fd, .aio_buf = buf, .aio_nbytes = count, .aio_offset = offset}; 620 | const int r = aio_read(&cb); 621 | if (unlikely(r)) 622 | return -1; 623 | 624 | return coq_wait_aio(q, &cb, self); 625 | } 626 | 627 | ssize_t 628 | coq_pwrite_aio(struct coq * const q, const int fd, const void * const buf, const size_t count, const off_t offset) 629 | { 630 | struct co * const self = co_self(); 631 | if (!self) 632 | return pwrite(fd, buf, count, offset); 633 | 634 | struct aiocb cb = { .aio_fildes = fd, .aio_buf = (void *)buf, .aio_nbytes = count, .aio_offset = offset}; 635 | const int r = aio_write(&cb); 636 | if (unlikely(r)) 637 | return -1; 638 | 639 | return coq_wait_aio(q, &cb, self); 640 | } 641 | // }}} aio 642 | 643 | // io_uring {{{ 644 | #if defined(LIBURING) 645 | static inline bool 646 | coq_uring_init(struct io_uring * const ring, const u32 depth) 647 | { 648 | struct io_uring_params p = {}; 649 | return 0 == io_uring_queue_init_params(depth, ring, &p); 650 | } 651 | 652 | struct io_uring * 653 | coq_uring_create(const u32 depth) 654 | { 655 | struct io_uring * const ring = malloc(sizeof(*ring)); 656 | if (coq_uring_init(ring, depth)) { 657 | return ring; 658 | } else { 659 | free(ring); 660 | return NULL; 661 | } 662 | } 663 | 664 | // returns a coq plus a uring at the end 665 | struct coq * 666 | coq_uring_create_pair(const u32 depth) 667 | { 668 | struct coq * const coq = calloc(1, sizeof(struct coq) + sizeof(struct io_uring)); 669 | if (coq_uring_init(coq->uring, depth)) { 670 | return coq; 671 | } else { 672 | free(coq); 673 | return NULL; 674 | } 675 | } 676 | 677 | void 678 | coq_uring_destroy(struct io_uring * const ring) 679 | { 680 | io_uring_queue_exit(ring); 681 | free(ring); 682 | } 683 | 684 | void 685 | coq_uring_destroy_pair(struct coq * const coq) 686 | { 687 | io_uring_queue_exit(coq->uring); 688 | coq_destroy(coq); 689 | } 690 | 691 | static bool 692 | cowq_process_uring(void * const priv) 693 | { 694 | struct io_uring * const ring = (typeof(ring))priv; 695 | struct io_uring_cqe * cqe = NULL; 696 | int ret = io_uring_wait_cqe(ring, &cqe); 697 | 698 | if (ret) 699 | return false; 700 | 701 | struct co * const co = (typeof(co))io_uring_cqe_get_data(cqe); 702 | debug_assert(co); 703 | co_enter(co, (u64)cqe); 704 | if (!co_valid(co)) 705 | co_destroy(co); 706 | return true; 707 | } 708 | 709 | ssize_t 710 | coq_pread_uring(struct coq * const q, struct io_uring * const ring0, 711 | const int fd, void * const buf, const size_t count, const off_t offset) 712 | { 713 | struct co * const self = co_self(); 714 | if (!self) 715 | return pread(fd, buf, count, offset); 716 | 717 | struct io_uring * const ring = ring0 ? ring0 : q->uring; 718 | struct io_uring_sqe * const sqe = io_uring_get_sqe(ring); 719 | if (sqe == NULL) 720 | return -1; 721 | 722 | struct iovec vec = {.iov_base = buf, .iov_len = count}; 723 | io_uring_prep_readv(sqe, fd, &vec, 1, offset); 724 | io_uring_sqe_set_data(sqe, self); 725 | io_uring_submit(ring); 726 | 727 | // prepare callback 728 | cowq_enqueue(q, cowq_process_uring, (void *)ring); 729 | 730 | // yield 731 | struct io_uring_cqe * const cqe = (typeof(cqe))co_back(0); 732 | debug_assert(cqe); 733 | const ssize_t ret = cqe->res; 734 | io_uring_cqe_seen(ring, cqe); 735 | return ret; 736 | } 737 | 738 | ssize_t 739 | coq_pwrite_uring(struct coq * const q, struct io_uring * const ring0, 740 | const int fd, const void * const buf, const size_t count, const off_t offset) 741 | { 742 | struct co * const self = co_self(); 743 | if (!self) 744 | return pwrite(fd, buf, count, offset); 745 | 746 | struct io_uring * const ring = ring0 ? ring0 : q->uring; 747 | struct io_uring_sqe * const sqe = io_uring_get_sqe(ring); 748 | if (sqe == NULL) 749 | return -1; 750 | 751 | struct iovec vec = {.iov_base = (void *)buf, .iov_len = count}; 752 | io_uring_prep_writev(sqe, fd, &vec, 1, offset); 753 | io_uring_sqe_set_data(sqe, self); 754 | io_uring_submit(ring); 755 | 756 | // prepare callback 757 | cowq_enqueue(q, cowq_process_uring, (void *)ring); 758 | 759 | // yield 760 | struct io_uring_cqe * const cqe = (typeof(cqe))co_back(0); 761 | debug_assert(cqe); 762 | const ssize_t ret = cqe->res; 763 | io_uring_cqe_seen(ring, cqe); 764 | return ret; 765 | } 766 | #endif // LIBURING 767 | // }}} io_uring 768 | 769 | // }}} coq 770 | 771 | // rcache {{{ 772 | // read-only cache 773 | #define RCACHE_NWAY ((16u)) 774 | #define RCACHE_VSHIFT128 ((2)) 775 | #define RCACHE_VSHIFT256 ((3)) 776 | #define RCACHE_VSHIFT512 ((4)) 777 | #define RCACHE_VWAY128 ((RCACHE_NWAY >> RCACHE_VSHIFT128)) 778 | #define RCACHE_VWAY256 ((RCACHE_NWAY >> RCACHE_VSHIFT256)) 779 | #define RCACHE_VWAY512 ((RCACHE_NWAY >> RCACHE_VSHIFT512)) 780 | #define RCACHE_MASK ((RCACHE_NWAY - 1)) 781 | #define RCACHE_MAXHIST ((UINT8_MAX - 1)) 782 | struct rcache_group { 783 | u8 hist[RCACHE_NWAY]; // 1x16=16B 784 | spinlock lock; // 4B 785 | au32 valid_bits; // 4B 786 | au32 write_bits; // 4B 787 | au32 dirty_bits; // 4B 788 | au16 refcnt[RCACHE_NWAY]; // 2x16=32B 789 | union { 790 | u32 tag[RCACHE_NWAY]; // 4x16=64B: high x-bit is fd; low y-bit is page-number (256MB max) 791 | m128 tagv128[RCACHE_VWAY128]; 792 | #if defined(__AVX2__) 793 | m256 tagv256[RCACHE_VWAY256]; 794 | #endif 795 | #if defined(__AVX512F__) 796 | m512 tagv512[RCACHE_VWAY512]; 797 | #endif 798 | }; 799 | }; 800 | 801 | static_assert((sizeof(struct rcache_group) % 64) == 0, "rcache_group size"); 802 | 803 | struct rcache { 804 | u8 * mem; 805 | struct rcache_group * groups; 806 | u32 group_mask; 807 | u32 nr_groups; 808 | u32 fd_shift; 809 | u32 pno_mask; 810 | u64 memsize; 811 | u64 gmemsize; 812 | struct bitmap * close_bm; 813 | }; 814 | 815 | struct rcache * 816 | rcache_create(const u64 size_mb, const u32 fd_bits) 817 | { 818 | debug_assert(size_mb && fd_bits); 819 | const u64 cachesz = bits_p2_up_u64(size_mb) << 20; 820 | const u64 npages = cachesz / PGSZ; 821 | const u64 ngroups = npages / (u64)RCACHE_NWAY; 822 | if (ngroups > UINT32_MAX) 823 | return NULL; 824 | struct rcache * const c = calloc(1, sizeof(*c)); 825 | if (!c) 826 | return NULL; 827 | c->mem = pages_alloc_best(cachesz, true, &c->memsize); // can use 1GB huge page 828 | if (!c->mem) { 829 | free(c); 830 | return NULL; 831 | } 832 | c->groups = pages_alloc_best(ngroups * sizeof(struct rcache_group), false, &c->gmemsize); 833 | if (!c->groups) { 834 | pages_unmap(c->mem, c->memsize); 835 | free(c); 836 | return NULL; 837 | } 838 | c->group_mask = (u32)ngroups - 1; 839 | c->nr_groups = (u32)ngroups; 840 | 841 | c->fd_shift = 32 - fd_bits; 842 | c->pno_mask = (1u << c->fd_shift) - 1u; 843 | 844 | c->close_bm = bitmap_create(1lu << fd_bits); 845 | debug_assert(c->close_bm); 846 | 847 | for (u64 i = 0; i < ngroups; i++) 848 | spinlock_init(&(c->groups[i].lock)); 849 | 850 | return c; 851 | } 852 | 853 | void 854 | rcache_destroy(struct rcache * const c) 855 | { 856 | free(c->close_bm); 857 | pages_unmap(c->mem, c->memsize); 858 | pages_unmap(c->groups, c->gmemsize); 859 | free(c); 860 | } 861 | 862 | static inline void 863 | rcache_read(int fd, void *pg, u32 pno) 864 | { 865 | struct coq * const coq = coq_current(); 866 | if (coq) { 867 | #if defined(LIBURING) 868 | if (coq_pread_uring(coq, NULL, fd, pg, PGSZ, PGSZ * pno) != PGSZ) 869 | debug_die(); 870 | #else 871 | if (coq_pread_aio(coq, fd, pg, PGSZ, PGSZ * pno) != PGSZ) 872 | debug_die(); 873 | #endif // LIBURING 874 | } else { // regular pread 875 | if (pread(fd, pg, PGSZ, PGSZ * pno) != PGSZ) 876 | debug_die(); 877 | } 878 | } 879 | 880 | static inline int 881 | rcache_tag_to_fd(struct rcache * const c, const u32 tag) 882 | { 883 | return (int)(tag >> c->fd_shift); 884 | } 885 | 886 | static inline u32 887 | rcache_tag(struct rcache * const c, const int fd, const u32 pno) 888 | { 889 | debug_assert(fd > 0); // please don't use stdin 890 | debug_assert((u32)__builtin_clz((u32)fd) >= c->fd_shift); 891 | debug_assert(pno <= c->pno_mask); 892 | 893 | const u32 tag = (((u32)fd) << c->fd_shift) | pno; 894 | return tag; 895 | } 896 | 897 | static inline u32 898 | rcache_hash(const u32 tag) 899 | { 900 | return crc32c_u32(0x0D15EA5Eu, tag); 901 | } 902 | 903 | static inline u8 * 904 | rcache_page(struct rcache * const c, const u32 gid, const u32 i) 905 | { 906 | return c->mem + (PGSZ * (gid * RCACHE_NWAY + i)); 907 | } 908 | 909 | // thread-unsafe 910 | void 911 | rcache_close_lazy(struct rcache * const c, const int fd) 912 | { 913 | debug_assert(bitmap_test(c->close_bm, (u64)fd) == false); 914 | bitmap_set1(c->close_bm, (u64)fd); 915 | } 916 | 917 | // thread-unsafe 918 | u64 919 | rcache_close_flush(struct rcache * const c) 920 | { 921 | struct bitmap * const bm = c->close_bm; 922 | const u64 count = bitmap_count(bm); 923 | if (count == 0) 924 | return 0; 925 | 926 | for (u32 i = 0; i < c->nr_groups; i++) { 927 | struct rcache_group * const g = &(c->groups[i]); 928 | spinlock_lock(&(g->lock)); 929 | for (u32 j = 0; j < RCACHE_NWAY; j++) { 930 | const int fd = rcache_tag_to_fd(c, g->tag[j]); 931 | if (bitmap_test(bm, (u64)fd)) { 932 | g->tag[j] = 0; 933 | g->hist[j] = 0; 934 | debug_assert(g->refcnt[j] == 0); 935 | } 936 | } 937 | spinlock_unlock(&(g->lock)); 938 | } 939 | while (bitmap_count(bm)) { 940 | const u64 bit = bitmap_first(bm); 941 | close((int)bit); 942 | bitmap_set0(bm, bit); 943 | } 944 | return count; 945 | } 946 | 947 | // invalidate cache and close(fd) 948 | void 949 | rcache_close(struct rcache * const c, const int fd) 950 | { 951 | for (u32 i = 0; i < c->nr_groups; i++) { 952 | struct rcache_group * const g = &(c->groups[i]); 953 | spinlock_lock(&(g->lock)); 954 | for (u32 j = 0; j < RCACHE_NWAY; j++) { 955 | if (rcache_tag_to_fd(c, g->tag[j]) == fd) { 956 | g->tag[j] = 0; 957 | g->hist[j] = 0; 958 | debug_assert(g->refcnt[j] == 0); 959 | } 960 | } 961 | spinlock_unlock(&(g->lock)); 962 | } 963 | close(fd); 964 | } 965 | 966 | static inline void 967 | rcache_pause(void) 968 | { 969 | struct coq * const coq = coq_current(); 970 | if (coq) 971 | coq_idle(coq); 972 | else 973 | cpu_pause(); 974 | } 975 | 976 | // lock has been acquired 977 | // read-only; return a page that has zero reference 978 | static u32 979 | rcache_search_victim(struct rcache_group * const g, const u32 i0) 980 | { 981 | // search unused page 982 | #pragma nounroll 983 | do { 984 | u32 imin = i0; 985 | u16 cmin = UINT16_MAX; 986 | u8 hmin = UINT8_MAX; 987 | #pragma nounroll 988 | for (u32 k = 0; k < RCACHE_NWAY; k++) { 989 | const u32 i = (k + i0) & RCACHE_MASK; 990 | if (g->hist[i] < hmin && atomic_load_explicit(&(g->refcnt[i]), MO_CONSUME) == 0) { 991 | // refcnt is 0 but we may still have a better choice 992 | imin = i; 993 | cmin = 0; 994 | hmin = g->hist[i]; 995 | } 996 | } 997 | if (cmin == 0) // found a victim 998 | return imin; 999 | 1000 | rcache_pause(); 1001 | } while (true); 1002 | } 1003 | 1004 | static void * 1005 | rcache_hit_i(struct rcache * const c, const u32 gid, struct rcache_group * const g, const u32 i) 1006 | { 1007 | const u8 hist0 = g->hist[i]; 1008 | if (hist0 < RCACHE_MAXHIST) 1009 | g->hist[i] = hist0 + 1; 1010 | atomic_fetch_add_explicit(&(g->refcnt[i]), 1, MO_ACQUIRE); 1011 | spinlock_unlock(&(g->lock)); 1012 | // wait if invalid 1013 | const u32 vbit = 1u << i; 1014 | while ((atomic_load_explicit(&(g->valid_bits), MO_CONSUME) & vbit) == 0) 1015 | rcache_pause(); 1016 | return rcache_page(c, gid, i); 1017 | } 1018 | 1019 | static void * 1020 | rcache_hit(struct rcache * const c, const u32 tag, const u32 gid, struct rcache_group * const g) 1021 | { 1022 | #if defined(__x86_64__) 1023 | #if defined(__AVX512F__) 1024 | const m512 tmpv = _mm512_set1_epi32((s32)tag); 1025 | for (u32 v = 0; v < RCACHE_VWAY512; v++) { 1026 | const u32 m = (u32)_mm512_cmpeq_epu32_mask(tmpv, g->tagv512[v]); 1027 | if (m) { 1028 | const u32 i = (v << RCACHE_VSHIFT512) + ((u32)__builtin_ctz(m)); 1029 | return rcache_hit_i(c, gid, g, i); 1030 | } 1031 | } 1032 | #elif defined(__AVX2__) 1033 | const m256 tmpv = _mm256_set1_epi32((s32)tag); 1034 | for (u32 v = 0; v < RCACHE_VWAY256; v++) { 1035 | const u32 m = (u32)_mm256_movemask_epi8(_mm256_cmpeq_epi32(tmpv, g->tagv256[v])); 1036 | if (m) { 1037 | const u32 i = (v << RCACHE_VSHIFT256) + ((u32)__builtin_ctz(m) >> 2); 1038 | return rcache_hit_i(c, gid, g, i); 1039 | } 1040 | } 1041 | #else // SSE4.2 1042 | const m128 tmpv = _mm_set1_epi32((s32)tag); 1043 | for (u32 v = 0; v < RCACHE_VWAY128; v++) { 1044 | const u32 m = (u32)_mm_movemask_epi8(_mm_cmpeq_epi32(tmpv, g->tagv128[v])); 1045 | if (m) { 1046 | const u32 i = (v << RCACHE_VSHIFT128) + ((u32)__builtin_ctz(m) >> 2); 1047 | return rcache_hit_i(c, gid, g, i); 1048 | } 1049 | } 1050 | #endif // __AVX512F__ 1051 | #else // TODO: AArch64 SIMD 1052 | const u32 i0 = tag & RCACHE_MASK; 1053 | for (u32 k = 0; k < RCACHE_NWAY; k++) { 1054 | const u32 i = (k + i0) & RCACHE_MASK; 1055 | if (g->tag[i] == tag) // hit 1056 | return rcache_hit_i(c, gid, g, i); 1057 | } 1058 | #endif // __x86_64__ 1059 | // still locked 1060 | return NULL; 1061 | } 1062 | 1063 | static __thread u64 rcache_stat_reads = 0; 1064 | void * 1065 | rcache_acquire(struct rcache * const c, const int fd, const u32 pno) 1066 | { 1067 | const u32 tag = rcache_tag(c, fd, pno); 1068 | const u32 gid = rcache_hash(tag) & c->group_mask; 1069 | struct rcache_group * const g = &(c->groups[gid]); 1070 | 1071 | spinlock_lock(&(g->lock)); 1072 | void * const ret1 = rcache_hit(c, tag, gid, g); 1073 | if (ret1) 1074 | return ret1; 1075 | 1076 | const u32 iv = rcache_search_victim(g, tag & RCACHE_MASK); 1077 | debug_assert(g->refcnt[iv] == 0); 1078 | 1079 | void * const pg = rcache_page(c, gid, iv); 1080 | atomic_store_explicit(&(g->refcnt[iv]), 1, MO_RELAXED); 1081 | g->tag[iv] = tag; 1082 | g->hist[iv] = 0; 1083 | const u32 vbit = 1u << iv; 1084 | atomic_fetch_and_explicit(&(g->valid_bits), ~vbit, MO_ACQUIRE); // clear bit 1085 | spinlock_unlock(&(g->lock)); 1086 | // perform I/O after releasing the lock 1087 | rcache_read(fd, pg, pno); // must succeed 1088 | rcache_stat_reads++; 1089 | atomic_fetch_or_explicit(&(g->valid_bits), vbit, MO_RELEASE); // clear bit 1090 | return pg; 1091 | } 1092 | 1093 | void 1094 | rcache_retain(struct rcache * const c, const void * const buf) 1095 | { 1096 | const u64 tmp = (((u64)buf) - ((u64)c->mem)) / PGSZ; 1097 | const u32 gid = (u32)(tmp / RCACHE_NWAY); 1098 | const u32 i = tmp & RCACHE_MASK; 1099 | 1100 | struct rcache_group * const g = &(c->groups[gid]); 1101 | debug_assert(g->refcnt[i]); 1102 | atomic_fetch_add_explicit(&(g->refcnt[i]), 1, MO_ACQUIRE); 1103 | } 1104 | 1105 | void 1106 | rcache_release(struct rcache * const c, const void * const buf) 1107 | { 1108 | const u64 tmp = (((u64)buf) - ((u64)c->mem)) / PGSZ; 1109 | const u32 gid = (u32)(tmp / RCACHE_NWAY); 1110 | const u32 i = tmp & RCACHE_MASK; 1111 | 1112 | struct rcache_group * const g = &(c->groups[gid]); 1113 | debug_assert(g->refcnt[i]); 1114 | atomic_fetch_sub_explicit(&(g->refcnt[i]), 1, MO_RELEASE); 1115 | } 1116 | 1117 | inline void 1118 | rcache_thread_stat_reset(void) 1119 | { 1120 | rcache_stat_reads = 0; 1121 | } 1122 | 1123 | inline u64 1124 | rcache_thread_stat_reads(void) 1125 | { 1126 | return rcache_stat_reads; 1127 | } 1128 | // }}} rcache 1129 | 1130 | // vim:fdm=marker 1131 | -------------------------------------------------------------------------------- /blkio.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | #include "lib.h" 9 | #if defined(LIBURING) 10 | #include 11 | #endif // LIBURING 12 | 13 | #ifdef __cplusplus 14 | extern "C" { 15 | #endif 16 | 17 | // wring {{{ 18 | struct wring; 19 | 20 | // iosz: fixed write size; must be a multiple of PGSZ 21 | extern struct wring * 22 | wring_create(const int fd, const u32 iosz, const u32 depth); 23 | 24 | extern void 25 | wring_update_fd(struct wring * const wring, const int fd); 26 | 27 | extern void 28 | wring_destroy(struct wring * const wring); 29 | 30 | extern void * 31 | wring_acquire(struct wring * const wring); 32 | 33 | // write part of the buf 34 | extern void 35 | wring_write_partial(struct wring * const wring, const off_t off, 36 | void * const buf, const size_t buf_off, const u32 size); 37 | 38 | extern void 39 | wring_write(struct wring * const wring, const off_t off, void * const buf); 40 | 41 | // flush the queue and wait for completion 42 | extern void 43 | wring_flush(struct wring * const wring); 44 | 45 | // send an fsync but does not wait for completion 46 | extern void 47 | wring_fsync(struct wring * const wring); 48 | // }}} wring 49 | 50 | // coq {{{ 51 | 52 | struct coq; 53 | typedef bool (*cowq_func) (void * priv); 54 | 55 | extern struct coq * 56 | coq_create(void); 57 | 58 | extern void 59 | coq_destroy(struct coq * const coq); 60 | 61 | // prefer io_uring on Linux; fall back to POSIX AIO 62 | extern struct coq * 63 | coq_create_auto(const u32 depth); 64 | 65 | extern void 66 | coq_destroy_auto(struct coq * const coq); 67 | 68 | extern u32 69 | corq_enqueue(struct coq * const q, struct co * const co); 70 | 71 | extern u32 72 | cowq_enqueue(struct coq * const q, cowq_func exec, void * const priv); 73 | 74 | extern void 75 | cowq_remove(struct coq * const q, const u32 i); 76 | 77 | extern void 78 | coq_yield(struct coq * const q); 79 | 80 | extern void 81 | coq_idle(struct coq * const q); 82 | 83 | extern void 84 | coq_run(struct coq * const q); 85 | 86 | extern void 87 | coq_install(struct coq * const q); 88 | 89 | extern void 90 | coq_uninstall(void); 91 | 92 | extern struct coq * 93 | coq_current(void); 94 | 95 | extern ssize_t 96 | coq_pread_aio(struct coq * const q, const int fd, void * const buf, const size_t count, const off_t offset); 97 | 98 | extern ssize_t 99 | coq_pwrite_aio(struct coq * const q, const int fd, const void * const buf, const size_t count, const off_t offset); 100 | 101 | #if defined(LIBURING) 102 | // io_uring-specific 103 | extern struct io_uring * 104 | coq_uring_create(const u32 depth); 105 | 106 | // use ring==NULL in pread_uring and pwrite_uring 107 | extern struct coq * 108 | coq_uring_create_pair(const u32 depth); 109 | 110 | extern void 111 | coq_uring_destroy(struct io_uring * const ring); 112 | 113 | extern void 114 | coq_uring_destroy_pair(struct coq * const coq); 115 | 116 | extern ssize_t 117 | coq_pread_uring(struct coq * const q, struct io_uring * const ring, 118 | const int fd, void * const buf, const size_t count, const off_t offset); 119 | 120 | extern ssize_t 121 | coq_pwrite_uring(struct coq * const q, struct io_uring * const ring, 122 | const int fd, const void * const buf, const size_t count, const off_t offset); 123 | #endif // LIBURING 124 | // }}} coq 125 | 126 | // rcache {{{ 127 | extern struct rcache * 128 | rcache_create(const u64 size_mb, const u32 fd_bits); 129 | 130 | extern void 131 | rcache_destroy(struct rcache * const c); 132 | 133 | extern void 134 | rcache_close_lazy(struct rcache * const c, const int fd); 135 | 136 | extern u64 137 | rcache_close_flush(struct rcache * const c); 138 | 139 | extern void 140 | rcache_close(struct rcache * const c, const int fd); 141 | 142 | extern void * 143 | rcache_acquire(struct rcache * const c, const int fd, const u32 pageid); 144 | 145 | extern void 146 | rcache_retain(struct rcache * const c, const void * const buf); 147 | 148 | extern void 149 | rcache_release(struct rcache * const c, const void * const buf); 150 | 151 | extern void 152 | rcache_thread_stat_reset(void); 153 | 154 | extern u64 155 | rcache_thread_stat_reads(void); 156 | // }}} rcache 157 | 158 | #ifdef __cplusplus 159 | } 160 | #endif 161 | // vim:fdm=marker 162 | -------------------------------------------------------------------------------- /ctypes.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | // C types only; C++ source code don't use this 9 | 10 | #include 11 | #include 12 | 13 | /* C11 atomic types */ 14 | typedef atomic_bool abool; 15 | 16 | typedef atomic_uchar au8; 17 | typedef atomic_ushort au16; 18 | typedef atomic_uint au32; 19 | typedef atomic_ulong au64; 20 | static_assert(sizeof(au8) == 1, "sizeof(au8)"); 21 | static_assert(sizeof(au16) == 2, "sizeof(au16)"); 22 | static_assert(sizeof(au32) == 4, "sizeof(au32)"); 23 | static_assert(sizeof(au64) == 8, "sizeof(au64)"); 24 | 25 | typedef atomic_char as8; 26 | typedef atomic_short as16; 27 | typedef atomic_int as32; 28 | typedef atomic_long as64; 29 | static_assert(sizeof(as8) == 1, "sizeof(as8)"); 30 | static_assert(sizeof(as16) == 2, "sizeof(as16)"); 31 | static_assert(sizeof(as32) == 4, "sizeof(as32)"); 32 | static_assert(sizeof(as64) == 8, "sizeof(as64)"); 33 | 34 | // shorten long names 35 | #define MO_RELAXED memory_order_relaxed 36 | #define MO_CONSUME memory_order_consume 37 | #define MO_ACQUIRE memory_order_acquire 38 | #define MO_RELEASE memory_order_release 39 | #define MO_ACQ_REL memory_order_acq_rel 40 | #define MO_SEQ_CST memory_order_seq_cst 41 | -------------------------------------------------------------------------------- /kv.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | #ifdef __cplusplus 9 | extern "C" { 10 | #endif 11 | 12 | // crc32c {{{ 13 | #define KV_CRC32C_SEED ((0xDEADBEEFu)) 14 | 15 | extern u32 16 | kv_crc32c(const void * const ptr, u32 len); 17 | 18 | extern u64 19 | kv_crc32c_extend(const u32 crc32c); 20 | // }}} crc32c 21 | 22 | // kv {{{ 23 | 24 | // struct {{{ 25 | /* 26 | * Some internal union names can be ignored: 27 | * struct kv { 28 | * u32 klen; 29 | * u32 vlen; 30 | * u64 hash; 31 | * u8 kv[]; 32 | * }; 33 | */ 34 | struct kv { 35 | union { // the first u64 36 | u64 kvlen; 37 | struct { 38 | u32 klen; 39 | union { u32 vlen; u32 refcnt; }; 40 | }; 41 | }; 42 | union { 43 | u64 hash; // hashvalue of the key 44 | u64 priv; // can hide a value here if hash is not used 45 | void * privptr; 46 | struct { u32 hashlo; u32 hashhi; }; // little endian 47 | struct { u32 privlo; u32 privhi; }; 48 | }; 49 | u8 kv[0]; // len(kv) == klen + vlen 50 | } __attribute__((packed)); 51 | 52 | struct kref { 53 | u32 len; 54 | union { u32 hash32; u32 priv; }; 55 | const u8 * ptr; 56 | } __attribute__((packed)); 57 | 58 | struct kvref { 59 | const u8 * kptr; // read-only 60 | const u8 * vptr; // read-only 61 | struct kv hdr; // hdr.kv[] is invalid 62 | }; 63 | // }}} struct 64 | 65 | // kv {{{ 66 | typedef int (*kv_kv_cmp_func)(const struct kv *, const struct kv *); 67 | 68 | extern size_t 69 | kv_size(const struct kv * const kv); 70 | 71 | extern size_t 72 | kv_size_align(const struct kv * const kv, const u64 align); 73 | 74 | extern size_t 75 | key_size(const struct kv * const key); 76 | 77 | extern size_t 78 | key_size_align(const struct kv * const key, const u64 align); 79 | 80 | extern void 81 | kv_update_hash(struct kv * const kv); 82 | 83 | extern void 84 | kv_refill_value(struct kv * const kv, const void * const value, const u32 vlen); 85 | 86 | extern void 87 | kv_refill(struct kv * const kv, const void * const key, const u32 klen, 88 | const void * const value, const u32 vlen); 89 | 90 | extern void 91 | kv_refill_str(struct kv * const kv, const char * const key, 92 | const void * const value, const u32 vlen); 93 | 94 | extern void 95 | kv_refill_str_str(struct kv * const kv, const char * const key, 96 | const char * const value); 97 | 98 | // the u64 key is filled in big-endian byte order 99 | extern void 100 | kv_refill_u64(struct kv * const kv, const u64 key, const void * const value, const u32 vlen); 101 | 102 | extern void 103 | kv_refill_hex32(struct kv * const kv, const u32 hex, const void * const value, const u32 vlen); 104 | 105 | extern void 106 | kv_refill_hex64(struct kv * const kv, const u64 hex, const void * const value, const u32 vlen); 107 | 108 | extern void 109 | kv_refill_hex64_klen(struct kv * const kv, const u64 hex, const u32 klen, 110 | const void * const value, const u32 vlen); 111 | 112 | extern void 113 | kv_refill_kref(struct kv * const kv, const struct kref * const kref); 114 | 115 | extern void 116 | kv_refill_kref_v(struct kv * const kv, const struct kref * const kref, 117 | const void * const value, const u32 vlen); 118 | 119 | extern struct kref 120 | kv_kref(const struct kv * const key); 121 | 122 | extern struct kv * 123 | kv_create(const void * const key, const u32 klen, const void * const value, const u32 vlen); 124 | 125 | extern struct kv * 126 | kv_create_str(const char * const key, const void * const value, const u32 vlen); 127 | 128 | extern struct kv * 129 | kv_create_str_str(const char * const key, const char * const value); 130 | 131 | extern struct kv * 132 | kv_create_kref(const struct kref * const kref, const void * const value, const u32 vlen); 133 | 134 | // a static kv with klen == 0 135 | extern const struct kv * 136 | kv_null(void); 137 | 138 | extern struct kv * 139 | kv_dup(const struct kv * const kv); 140 | 141 | extern struct kv * 142 | kv_dup_key(const struct kv * const kv); 143 | 144 | extern struct kv * 145 | kv_dup2(const struct kv * const from, struct kv * const to); 146 | 147 | extern struct kv * 148 | kv_dup2_key(const struct kv * const from, struct kv * const to); 149 | 150 | extern struct kv * 151 | kv_dup2_key_prefix(const struct kv * const from, struct kv * const to, const u32 plen); 152 | 153 | extern bool 154 | kv_match(const struct kv * const key1, const struct kv * const key2); 155 | 156 | extern bool 157 | kv_match_hash(const struct kv * const key1, const struct kv * const key2); 158 | 159 | extern bool 160 | kv_match_full(const struct kv * const kv1, const struct kv * const kv2); 161 | 162 | extern bool 163 | kv_match_kv128(const struct kv * const sk, const u8 * const kv128); 164 | 165 | extern int 166 | kv_compare(const struct kv * const kv1, const struct kv * const kv2); 167 | 168 | extern int 169 | kv_k128_compare(const struct kv * const sk, const u8 * const k128); 170 | 171 | extern int 172 | kv_kv128_compare(const struct kv * const sk, const u8 * const kv128); 173 | 174 | extern void 175 | kv_qsort(struct kv ** const kvs, const size_t nr); 176 | 177 | extern u32 178 | kv_key_lcp(const struct kv * const key1, const struct kv * const key2); 179 | 180 | extern u32 181 | kv_key_lcp_skip(const struct kv * const key1, const struct kv * const key2, const u32 lcp0); 182 | 183 | extern void 184 | kv_psort(struct kv ** const kvs, const u64 nr, const u64 tlo, const u64 thi); 185 | 186 | extern void * 187 | kv_vptr(struct kv * const kv); 188 | 189 | extern void * 190 | kv_kptr(struct kv * const kv); 191 | 192 | extern const void * 193 | kv_vptr_c(const struct kv * const kv); 194 | 195 | extern const void * 196 | kv_kptr_c(const struct kv * const kv); 197 | 198 | extern void 199 | kv_print(const struct kv * const kv, const char * const cmd, FILE * const out); 200 | // }}} kv 201 | 202 | // mm {{{ 203 | typedef struct kv * (* kvmap_mm_in_func)(struct kv * kv, void * priv); 204 | typedef struct kv * (* kvmap_mm_out_func)(struct kv * kv, struct kv * out); 205 | typedef void (* kvmap_mm_free_func)(struct kv * kv, void * priv); 206 | 207 | // manage internal kv data of kvmap 208 | struct kvmap_mm { 209 | // to create a private copy of "kv" 210 | // see put() functions 211 | kvmap_mm_in_func in; 212 | // to duplicate a private copy of "kv" to "out" 213 | // see get() and iter_peek() functions 214 | kvmap_mm_out_func out; 215 | // to free a kv 216 | // see del() and put() functions 217 | kvmap_mm_free_func free; 218 | void * priv; 219 | }; 220 | 221 | extern struct kv * 222 | kvmap_mm_in_noop(struct kv * const kv, void * const priv); 223 | 224 | extern struct kv * 225 | kvmap_mm_out_noop(struct kv * const kv, struct kv * const out); 226 | 227 | extern void 228 | kvmap_mm_free_noop(struct kv * const kv, void * const priv); 229 | 230 | extern struct kv * 231 | kvmap_mm_in_dup(struct kv * const kv, void * const priv); 232 | 233 | extern struct kv * 234 | kvmap_mm_out_dup(struct kv * const kv, struct kv * const out); 235 | 236 | extern void 237 | kvmap_mm_free_free(struct kv * const kv, void * const priv); 238 | 239 | // the default mm 240 | extern const struct kvmap_mm kvmap_mm_dup; // in:Dup, out:Dup, free:Free 241 | extern const struct kvmap_mm kvmap_mm_ndf; // in:Noop, out:Dup, free:Free 242 | // }}} mm 243 | 244 | // ref {{{ 245 | typedef int (*kref_kv_cmp_func)(const struct kref *, const struct kv *); 246 | 247 | // ptr and len only 248 | extern void 249 | kref_ref_raw(struct kref * const kref, const u8 * const ptr, const u32 len); 250 | 251 | // this calculates hash32 252 | extern void 253 | kref_ref_hash32(struct kref * const kref, const u8 * const ptr, const u32 len); 254 | 255 | extern void 256 | kref_update_hash32(struct kref * const kref); 257 | 258 | extern void 259 | kref_ref_kv(struct kref * const kref, const struct kv * const kv); 260 | 261 | extern void 262 | kref_ref_kv_hash32(struct kref * const kref, const struct kv * const kv); 263 | 264 | extern bool 265 | kref_match(const struct kref * const k1, const struct kref * const k2); 266 | 267 | extern bool 268 | kref_kv_match(const struct kref * const kref, const struct kv * const k); 269 | 270 | extern int 271 | kref_compare(const struct kref * const kref1, const struct kref * const kref2); 272 | 273 | extern int 274 | kref_kv_compare(const struct kref * const kref, const struct kv * const k); 275 | 276 | extern u32 277 | kref_lcp(const struct kref * const k1, const struct kref * const k2); 278 | 279 | extern u32 280 | kref_kv_lcp(const struct kref * const kref, const struct kv * const kv); 281 | 282 | extern int 283 | kref_k128_compare(const struct kref * const sk, const u8 * const k128); 284 | 285 | extern int 286 | kref_kv128_compare(const struct kref * const sk, const u8 * const kv128); 287 | 288 | extern const struct kref * 289 | kref_null(void); 290 | 291 | extern void 292 | kvref_ref_kv(struct kvref * const ref, struct kv * const kv); 293 | 294 | extern struct kv * 295 | kvref_dup2_kv(struct kvref * const ref, struct kv * const to); 296 | 297 | extern struct kv * 298 | kvref_dup2_key(struct kvref * const ref, struct kv * const to); 299 | 300 | extern int 301 | kvref_kv_compare(const struct kvref * const ref, const struct kv * const kv); 302 | // }}} ref 303 | 304 | // kv128 {{{ 305 | extern size_t 306 | kv128_estimate_kv(const struct kv * const kv); 307 | 308 | extern u8 * 309 | kv128_encode_kv(const struct kv * const kv, u8 * const out, size_t * const pesize); 310 | 311 | extern struct kv * 312 | kv128_decode_kv(const u8 * const ptr, struct kv * const out, size_t * const pesize); 313 | 314 | extern size_t 315 | kv128_size(const u8 * const ptr); 316 | // }}} kv128 317 | 318 | // }}} kv 319 | 320 | // kvmap {{{ 321 | 322 | // kvmap_api {{{ 323 | typedef void (* kv_inp_func)(struct kv * const curr, void * const priv); 324 | 325 | // the merge function should: 326 | // 1: return NULL if the origin kv is not changed at all 327 | // 2: return kv0 if updates has been applied in-place 328 | // 3: return a different kv if the original kv must be replaced 329 | // In an in-memory kvmap, 2==1 and no further action is needed 330 | // In a persistent kv store with a memtable, 2 will need an insertion if kv0 is not from the memtable 331 | typedef struct kv * (* kv_merge_func)(struct kv * const kv0, void * const priv); 332 | 333 | struct kvmap_api { 334 | // feature bits 335 | bool hashkey; // true: caller needs to provide correct hash in kv/kref 336 | bool ordered; // true: has iter_seek 337 | bool threadsafe; // true: support thread_safe access 338 | bool readonly; // true: no put() and del() 339 | bool irefsafe; // true: iter's kref/kvref can be safely accessed after iter_seek/iter_skip/iter_park 340 | bool unique; // provide unique keys, especially for iterators 341 | bool refpark; // ref has park() and resume() 342 | bool async; // XXX for testing KVell 343 | 344 | // put (aka put/upsert): return true on success; false on error 345 | // mm.in() controls how things move into the kvmap; the default mm make a copy with malloc() 346 | // mm.free() controls how old kv get disposed when replaced 347 | bool (* put) (void * const ref, struct kv * const kv); 348 | // get: search and return a kv if found, or NULL if not 349 | // with the default mm: malloc() if out == NULL; otherwise, use out as buffer 350 | // with custom kvmap_mm: mm.out() controls buffer; use with caution 351 | // caller should use the returned ptr even if out is provided 352 | struct kv * (* get) (void * const ref, const struct kref * const key, struct kv * const out); 353 | // probe: return true on found, false on not found 354 | bool (* probe) (void * const ref, const struct kref * const key); 355 | // del: return true on something deleted, false on not found 356 | // mm.free() controls how old kv get disposed when replaced 357 | bool (* del) (void * const ref, const struct kref * const key); 358 | // inp: inplace operation if key exists; otherwise return false; uf() is always executed even with NULL key 359 | // inpr/inpw acquires r/w locks respectively. 360 | // Note that in inpw() you can only change the value. 361 | bool (* inpr) (void * const ref, const struct kref * const key, kv_inp_func uf, void * const priv); 362 | bool (* inpw) (void * const ref, const struct kref * const key, kv_inp_func uf, void * const priv); 363 | // merge: put+callback on old/new keys; another name: read-modify-write 364 | // return true if successfull; return false on error 365 | bool (* merge) (void * const ref, const struct kref * const key, kv_merge_func uf, void * const priv); 366 | // delete-range: delete all keys from start (inclusive) to end (exclusive) 367 | u64 (* delr) (void * const ref, const struct kref * const start, const struct kref * const end); 368 | // make everything persist; for persistent maps only 369 | void (* sync) (void * const ref); 370 | 371 | // general guidelines for thread-safe iters: 372 | // - it is assumed that the key under the cursor is locked/freezed/immutable 373 | // - once created one must call iter_seek to make it valid 374 | // - the ownership of ref is given to the iter so ref should not be used until iter_destroy 375 | // - creating and use more than one iter based on a ref can cause deadlocks 376 | void * (* iter_create) (void * const ref); 377 | // move the cursor to the first key >= search-key; 378 | void (* iter_seek) (void * const iter, const struct kref * const key); 379 | // check if the cursor points to a valid key 380 | bool (* iter_valid) (void * const iter); 381 | // return the current key; copy to out if (out != NULL) 382 | // mm.out() controls copy-out 383 | struct kv * (* iter_peek) (void * const iter, struct kv * const out); 384 | // similar to peek but does not copy; return false if iter is invalid 385 | bool (* iter_kref) (void * const iter, struct kref * const kref); 386 | // similar to iter_kref but also provide the value 387 | bool (* iter_kvref) (void * const iter, struct kvref * const kvref); 388 | // iter_retain makes kref or kvref of the current iter remain valid until released 389 | // the returned opaque pointer should be provided when releasing the hold 390 | u64 (* iter_retain) (void * const iter); 391 | void (* iter_release) (void * const iter, const u64 opaque); 392 | // skip one element 393 | void (* iter_skip1) (void * const iter); 394 | // skip nr elements 395 | void (* iter_skip) (void * const iter, const u32 nr); 396 | // iter_next == iter_peek + iter_skip1 397 | struct kv * (* iter_next) (void * const iter, struct kv * const out); 398 | // perform inplace opeation if the current key is valid; return false if no current key 399 | // the uf() is always executed even with NULL key 400 | bool (* iter_inp) (void * const iter, kv_inp_func uf, void * const priv); 401 | // invalidate the iter to release any resources or locks 402 | // afterward, must call seek() again before accessing data 403 | void (* iter_park) (void * const iter); 404 | // destroy iter 405 | void (* iter_destroy) (void * const iter); 406 | 407 | // misc: 408 | // create refs for maps if required; always use use kvmap_ref() and kvmap_unref() 409 | // if there are ref/unref functions, ref-ptr should be used as map for all kv operations 410 | void * (* ref) (void * map); 411 | // return the original map 412 | void * (* unref) (void * ref); 413 | // pause access without unref; must call resume later before access index again 414 | void (* park) (void * ref); 415 | // resume access of ref; must be paired with a park() 416 | void (* resume) (void * ref); 417 | 418 | // UNSAFE functions: 419 | // empty the map 420 | void (* clean) (void * map); 421 | // erase everything 422 | void (* destroy) (void * map); 423 | // for debugging 424 | void (* fprint) (void * map, FILE * const out); 425 | }; 426 | 427 | // registry 428 | struct kvmap_api_reg { 429 | int nargs; // number of arguments after name 430 | const char * name; 431 | const char * args_msg; // see ...helper_message 432 | // multiple apis may share one create function 433 | // arguments: name (e.g., "rdb"), mm (usually NULL), the remaining args 434 | void * (*create)(const char *, const struct kvmap_mm *, char **); 435 | const struct kvmap_api * api; 436 | }; 437 | 438 | // call this function to register a kvmap_api 439 | extern void 440 | kvmap_api_register(const int nargs, const char * const name, const char * const args_msg, 441 | void * (*create)(const char *, const struct kvmap_mm *, char **), const struct kvmap_api * const api); 442 | 443 | extern void 444 | kvmap_api_helper_message(void); 445 | 446 | extern int 447 | kvmap_api_helper(int argc, char ** const argv, const struct kvmap_mm * const mm, 448 | const struct kvmap_api ** const api_out, void ** const map_out); 449 | // }}} kvmap_api 450 | 451 | // helpers {{{ 452 | extern void 453 | kvmap_inp_steal_kv(struct kv * const kv, void * const priv); 454 | 455 | extern void * 456 | kvmap_ref(const struct kvmap_api * const api, void * const map); 457 | 458 | extern void * 459 | kvmap_unref(const struct kvmap_api * const api, void * const ref); 460 | 461 | extern struct kv * 462 | kvmap_kv_get(const struct kvmap_api * const api, void * const ref, 463 | const struct kv * const key, struct kv * const out); 464 | 465 | extern bool 466 | kvmap_kv_probe(const struct kvmap_api * const api, void * const ref, 467 | const struct kv * const key); 468 | 469 | extern bool 470 | kvmap_kv_put(const struct kvmap_api * const api, void * const ref, 471 | struct kv * const kv); 472 | 473 | extern bool 474 | kvmap_kv_del(const struct kvmap_api * const api, void * const ref, 475 | const struct kv * const key); 476 | 477 | extern bool 478 | kvmap_kv_inpr(const struct kvmap_api * const api, void * const ref, 479 | const struct kv * const key, kv_inp_func uf, void * const priv); 480 | 481 | extern bool 482 | kvmap_kv_inpw(const struct kvmap_api * const api, void * const ref, 483 | const struct kv * const key, kv_inp_func uf, void * const priv); 484 | 485 | extern bool 486 | kvmap_kv_merge(const struct kvmap_api * const api, void * const ref, 487 | const struct kv * const key, kv_merge_func uf, void * const priv); 488 | 489 | extern u64 490 | kvmap_kv_delr(const struct kvmap_api * const api, void * const ref, 491 | const struct kv * const start, const struct kv * const end); 492 | 493 | extern void 494 | kvmap_kv_iter_seek(const struct kvmap_api * const api, void * const iter, 495 | const struct kv * const key); 496 | 497 | extern struct kv * 498 | kvmap_raw_get(const struct kvmap_api * const api, void * const ref, 499 | const u32 len, const u8 * const ptr, struct kv * const out); 500 | 501 | extern bool 502 | kvmap_raw_probe(const struct kvmap_api * const api, void * const ref, 503 | const u32 len, const u8 * const ptr); 504 | 505 | extern bool 506 | kvmap_raw_del(const struct kvmap_api * const api, void * const ref, 507 | const u32 len, const u8 * const ptr); 508 | 509 | extern bool 510 | kvmap_raw_inpr(const struct kvmap_api * const api, void * const ref, 511 | const u32 len, const u8 * const ptr, kv_inp_func uf, void * const priv); 512 | 513 | extern bool 514 | kvmap_raw_inpw(const struct kvmap_api * const api, void * const ref, 515 | const u32 len, const u8 * const ptr, kv_inp_func uf, void * const priv); 516 | 517 | extern void 518 | kvmap_raw_iter_seek(const struct kvmap_api * const api, void * const iter, 519 | const u32 len, const u8 * const ptr); 520 | 521 | extern u64 522 | kvmap_dump_keys(const struct kvmap_api * const api, void * const map, const int fd); 523 | // }}} helpers 524 | 525 | // }}} kvmap 526 | 527 | // miter {{{ 528 | // general-purpose merging iterator 529 | // api functions: 530 | // REQUIRED: 531 | // - iter_create 532 | // - iter_seek 533 | // - iter_peek 534 | // - iter_skip 535 | // - iter_destroy 536 | // - iter_kref 537 | // - iter_kvref 538 | // OPTIONAL (api-specific): 539 | // - ref/unref 540 | // - iter_park 541 | // - resume/park (need also set api->refpark) 542 | // OPTIONAL (performance): 543 | // - api->unique (faster miter_skip_unique) 544 | // - iter_retain/iter_release (less memcpy) 545 | 546 | struct miter; 547 | 548 | extern struct miter * 549 | miter_create(void); 550 | 551 | // caller owns the ref and the iter; miter will not destroy them 552 | // using the iter or the ref with an active miter can lead to undefined behavior 553 | extern bool 554 | miter_add_iter(struct miter * const miter, const struct kvmap_api * const api, void * const ref, void * const iter); 555 | 556 | // caller owns the ref; miter will create and destroy the iter 557 | // using the underlying ref with an active miter can lead to undefined behavior 558 | extern void * 559 | miter_add_ref(struct miter * const miter, const struct kvmap_api * const api, void * const ref); 560 | 561 | // miter will take a ref of the map, create an iter, and clean up everything 562 | // be careful of using another ref/iter in the same thread 563 | extern void * 564 | miter_add(struct miter * const miter, const struct kvmap_api * const api, void * const map); 565 | 566 | extern u32 567 | miter_rank(struct miter * const miter); 568 | 569 | extern void 570 | miter_seek(struct miter * const miter, const struct kref * const key); 571 | 572 | extern void 573 | miter_kv_seek(struct miter * const miter, const struct kv * const key); 574 | 575 | extern bool 576 | miter_valid(struct miter * const miter); 577 | 578 | extern struct kv * 579 | miter_peek(struct miter * const miter, struct kv * const out); 580 | 581 | extern bool 582 | miter_kref(struct miter * const miter, struct kref * const kref); 583 | 584 | extern bool 585 | miter_kvref(struct miter * const miter, struct kvref * const kvref); 586 | 587 | extern void 588 | miter_skip1(struct miter * const miter); 589 | 590 | extern void 591 | miter_skip(struct miter * const miter, const u32 nr); 592 | 593 | extern struct kv * 594 | miter_next(struct miter * const miter, struct kv * const out); 595 | 596 | extern void 597 | miter_skip_unique(struct miter * const miter); 598 | 599 | extern struct kv * 600 | miter_next_unique(struct miter * const miter, struct kv * const out); 601 | 602 | extern void 603 | miter_park(struct miter * const miter); 604 | 605 | extern void 606 | miter_clean(struct miter * const miter); 607 | 608 | extern void 609 | miter_destroy(struct miter * const miter); 610 | // }}} miter 611 | 612 | #ifdef __cplusplus 613 | } 614 | #endif 615 | // vim:fdm=marker 616 | -------------------------------------------------------------------------------- /lib.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | // includes {{{ 9 | // C headers 10 | #include 11 | #include 12 | #include 13 | #include 14 | #include 15 | #include 16 | #include 17 | #include 18 | #include 19 | 20 | // POSIX headers 21 | #include 22 | #include 23 | #include 24 | 25 | // Linux headers 26 | #include 27 | #include 28 | #include 29 | #include 30 | 31 | // SIMD 32 | #if defined(__x86_64__) 33 | #include 34 | #elif defined(__aarch64__) 35 | #include 36 | #include 37 | #endif 38 | // }}} includes 39 | 40 | #ifdef __cplusplus 41 | extern "C" { 42 | #endif 43 | 44 | // types {{{ 45 | typedef char s8; 46 | typedef short s16; 47 | typedef int s32; 48 | typedef long s64; 49 | typedef __int128_t s128; 50 | static_assert(sizeof(s8) == 1, "sizeof(s8)"); 51 | static_assert(sizeof(s16) == 2, "sizeof(s16)"); 52 | static_assert(sizeof(s32) == 4, "sizeof(s32)"); 53 | static_assert(sizeof(s64) == 8, "sizeof(s64)"); 54 | static_assert(sizeof(s128) == 16, "sizeof(s128)"); 55 | 56 | typedef unsigned char u8; 57 | typedef unsigned short u16; 58 | typedef unsigned int u32; 59 | typedef unsigned long u64; 60 | typedef __uint128_t u128; 61 | static_assert(sizeof(u8) == 1, "sizeof(u8)"); 62 | static_assert(sizeof(u16) == 2, "sizeof(u16)"); 63 | static_assert(sizeof(u32) == 4, "sizeof(u32)"); 64 | static_assert(sizeof(u64) == 8, "sizeof(u64)"); 65 | static_assert(sizeof(u128) == 16, "sizeof(u128)"); 66 | 67 | #if defined(__x86_64__) 68 | typedef __m128i m128; 69 | #if defined(__AVX2__) 70 | typedef __m256i m256; 71 | #endif // __AVX2__ 72 | #if defined(__AVX512F__) 73 | typedef __m512i m512; 74 | #endif // __AVX512F__ 75 | #elif defined(__aarch64__) 76 | typedef uint8x16_t m128; 77 | #else 78 | #error Need x86_64 or AArch64. 79 | #endif 80 | // }}} types 81 | 82 | // defs {{{ 83 | #define likely(____x____) __builtin_expect(____x____, 1) 84 | #define unlikely(____x____) __builtin_expect(____x____, 0) 85 | 86 | // ansi colors 87 | // 3X:fg; 4X:bg; 9X:light fg; 10X:light bg; 88 | // X can be one of the following colors: 89 | // 0:black; 1:red; 2:green; 3:yellow; 90 | // 4:blue; 5:magenta; 6:cyan; 7:white; 91 | #define TERMCLR(____code____) "\x1b[" #____code____ "m" 92 | // }}} defs 93 | 94 | // const {{{ 95 | #define PGSZ ((4096lu)) 96 | // }}} const 97 | 98 | // math {{{ 99 | extern u64 100 | mhash64(const u64 v); 101 | 102 | extern u32 103 | mhash32(const u32 v); 104 | 105 | extern u64 106 | gcd64(u64 a, u64 b); 107 | // }}} math 108 | 109 | // random {{{ 110 | extern u64 111 | random_u64(void); 112 | 113 | extern void 114 | srandom_u64(const u64 seed); 115 | 116 | extern double 117 | random_double(void); 118 | // }}} random 119 | 120 | // timing {{{ 121 | extern u64 122 | time_nsec(void); 123 | 124 | extern double 125 | time_sec(void); 126 | 127 | extern u64 128 | time_diff_nsec(const u64 last); 129 | 130 | extern double 131 | time_diff_sec(const double last); 132 | 133 | extern void 134 | time_stamp(char * str, const size_t size); 135 | 136 | extern void 137 | time_stamp2(char * str, const size_t size); 138 | // }}} timing 139 | 140 | // cpucache {{{ 141 | extern void 142 | cpu_pause(void); 143 | 144 | extern void 145 | cpu_mfence(void); 146 | 147 | extern void 148 | cpu_cfence(void); 149 | 150 | extern void 151 | cpu_prefetch0(const void * const ptr); 152 | 153 | extern void 154 | cpu_prefetch1(const void * const ptr); 155 | 156 | extern void 157 | cpu_prefetch2(const void * const ptr); 158 | 159 | extern void 160 | cpu_prefetch3(const void * const ptr); 161 | 162 | extern void 163 | cpu_prefetchw(const void * const ptr); 164 | // }}} cpucache 165 | 166 | // crc32c {{{ 167 | extern u32 168 | crc32c_u8(const u32 crc, const u8 v); 169 | 170 | extern u32 171 | crc32c_u16(const u32 crc, const u16 v); 172 | 173 | extern u32 174 | crc32c_u32(const u32 crc, const u32 v); 175 | 176 | extern u32 177 | crc32c_u64(const u32 crc, const u64 v); 178 | 179 | // 1 <= nr <= 3 180 | extern u32 181 | crc32c_inc_123(const u8 * buf, u32 nr, u32 crc); 182 | 183 | // nr % 4 == 0 184 | extern u32 185 | crc32c_inc_x4(const u8 * buf, u32 nr, u32 crc); 186 | 187 | extern u32 188 | crc32c_inc(const u8 * buf, u32 nr, u32 crc); 189 | // }}} crc32c 190 | 191 | // debug {{{ 192 | extern void 193 | debug_break(void); 194 | 195 | extern void 196 | debug_backtrace(void); 197 | 198 | extern void 199 | watch_u64_usr1(u64 * const ptr); 200 | 201 | #ifndef NDEBUG 202 | extern void 203 | debug_assert(const bool v); 204 | #else 205 | #define debug_assert(expr) ((void)0) 206 | #endif 207 | 208 | __attribute__((noreturn)) 209 | extern void 210 | debug_die(void); 211 | 212 | __attribute__((noreturn)) 213 | extern void 214 | debug_die_perror(void); 215 | 216 | extern void 217 | debug_dump_maps(FILE * const out); 218 | 219 | extern bool 220 | debug_perf_switch(void); 221 | // }}} debug 222 | 223 | // mm {{{ 224 | #ifdef ALLOCFAIL 225 | extern bool 226 | alloc_fail(void); 227 | #endif 228 | 229 | extern void * 230 | xalloc(const size_t align, const size_t size); 231 | 232 | extern void * 233 | yalloc(const size_t size); 234 | 235 | extern void ** 236 | malloc_2d(const size_t nr, const size_t size); 237 | 238 | extern void ** 239 | calloc_2d(const size_t nr, const size_t size); 240 | 241 | extern void 242 | pages_unmap(void * const ptr, const size_t size); 243 | 244 | extern void 245 | pages_lock(void * const ptr, const size_t size); 246 | 247 | /* hugepages */ 248 | // force posix allocators: -DVALGRIND_MEMCHECK 249 | extern void * 250 | pages_alloc_4kb(const size_t nr_4kb); 251 | 252 | extern void * 253 | pages_alloc_2mb(const size_t nr_2mb); 254 | 255 | extern void * 256 | pages_alloc_1gb(const size_t nr_1gb); 257 | 258 | extern void * 259 | pages_alloc_best(const size_t size, const bool try_1gb, u64 * const size_out); 260 | // }}} mm 261 | 262 | // process/thread {{{ 263 | extern void 264 | thread_get_name(const pthread_t pt, char * const name, const size_t len); 265 | 266 | extern void 267 | thread_set_name(const pthread_t pt, const char * const name); 268 | 269 | extern long 270 | process_get_rss(void); 271 | 272 | extern u32 273 | process_affinity_count(void); 274 | 275 | extern u32 276 | process_getaffinity_list(const u32 max, u32 * const cores); 277 | 278 | extern void 279 | thread_setaffinity_list(const u32 nr, const u32 * const list); 280 | 281 | extern void 282 | thread_pin(const u32 cpu); 283 | 284 | extern u64 285 | process_cpu_time_usec(void); 286 | 287 | // if args == true, argx is void ** 288 | // if args == false, argx is void * 289 | extern u64 290 | thread_fork_join(u32 nr, void *(*func) (void *), const bool args, void * const argx); 291 | 292 | extern int 293 | thread_create_at(const u32 cpu, pthread_t * const thread, void *(*start_routine) (void *), void * const arg); 294 | // }}} process/thread 295 | 296 | // locking {{{ 297 | typedef union { 298 | u32 opaque; 299 | } spinlock; 300 | 301 | extern void 302 | spinlock_init(spinlock * const lock); 303 | 304 | extern void 305 | spinlock_lock(spinlock * const lock); 306 | 307 | extern bool 308 | spinlock_trylock(spinlock * const lock); 309 | 310 | extern void 311 | spinlock_unlock(spinlock * const lock); 312 | 313 | typedef union { 314 | u32 opaque; 315 | } rwlock; 316 | 317 | extern void 318 | rwlock_init(rwlock * const lock); 319 | 320 | extern bool 321 | rwlock_trylock_read(rwlock * const lock); 322 | 323 | // low-priority reader-lock; use with trylock_write_hp 324 | extern bool 325 | rwlock_trylock_read_lp(rwlock * const lock); 326 | 327 | extern bool 328 | rwlock_trylock_read_nr(rwlock * const lock, u16 nr); 329 | 330 | extern void 331 | rwlock_lock_read(rwlock * const lock); 332 | 333 | extern void 334 | rwlock_unlock_read(rwlock * const lock); 335 | 336 | extern bool 337 | rwlock_trylock_write(rwlock * const lock); 338 | 339 | extern bool 340 | rwlock_trylock_write_nr(rwlock * const lock, u16 nr); 341 | 342 | extern void 343 | rwlock_lock_write(rwlock * const lock); 344 | 345 | // writer has higher priority; new readers are blocked 346 | extern bool 347 | rwlock_trylock_write_hp(rwlock * const lock); 348 | 349 | extern bool 350 | rwlock_trylock_write_hp_nr(rwlock * const lock, u16 nr); 351 | 352 | extern void 353 | rwlock_lock_write_hp(rwlock * const lock); 354 | 355 | extern void 356 | rwlock_unlock_write(rwlock * const lock); 357 | 358 | extern void 359 | rwlock_write_to_read(rwlock * const lock); 360 | 361 | typedef union { 362 | u64 opqaue[8]; 363 | } mutex; 364 | 365 | extern void 366 | mutex_init(mutex * const lock); 367 | 368 | extern void 369 | mutex_lock(mutex * const lock); 370 | 371 | extern bool 372 | mutex_trylock(mutex * const lock); 373 | 374 | extern void 375 | mutex_unlock(mutex * const lock); 376 | 377 | extern void 378 | mutex_deinit(mutex * const lock); 379 | // }}} locking 380 | 381 | // coroutine {{{ 382 | extern u64 co_switch_stack(u64 * const saversp, const u64 newrsp, const u64 retval); 383 | 384 | struct co; 385 | 386 | extern struct co * 387 | co_create(const u64 stacksize, void * func, void * priv, u64 * const host); 388 | 389 | extern void 390 | co_reuse(struct co * const co, void * func, void * priv, u64 * const host); 391 | 392 | extern struct co * 393 | co_fork(void * func, void * priv); 394 | 395 | extern void * 396 | co_priv(void); 397 | 398 | extern u64 399 | co_enter(struct co * const to, const u64 retval); 400 | 401 | extern u64 402 | co_switch_to(struct co * const to, const u64 retval); 403 | 404 | extern u64 405 | co_back(const u64 retval); 406 | 407 | extern void 408 | co_exit(const u64 retval); 409 | 410 | extern bool 411 | co_valid(struct co * const co); 412 | 413 | extern struct co * 414 | co_self(void); 415 | 416 | extern void 417 | co_destroy(struct co * const co); 418 | 419 | struct corr; 420 | 421 | extern struct corr * 422 | corr_create(const u64 stacksize, void * func, void * priv, u64 * const host); 423 | 424 | extern struct corr * 425 | corr_link(const u64 stacksize, void * func, void * priv, struct corr * const prev); 426 | 427 | extern void 428 | corr_reuse(struct corr * const co, void * func, void * priv, u64 * const host); 429 | 430 | extern void 431 | corr_relink(struct corr * const co, void * func, void * priv, struct corr * const prev); 432 | 433 | extern void 434 | corr_enter(struct corr * const co); 435 | 436 | extern void 437 | corr_yield(void); 438 | 439 | extern void 440 | corr_exit(void); 441 | 442 | extern void 443 | corr_destroy(struct corr * const co); 444 | // }}} coroutine 445 | 446 | // bits {{{ 447 | extern u32 448 | bits_reverse_u32(const u32 v); 449 | 450 | extern u64 451 | bits_reverse_u64(const u64 v); 452 | 453 | extern u64 454 | bits_rotl_u64(const u64 v, const u8 n); 455 | 456 | extern u64 457 | bits_rotr_u64(const u64 v, const u8 n); 458 | 459 | extern u32 460 | bits_rotl_u32(const u32 v, const u8 n); 461 | 462 | extern u32 463 | bits_rotr_u32(const u32 v, const u8 n); 464 | 465 | extern u64 466 | bits_p2_up_u64(const u64 v); 467 | 468 | extern u32 469 | bits_p2_up_u32(const u32 v); 470 | 471 | extern u64 472 | bits_p2_down_u64(const u64 v); 473 | 474 | extern u32 475 | bits_p2_down_u32(const u32 v); 476 | 477 | extern u64 478 | bits_round_up(const u64 v, const u8 power); 479 | 480 | extern u64 481 | bits_round_up_a(const u64 v, const u64 a); 482 | 483 | extern u64 484 | bits_round_down(const u64 v, const u8 power); 485 | 486 | extern u64 487 | bits_round_down_a(const u64 v, const u64 a); 488 | // }}} bits 489 | 490 | // simd {{{ 491 | extern u32 492 | m128_movemask_u8(const m128 v); 493 | 494 | // extern u32 495 | //m128_movemask_u16(const m128 v); 496 | // 497 | // extern u32 498 | //m128_movemask_u32(const m128 v); 499 | // }}} simd 500 | 501 | // vi128 {{{ 502 | extern u32 503 | vi128_estimate_u32(const u32 v); 504 | 505 | extern u8 * 506 | vi128_encode_u32(u8 * dst, u32 v); 507 | 508 | extern const u8 * 509 | vi128_decode_u32(const u8 * src, u32 * const out); 510 | 511 | extern u32 512 | vi128_estimate_u64(const u64 v); 513 | 514 | extern u8 * 515 | vi128_encode_u64(u8 * dst, u64 v); 516 | 517 | extern const u8 * 518 | vi128_decode_u64(const u8 * src, u64 * const out); 519 | // }}} vi128 520 | 521 | // misc {{{ 522 | // TODO: only works on little endian? 523 | struct entry13 { // what a beautiful name 524 | union { 525 | u16 e1; 526 | struct { // easy for debugging 527 | u64 e1_64:16; 528 | u64 e3:48; 529 | }; 530 | u64 v64; 531 | void * ptr; 532 | }; 533 | }; 534 | 535 | static_assert(sizeof(struct entry13) == 8, "sizeof(entry13) != 8"); 536 | 537 | // directly access read .e1 and .e3 538 | // directly write .e1 539 | // use entry13_update() to update the entire entry 540 | 541 | extern struct entry13 542 | entry13(const u16 e1, const u64 e3); 543 | 544 | extern void 545 | entry13_update_e3(struct entry13 * const e, const u64 e3); 546 | 547 | extern void * 548 | u64_to_ptr(const u64 v); 549 | 550 | extern u64 551 | ptr_to_u64(const void * const ptr); 552 | 553 | extern size_t 554 | m_usable_size(void * const ptr); 555 | 556 | extern size_t 557 | fdsize(const int fd); 558 | 559 | extern u32 560 | memlcp(const u8 * const p1, const u8 * const p2, const u32 max); 561 | 562 | __attribute__ ((format (printf, 2, 3))) 563 | extern void 564 | logger_printf(const int fd, const char * const fmt, ...); 565 | // }}} misc 566 | 567 | // bitmap {{{ 568 | struct bitmap; 569 | 570 | extern struct bitmap * 571 | bitmap_create(const u64 nbits); 572 | 573 | extern void 574 | bitmap_init(struct bitmap * const bm, const u64 nbits); 575 | 576 | extern bool 577 | bitmap_test(const struct bitmap * const bm, const u64 idx); 578 | 579 | extern bool 580 | bitmap_test_all1(struct bitmap * const bm); 581 | 582 | extern bool 583 | bitmap_test_all0(struct bitmap * const bm); 584 | 585 | extern void 586 | bitmap_set1(struct bitmap * const bm, const u64 idx); 587 | 588 | extern void 589 | bitmap_set0(struct bitmap * const bm, const u64 idx); 590 | 591 | extern void 592 | bitmap_set1_safe64(struct bitmap * const bm, const u64 idx); 593 | 594 | extern void 595 | bitmap_set0_safe64(struct bitmap * const bm, const u64 idx); 596 | 597 | extern u64 598 | bitmap_count(struct bitmap * const bm); 599 | 600 | extern u64 601 | bitmap_first(struct bitmap * const bm); 602 | 603 | extern void 604 | bitmap_set_all1(struct bitmap * const bm); 605 | 606 | extern void 607 | bitmap_set_all0(struct bitmap * const bm); 608 | // }}} bitmap 609 | 610 | // slab {{{ 611 | struct slab; 612 | 613 | extern struct slab * 614 | slab_create(const u64 obj_size, const u64 blk_size); 615 | 616 | extern bool 617 | slab_reserve_unsafe(struct slab * const slab, const u64 nr); 618 | 619 | extern void * 620 | slab_alloc_unsafe(struct slab * const slab); 621 | 622 | extern void * 623 | slab_alloc_safe(struct slab * const slab); 624 | 625 | extern void 626 | slab_free_unsafe(struct slab * const slab, void * const ptr); 627 | 628 | extern void 629 | slab_free_safe(struct slab * const slab, void * const ptr); 630 | 631 | extern void 632 | slab_free_all(struct slab * const slab); 633 | 634 | extern u64 635 | slab_get_nalloc(struct slab * const slab); 636 | 637 | extern void 638 | slab_destroy(struct slab * const slab); 639 | // }}} slab 640 | 641 | // qsort {{{ 642 | extern int 643 | compare_u16(const void * const p1, const void * const p2); 644 | 645 | extern void 646 | qsort_u16(u16 * const array, const size_t nr); 647 | 648 | extern u16 * 649 | bsearch_u16(const u16 v, const u16 * const array, const size_t nr); 650 | 651 | extern void 652 | shuffle_u16(u16 * const array, const u64 nr); 653 | 654 | extern int 655 | compare_u32(const void * const p1, const void * const p2); 656 | 657 | extern void 658 | qsort_u32(u32 * const array, const size_t nr); 659 | 660 | extern u32 * 661 | bsearch_u32(const u32 v, const u32 * const array, const size_t nr); 662 | 663 | extern void 664 | shuffle_u32(u32 * const array, const u64 nr); 665 | 666 | extern int 667 | compare_u64(const void * const p1, const void * const p2); 668 | 669 | extern void 670 | qsort_u64(u64 * const array, const size_t nr); 671 | 672 | extern u64 * 673 | bsearch_u64(const u64 v, const u64 * const array, const size_t nr); 674 | 675 | extern void 676 | shuffle_u64(u64 * const array, const u64 nr); 677 | 678 | extern int 679 | compare_double(const void * const p1, const void * const p2); 680 | 681 | extern void 682 | qsort_double(double * const array, const size_t nr); 683 | 684 | extern void 685 | qsort_u64_sample(const u64 * const array0, const u64 nr, const u64 res, FILE * const out); 686 | 687 | extern void 688 | qsort_double_sample(const double * const array0, const u64 nr, const u64 res, FILE * const out); 689 | // }}} qsort 690 | 691 | // xlog {{{ 692 | struct xlog; 693 | 694 | extern struct xlog * 695 | xlog_create(const u64 nr_init, const u64 unit_size); 696 | 697 | extern void 698 | xlog_append(struct xlog * const xlog, const void * const rec); 699 | 700 | extern void 701 | xlog_append_cycle(struct xlog * const xlog, const void * const rec); 702 | 703 | extern void 704 | xlog_reset(struct xlog * const xlog); 705 | 706 | extern u64 707 | xlog_read(struct xlog * const xlog, void * const buf, const u64 nr_max); 708 | 709 | extern void 710 | xlog_dump(struct xlog * const xlog, FILE * const out); 711 | 712 | extern void 713 | xlog_destroy(struct xlog * const xlog); 714 | 715 | struct xlog_iter; 716 | 717 | extern struct xlog_iter * 718 | xlog_iter_create(const struct xlog * const xlog); 719 | 720 | extern bool 721 | xlog_iter_next(struct xlog_iter * const iter, void * const out); 722 | // free iter after use 723 | // }}} xlog 724 | 725 | // string {{{ 726 | // XXX strdec_ and strhex_ functions does not append the trailing '\0' to the output string 727 | // size of out should be >= 10 728 | extern void 729 | strdec_32(void * const out, const u32 v); 730 | 731 | // size of out should be >= 20 732 | extern void 733 | strdec_64(void * const out, const u64 v); 734 | 735 | // size of out should be >= 8 736 | extern void 737 | strhex_32(void * const out, const u32 v); 738 | 739 | // size of out should be >= 16 740 | extern void 741 | strhex_64(void * const out, const u64 v); 742 | 743 | extern u64 744 | a2u64(const void * const str); 745 | 746 | extern u32 747 | a2u32(const void * const str); 748 | 749 | extern s64 750 | a2s64(const void * const str); 751 | 752 | extern s32 753 | a2s32(const void * const str); 754 | 755 | extern void 756 | str_print_hex(FILE * const out, const void * const data, const u32 len); 757 | 758 | extern void 759 | str_print_dec(FILE * const out, const void * const data, const u32 len); 760 | 761 | // user should free returned ptr (and nothing else) after use 762 | extern char ** 763 | strtoks(const char * const str, const char * const delim); 764 | 765 | extern u32 766 | strtoks_count(const char * const * const toks); 767 | // }}} string 768 | 769 | // qsbr {{{ 770 | // QSBR vs EBR (Quiescent-State vs Epoch Based Reclaimation) 771 | // QSBR: readers just use qsbr_update -> qsbr_update -> ... repeatedly 772 | // EBR: readers use qsbr_update -> qsbr_park -> qsbr_resume -> qsbr_update -> ... 773 | // The advantage of EBR is qsbr_park can happen much earlier than the next qsbr_update 774 | // The disadvantage is the extra cost, a pair of park/resume is used in every iteration 775 | struct qsbr; 776 | struct qsbr_ref { 777 | #ifdef QSBR_DEBUG 778 | u64 debug[16]; 779 | #endif 780 | u64 opaque[3]; 781 | }; 782 | 783 | extern struct qsbr * 784 | qsbr_create(void); 785 | 786 | // every READER accessing the shared data must first register itself with the qsbr 787 | extern bool 788 | qsbr_register(struct qsbr * const q, struct qsbr_ref * const qref); 789 | 790 | extern void 791 | qsbr_unregister(struct qsbr * const q, struct qsbr_ref * const qref); 792 | 793 | // For READER: mark the beginning of critical section; like rcu_read_lock() 794 | extern void 795 | qsbr_update(struct qsbr_ref * const qref, const u64 v); 796 | 797 | // temporarily stop access the shared data to avoid blocking writers 798 | // READER can use qsbr_park (like rcu_read_unlock()) in conjunction with qsbr_update 799 | // qsbr_park is roughly equivalent to qsbr_unregister, but faster 800 | extern void 801 | qsbr_park(struct qsbr_ref * const qref); 802 | 803 | // undo the effect of qsbr_park; must use it between qsbr_park and qsbr_update 804 | // qsbr_resume is roughly equivalent to qsbr_register, but faster 805 | extern void 806 | qsbr_resume(struct qsbr_ref * const qref); 807 | 808 | // WRITER: wait until all the readers have announced v=target with qsbr_update 809 | extern void 810 | qsbr_wait(struct qsbr * const q, const u64 target); 811 | 812 | extern void 813 | qsbr_destroy(struct qsbr * const q); 814 | // }}} qsbr 815 | 816 | #ifdef __cplusplus 817 | } 818 | #endif 819 | // vim:fdm=marker 820 | -------------------------------------------------------------------------------- /remixdb.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | #include 9 | #include 10 | 11 | #ifdef __cplusplus 12 | extern "C" { 13 | #endif 14 | 15 | struct xdb; 16 | struct xdb_ref; 17 | struct xdb_iter; 18 | 19 | extern struct xdb * 20 | remixdb_open(const char * const dir, const size_t cache_size_mb, const size_t mt_size_mb); 21 | 22 | extern struct xdb * 23 | remixdb_open_compact(const char * const dir, const size_t cache_size_mb, const size_t mt_size_mb); 24 | 25 | extern struct xdb_ref * 26 | remixdb_ref(struct xdb * const xdb); 27 | 28 | extern void 29 | remixdb_unref(struct xdb_ref * const ref); 30 | 31 | extern void 32 | remixdb_close(struct xdb * const xdb); 33 | 34 | extern bool 35 | remixdb_set(struct xdb_ref * const ref, const void * const kbuf, const uint32_t klen, 36 | const void * const vbuf, const uint32_t vlen); 37 | 38 | extern bool 39 | remixdb_del(struct xdb_ref * const ref, const void * const kbuf, const uint32_t klen); 40 | 41 | extern bool 42 | remixdb_probe(struct xdb_ref * const ref, const void * const kbuf, const uint32_t klen); 43 | 44 | extern bool 45 | remixdb_get(struct xdb_ref * const ref, const void * const kbuf, const uint32_t klen, 46 | void * const vbuf_out, uint32_t * const vlen_out); 47 | 48 | extern void 49 | remixdb_sync(struct xdb_ref * const ref); 50 | 51 | extern struct xdb_iter * 52 | remixdb_iter_create(struct xdb_ref * const ref); 53 | 54 | extern void 55 | remixdb_iter_seek(struct xdb_iter * const iter, const void * const kbuf, const uint32_t klen); 56 | 57 | extern bool 58 | remixdb_iter_valid(struct xdb_iter * const iter); 59 | 60 | extern bool 61 | remixdb_iter_peek(struct xdb_iter * const iter, 62 | void * const kbuf_out, uint32_t * const klen_out, 63 | void * const vbuf_out, uint32_t * const vlen_out); 64 | 65 | extern void 66 | remixdb_iter_skip(struct xdb_iter * const iter, const uint32_t nr); 67 | 68 | extern void 69 | remixdb_iter_park(struct xdb_iter * const iter); 70 | 71 | extern void 72 | remixdb_iter_destroy(struct xdb_iter * const iter); 73 | 74 | #ifdef __cplusplus 75 | } 76 | #endif 77 | // vim:fdm=marker 78 | -------------------------------------------------------------------------------- /remixdb.strip: -------------------------------------------------------------------------------- 1 | -K remixdb_open 2 | -K remixdb_open_compact 3 | -K remixdb_close 4 | -K remixdb_ref 5 | -K remixdb_unref 6 | -K remixdb_put 7 | -K remixdb_del 8 | -K remixdb_get 9 | -K remixdb_probe 10 | -K remixdb_sync 11 | -K remixdb_iter_create 12 | -K remixdb_iter_destroy 13 | -K remixdb_iter_valid 14 | -K remixdb_iter_seek 15 | -K remixdb_iter_park 16 | -K remixdb_iter_peek 17 | -K remixdb_iter_skip 18 | -------------------------------------------------------------------------------- /sotest.c: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #include 7 | #include 8 | 9 | #include "remixdb.h" 10 | 11 | int 12 | main(int argc, char ** argv) 13 | { 14 | (void)argc; 15 | (void)argv; 16 | struct xdb * const xdb = remixdb_open("/tmp/xdbdemo", 256, 256); // blockcache=256MB, MemTable=256MB 17 | struct xdb_ref * const ref = remixdb_ref(xdb); 18 | 19 | bool r; 20 | 21 | r = remixdb_set(ref, "remix", 5, "easy", 4); 22 | printf("remixdb_set remix easy %c\n", r?'T':'F'); 23 | 24 | r = remixdb_set(ref, "time_travel", 11, "impossible", 10); 25 | printf("remixdb_set time_travel impossible %c\n", r?'T':'F'); 26 | 27 | r = remixdb_del(ref, "time_travel", 11); 28 | printf("remixdb_del time_travel %c\n", r?'T':'F'); 29 | 30 | r = remixdb_probe(ref, "time_travel", 11); 31 | printf("remixdb_probe time_travel %c\n", r?'T':'F'); 32 | 33 | uint32_t klen_out = 0; 34 | char kbuf_out[8] = {}; 35 | uint32_t vlen_out = 0; 36 | char vbuf_out[8] = {}; 37 | r = remixdb_get(ref, "remix", 5, vbuf_out, &vlen_out); 38 | printf("remixdb_get remix %c %u %.*s\n", r?'T':'F', vlen_out, vlen_out, vbuf_out); 39 | 40 | // prepare a few keys for range ops 41 | r = remixdb_set(ref, "00", 2, "0_value", 7); 42 | r = remixdb_set(ref, "11", 2, "1_value", 7); 43 | r = remixdb_set(ref, "22", 2, "2_value", 7); 44 | 45 | struct xdb_iter * const iter = remixdb_iter_create(ref); 46 | 47 | remixdb_iter_seek(iter, NULL, 0); // seek to the head 48 | printf("remixdb_iter_seek \"\"\n"); 49 | while (remixdb_iter_valid(iter)) { 50 | r = remixdb_iter_peek(iter, kbuf_out, &klen_out, vbuf_out, &vlen_out); 51 | if (r) { 52 | printf("remixdb_iter_peek klen=%u key=%.*s vlen=%u value=%.*s\n", 53 | klen_out, klen_out, kbuf_out, vlen_out, vlen_out, vbuf_out); 54 | } else { 55 | printf("ERROR!\n"); 56 | } 57 | remixdb_iter_skip(iter, 1); 58 | } 59 | 60 | // call iter_park if you will go idle but want to use the iter later 61 | // don't need to call iter_park if you're actively using iter 62 | remixdb_iter_park(iter); 63 | usleep(10); 64 | 65 | remixdb_iter_seek(iter, "0", 1); 66 | printf("remixdb_iter_seek \"0\"\n"); 67 | // this time we don't want to copy the value 68 | r = remixdb_iter_peek(iter, kbuf_out, &klen_out, NULL, NULL); 69 | if (r){ 70 | printf("remixdb_iter_peek klen=%u key=%.*s\n", klen_out, klen_out, kbuf_out); 71 | } else { 72 | printf("ERROR: iter_peek failed\n"); 73 | } 74 | 75 | remixdb_iter_destroy(iter); 76 | // there must be no active iter when calling unref() 77 | remixdb_unref(ref); 78 | 79 | // unsafe operations: should have released all references 80 | remixdb_close(xdb); // destroy also calls clean interally 81 | return 0; 82 | } 83 | -------------------------------------------------------------------------------- /sst.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | #include "blkio.h" 9 | 10 | #ifdef __cplusplus 11 | extern "C" { 12 | #endif 13 | 14 | #define SST_VLEN_TS ((0x10000u)) // tomb stone 15 | #define SST_VLEN_MASK ((0xffffu)) // real vlen == vlen & 0xffff 16 | 17 | // kv {{{ 18 | extern size_t 19 | sst_kv_vi128_estimate(const struct kv * const kv); 20 | 21 | extern u8 * 22 | sst_kv_vi128_encode(u8 * ptr, const struct kv * const kv); 23 | 24 | extern size_t 25 | sst_kv_size(const struct kv * const kv); 26 | 27 | extern struct kv * 28 | sst_kvref_dup2_kv(struct kvref * const kvref, struct kv * const out); 29 | // }}} kv 30 | 31 | // mm {{{ 32 | 33 | extern struct kv * 34 | kvmap_mm_in_ts(struct kv * const kv, void * const priv); 35 | 36 | extern struct kv * 37 | kvmap_mm_out_ts(struct kv * const kv, struct kv * const out); 38 | 39 | extern const struct kvmap_mm kvmap_mm_ts; 40 | // }}} mm 41 | 42 | // sst {{{ 43 | struct sst; 44 | 45 | extern struct sst * 46 | sst_open(const char * const dirname, const u64 seq, const u32 way); 47 | 48 | extern const struct sst_meta * 49 | sst_meta(struct sst * const sst); 50 | 51 | extern void 52 | sst_rcache(struct sst * const sst, struct rcache * const rc); 53 | 54 | extern struct kv * 55 | sst_get(struct sst * const map, const struct kref * const key, struct kv * const out); 56 | 57 | extern bool 58 | sst_probe(struct sst* const map, const struct kref * const key); 59 | 60 | extern struct kv * 61 | sst_first_key(struct sst * const map, struct kv * const out); 62 | 63 | extern struct kv * 64 | sst_last_key(struct sst * const map, struct kv * const out); 65 | 66 | extern void 67 | sst_destroy(struct sst * const map); 68 | 69 | extern void 70 | sst_dump(struct sst * const sst, const char * const fn); 71 | 72 | extern void 73 | sst_fprint(struct sst * const map, FILE * const out); 74 | 75 | struct sst_iter; 76 | 77 | extern struct sst_iter * 78 | sst_iter_create(struct sst * const sst); 79 | 80 | extern bool 81 | sst_iter_ts(struct sst_iter * const iter); 82 | 83 | extern void 84 | sst_iter_seek(struct sst_iter * const iter, const struct kref * const key); 85 | 86 | extern void 87 | sst_iter_seek_null(struct sst_iter * const iter); 88 | 89 | extern bool 90 | sst_iter_valid(struct sst_iter * const iter); 91 | 92 | extern struct kv * 93 | sst_iter_peek(struct sst_iter * const iter, struct kv * const out); 94 | 95 | extern bool 96 | sst_iter_kref(struct sst_iter * const iter, struct kref * const kref); 97 | 98 | extern bool 99 | sst_iter_kvref(struct sst_iter * const iter, struct kvref * const kvref); 100 | 101 | extern u64 102 | sst_iter_retain(struct sst_iter * const iter); 103 | 104 | extern void 105 | sst_iter_release(struct sst_iter * const iter, const u64 opaque); 106 | 107 | extern void 108 | sst_iter_skip1(struct sst_iter * const iter); 109 | 110 | extern void 111 | sst_iter_skip(struct sst_iter * const iter, const u32 nr); 112 | 113 | extern struct kv * 114 | sst_iter_next(struct sst_iter * const iter, struct kv * const out); 115 | 116 | extern void 117 | sst_iter_park(struct sst_iter * const iter); 118 | 119 | u64 120 | sst_iter_retain(struct sst_iter * const iter); 121 | 122 | void 123 | sst_iter_release(struct sst_iter * const iter, const u64 opaque); 124 | 125 | extern void 126 | sst_iter_destroy(struct sst_iter * const iter); 127 | // }}} sst 128 | 129 | // build-sst {{{ 130 | // api contains sorted keys and supports iter_next(). 131 | // all keys in the map_api will be added to the sstable. 132 | extern u64 133 | sst_build(const char * const dirname, struct miter * const miter, 134 | const u64 seq, const u32 way, const u32 maxblkid0, const bool del, const bool ckeys, 135 | const struct kv * const k0, const struct kv * const kz); 136 | // }}} build-sst 137 | 138 | // msstx {{{ 139 | // msst (multi-sst) 140 | struct msst; 141 | struct msstx_iter; 142 | 143 | // msstx 144 | extern struct msst * 145 | msstx_open(const char * const dirname, const u64 seq, const u32 nway); 146 | 147 | extern void 148 | msst_rcache(struct msst * const msst, struct rcache * const rc); 149 | 150 | extern void 151 | msstx_destroy(struct msst * const msst); 152 | 153 | extern struct msstx_iter * 154 | msstx_iter_create(struct msst * const msst); 155 | 156 | extern struct kv * 157 | msstx_get(struct msst * const msst, const struct kref * const key, struct kv * const out); 158 | 159 | extern bool 160 | msstx_probe(struct msst * const msst, const struct kref * const key); 161 | 162 | extern bool 163 | msstx_iter_valid(struct msstx_iter * const iter); 164 | 165 | extern void 166 | msstx_iter_seek(struct msstx_iter * const iter, const struct kref * const key); 167 | 168 | extern void 169 | msstx_iter_seek_null(struct msstx_iter * const iter); 170 | 171 | extern struct kv * 172 | msstx_iter_peek(struct msstx_iter * const iter, struct kv * const out); 173 | 174 | extern bool 175 | msstx_iter_kref(struct msstx_iter * const iter, struct kref * const kref); 176 | 177 | extern bool 178 | msstx_iter_kvref(struct msstx_iter * const iter, struct kvref * const kvref); 179 | 180 | extern u64 181 | msstx_iter_retain(struct msstx_iter * const iter); 182 | 183 | extern void 184 | msstx_iter_release(struct msstx_iter * const iter, const u64 opaque); 185 | 186 | extern void 187 | msstx_iter_skip1(struct msstx_iter * const iter); 188 | 189 | extern void 190 | msstx_iter_skip(struct msstx_iter * const iter, const u32 nr); 191 | 192 | extern struct kv * 193 | msstx_iter_next(struct msstx_iter * const iter, struct kv * const out); 194 | 195 | extern void 196 | msstx_iter_park(struct msstx_iter * const iter); 197 | 198 | extern void 199 | msstx_iter_destroy(struct msstx_iter * const iter); 200 | // }}} msstx 201 | 202 | // ssty {{{ 203 | struct ssty; 204 | 205 | extern struct ssty * 206 | ssty_open(const char * const dirname, const u64 seq, const u32 nway); 207 | 208 | extern void 209 | ssty_destroy(struct ssty * const ssty); 210 | 211 | extern void 212 | ssty_fprint(struct ssty * const ssty, FILE * const fout); 213 | // }}} ssty 214 | 215 | // mssty {{{ 216 | struct mssty_ref; 217 | struct mssty_iter; 218 | 219 | extern bool 220 | mssty_open_y(const char * const dirname, struct msst * const msst); 221 | 222 | extern struct msst * 223 | mssty_open(const char * const dirname, const u64 seq, const u32 nway); 224 | 225 | extern void 226 | mssty_destroy(struct msst * const msst); 227 | 228 | extern void 229 | mssty_fprint(struct msst * const msst, FILE * const fout); 230 | 231 | extern struct mssty_ref * 232 | mssty_ref(struct msst * const msst); 233 | 234 | extern struct msst * 235 | mssty_unref(struct mssty_ref * const ref); 236 | 237 | extern struct mssty_iter * 238 | mssty_iter_create(struct mssty_ref * const ref); 239 | 240 | extern bool 241 | mssty_iter_valid(struct mssty_iter * const iter); 242 | 243 | extern void 244 | mssty_iter_seek(struct mssty_iter * const iter, const struct kref * const key); 245 | 246 | extern void 247 | mssty_iter_seek_null(struct mssty_iter * const iter); 248 | 249 | extern void 250 | mssty_iter_seek_near(struct mssty_iter * const iter, const struct kref * const key, const bool bsearch_keys); 251 | 252 | extern struct kv * 253 | mssty_iter_peek(struct mssty_iter * const iter, struct kv * const out); 254 | 255 | extern bool 256 | mssty_iter_kref(struct mssty_iter * const iter, struct kref * const kref); 257 | 258 | extern bool 259 | mssty_iter_kvref(struct mssty_iter * const iter, struct kvref * const kvref); 260 | 261 | extern u64 262 | mssty_iter_retain(struct mssty_iter * const iter); 263 | 264 | extern void 265 | mssty_iter_release(struct mssty_iter * const iter, const u64 opaque); 266 | 267 | extern void 268 | mssty_iter_skip1(struct mssty_iter * const iter); 269 | 270 | extern void 271 | mssty_iter_skip(struct mssty_iter * const iter, const u32 nr); 272 | 273 | extern struct kv * 274 | mssty_iter_next(struct mssty_iter * const iter, struct kv * const out); 275 | 276 | extern void 277 | mssty_iter_park(struct mssty_iter * const iter); 278 | 279 | extern void 280 | mssty_iter_destroy(struct mssty_iter * const iter); 281 | 282 | // ts iter: ignore a key if its newest version is a tombstone 283 | extern bool 284 | mssty_iter_ts(struct mssty_iter * const iter); 285 | 286 | extern void 287 | mssty_iter_seek_ts(struct mssty_iter * const iter, const struct kref * const key); 288 | 289 | extern void 290 | mssty_iter_skip1_ts(struct mssty_iter * const iter); 291 | 292 | extern void 293 | mssty_iter_skip_ts(struct mssty_iter * const iter, const u32 nr); 294 | 295 | extern struct kv * 296 | mssty_iter_next_ts(struct mssty_iter * const iter, struct kv * const out); 297 | 298 | // dup iter: return all versions, including old keys and tombstones 299 | extern struct kv * 300 | mssty_iter_peek_dup(struct mssty_iter * const iter, struct kv * const out); 301 | 302 | extern void 303 | mssty_iter_skip1_dup(struct mssty_iter * const iter); 304 | 305 | extern void 306 | mssty_iter_skip_dup(struct mssty_iter * const iter, const u32 nr); 307 | 308 | extern struct kv * 309 | mssty_iter_next_dup(struct mssty_iter * const iter, struct kv * const out); 310 | 311 | extern bool 312 | mssty_iter_kref_dup(struct mssty_iter * const iter, struct kref * const kref); 313 | 314 | extern bool 315 | mssty_iter_kvref_dup(struct mssty_iter * const iter, struct kvref * const kvref); 316 | 317 | // mssty_get can return tombstone 318 | extern struct kv * 319 | mssty_get(struct mssty_ref * const ref, const struct kref * const key, struct kv * const out); 320 | 321 | // mssty_probe can return tombstone 322 | extern bool 323 | mssty_probe(struct mssty_ref * const ref, const struct kref * const key); 324 | 325 | // return NULL for tomestone 326 | extern struct kv * 327 | mssty_get_ts(struct mssty_ref * const ref, const struct kref * const key, struct kv * const out); 328 | 329 | // return false for tomestone 330 | extern bool 331 | mssty_probe_ts(struct mssty_ref * const ref, const struct kref * const key); 332 | 333 | extern bool 334 | mssty_get_value_ts(struct mssty_ref * const ref, const struct kref * const key, 335 | void * const vbuf_out, u32 * const vlen_out); 336 | 337 | extern struct kv * 338 | mssty_first(struct msst * const msst, struct kv * const out); 339 | 340 | extern struct kv * 341 | mssty_last(struct msst * const msst, struct kv * const out); 342 | 343 | extern void 344 | mssty_dump(struct msst * const msst, const char * const fn); 345 | // }}} mssty 346 | 347 | // build-ssty {{{ 348 | // build extended metadata based on a set of sstables. 349 | // y0 and way0 are optional for speeding up the sorting 350 | extern u32 351 | ssty_build(const char * const dirname, struct msst * const msst, 352 | const u64 seq, const u32 way, struct msst * const y0, const u32 way0, const bool tags); 353 | // }}} build-ssty 354 | 355 | // msstv {{{ 356 | struct msstv; 357 | struct msstv_iter; 358 | struct msstv_ref; 359 | 360 | extern struct msstv * 361 | msstv_create(const u64 nslots, const u64 version); 362 | 363 | extern void 364 | msstv_append(struct msstv * const v, struct msst * const msst, const struct kv * const anchor); 365 | 366 | extern void 367 | msstv_rcache(struct msstv * const v, struct rcache * const rc); 368 | 369 | extern void 370 | msstv_destroy(struct msstv * const v); 371 | 372 | extern struct msstv * 373 | msstv_open(const char * const dirname, const char * const filename); 374 | 375 | extern struct msstv * 376 | msstv_open_version(const char * const dirname, const u64 version); 377 | 378 | extern struct msstv_ref * 379 | msstv_ref(struct msstv * const v); 380 | 381 | extern struct msstv * 382 | msstv_unref(struct msstv_ref * const ref); 383 | 384 | extern struct kv * 385 | msstv_get(struct msstv_ref * const ref, const struct kref * const key, struct kv * const out); 386 | 387 | extern bool 388 | msstv_probe(struct msstv_ref * const ref, const struct kref * const key); 389 | 390 | // return NULL for tomestone 391 | extern struct kv * 392 | msstv_get_ts(struct msstv_ref * const ref, const struct kref * const key, struct kv * const out); 393 | 394 | // return false for tomestone 395 | extern bool 396 | msstv_probe_ts(struct msstv_ref * const ref, const struct kref * const key); 397 | 398 | extern bool 399 | msstv_get_value_ts(struct msstv_ref * const ref, const struct kref * const key, 400 | void * const vbuf_out, u32 * const vlen_out); 401 | 402 | extern struct msstv_iter * 403 | msstv_iter_create(struct msstv_ref * const ref); 404 | 405 | extern bool 406 | msstv_iter_valid(struct msstv_iter * const vi); 407 | 408 | extern void 409 | msstv_iter_seek(struct msstv_iter * const vi, const struct kref * const key); 410 | 411 | extern struct kv * 412 | msstv_iter_peek(struct msstv_iter * const vi, struct kv * const out); 413 | 414 | extern bool 415 | msstv_iter_kref(struct msstv_iter * const vi, struct kref * const kref); 416 | 417 | extern bool 418 | msstv_iter_kvref(struct msstv_iter * const vi, struct kvref * const kvref); 419 | 420 | extern u64 421 | msstv_iter_retain(struct msstv_iter * const vi); 422 | 423 | extern void 424 | msstv_iter_release(struct msstv_iter * const vi, const u64 opaque); 425 | 426 | extern void 427 | msstv_iter_skip1(struct msstv_iter * const vi); 428 | 429 | extern void 430 | msstv_iter_skip(struct msstv_iter * const vi, const u32 nr); 431 | 432 | extern struct kv * 433 | msstv_iter_next(struct msstv_iter * const vi, struct kv * const out); 434 | 435 | extern void 436 | msstv_iter_park(struct msstv_iter * const vi); 437 | 438 | extern bool 439 | msstv_iter_ts(struct msstv_iter * const vi); 440 | 441 | extern void 442 | msstv_iter_seek_ts(struct msstv_iter * const vi, const struct kref * const key); 443 | 444 | extern void 445 | msstv_iter_skip1_ts(struct msstv_iter * const vi); 446 | 447 | extern void 448 | msstv_iter_skip_ts(struct msstv_iter * const vi, const u32 nr); 449 | 450 | extern struct kv * 451 | msstv_iter_next_ts(struct msstv_iter * const vi, struct kv * const out); 452 | 453 | extern void 454 | msstv_fprint(struct msstv * const v, FILE * const out); 455 | 456 | extern void 457 | msstv_iter_destroy(struct msstv_iter * const vi); 458 | 459 | // UNSAFE! 460 | // return the anchors of msstv terminated with NULL 461 | // the returned pointer should be freed after use 462 | // must use when holding a msstv 463 | // anchor->vlen: 0: accepted; 1: rejected 464 | extern struct kv ** 465 | msstv_anchors(struct msstv * const v); 466 | // }}} msstv 467 | 468 | // msstz {{{ 469 | struct msstz; 470 | 471 | extern struct msstz * 472 | msstz_open(const char * const dirname, const u64 cache_size_mb, const bool ckeys, const bool tags); 473 | 474 | extern void 475 | msstz_destroy(struct msstz * const z); 476 | 477 | extern int 478 | msstz_logfd(struct msstz * const z); 479 | 480 | // return number of bytes written since opened 481 | extern u64 482 | msstz_stat_writes(struct msstz * const z); 483 | 484 | extern u64 485 | msstz_stat_reads(struct msstz * const z); 486 | 487 | // default is 0 488 | extern void 489 | msstz_set_minsz(struct msstz * const z, const u64 minsz); 490 | 491 | extern u64 492 | msstz_version(struct msstz * const z); 493 | 494 | extern struct msstv * 495 | msstz_getv(struct msstz * const z); 496 | 497 | extern void 498 | msstz_putv(struct msstz * const z, struct msstv * const v); 499 | 500 | typedef void (*msstz_range_cb)(void * priv, const bool accepted, const struct kv * k0, const struct kv * kz); 501 | 502 | extern void 503 | msstz_comp(struct msstz * const z, const struct kvmap_api * const api1, void * const map1, 504 | const u32 nr_workers, const u32 co_per_worker, const u64 max_reject); 505 | // }}} msstz 506 | 507 | // api {{{ 508 | extern const struct kvmap_api kvmap_api_sst; 509 | extern const struct kvmap_api kvmap_api_msstx; 510 | extern const struct kvmap_api kvmap_api_mssty; 511 | extern const struct kvmap_api kvmap_api_mssty_ts; 512 | extern const struct kvmap_api kvmap_api_msstv; 513 | extern const struct kvmap_api kvmap_api_msstv_ts; 514 | // }}} api 515 | 516 | #ifdef __cplusplus 517 | } 518 | #endif 519 | // vim:fdm=marker 520 | -------------------------------------------------------------------------------- /wh.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | #ifdef __cplusplus 9 | extern "C" { 10 | #endif 11 | 12 | struct wormhole; 13 | struct wormref; 14 | 15 | // wormhole {{{ 16 | // the wh created by wormhole_create() can work with all of safe/unsafe operations. 17 | extern struct wormhole * 18 | wormhole_create(const struct kvmap_mm * const mm); 19 | 20 | // the wh created by whunsafe_create() can only work with the unsafe operations. 21 | extern struct wormhole * 22 | whunsafe_create(const struct kvmap_mm * const mm); 23 | 24 | extern struct kv * 25 | wormhole_get(struct wormref * const ref, const struct kref * const key, struct kv * const out); 26 | 27 | extern bool 28 | wormhole_probe(struct wormref * const ref, const struct kref * const key); 29 | 30 | extern bool 31 | wormhole_put(struct wormref * const ref, struct kv * const kv); 32 | 33 | extern bool 34 | wormhole_merge(struct wormref * const ref, const struct kref * const kref, 35 | kv_merge_func uf, void * const priv); 36 | 37 | extern bool 38 | wormhole_inpr(struct wormref * const ref, const struct kref * const key, 39 | kv_inp_func uf, void * const priv); 40 | 41 | extern bool 42 | wormhole_inpw(struct wormref * const ref, const struct kref * const key, 43 | kv_inp_func uf, void * const priv); 44 | 45 | extern bool 46 | wormhole_del(struct wormref * const ref, const struct kref * const key); 47 | 48 | extern u64 49 | wormhole_delr(struct wormref * const ref, const struct kref * const start, 50 | const struct kref * const end); 51 | 52 | extern struct wormhole_iter * 53 | wormhole_iter_create(struct wormref * const ref); 54 | 55 | extern void 56 | wormhole_iter_seek(struct wormhole_iter * const iter, const struct kref * const key); 57 | 58 | extern bool 59 | wormhole_iter_valid(struct wormhole_iter * const iter); 60 | 61 | extern struct kv * 62 | wormhole_iter_peek(struct wormhole_iter * const iter, struct kv * const out); 63 | 64 | extern bool 65 | wormhole_iter_kref(struct wormhole_iter * const iter, struct kref * const kref); 66 | 67 | extern bool 68 | wormhole_iter_kvref(struct wormhole_iter * const iter, struct kvref * const kvref); 69 | 70 | extern void 71 | wormhole_iter_skip1(struct wormhole_iter * const iter); 72 | 73 | extern void 74 | wormhole_iter_skip(struct wormhole_iter * const iter, const u32 nr); 75 | 76 | extern struct kv * 77 | wormhole_iter_next(struct wormhole_iter * const iter, struct kv * const out); 78 | 79 | extern bool 80 | wormhole_iter_inp(struct wormhole_iter * const iter, kv_inp_func uf, void * const priv); 81 | 82 | extern void 83 | wormhole_iter_park(struct wormhole_iter * const iter); 84 | 85 | extern void 86 | wormhole_iter_destroy(struct wormhole_iter * const iter); 87 | 88 | extern struct wormref * 89 | wormhole_ref(struct wormhole * const map); 90 | 91 | extern struct wormhole * 92 | wormhole_unref(struct wormref * const ref); 93 | 94 | extern void 95 | wormhole_park(struct wormref * const ref); 96 | 97 | extern void 98 | wormhole_resume(struct wormref * const ref); 99 | 100 | extern void 101 | wormhole_refresh_qstate(struct wormref * const ref); 102 | 103 | // clean with more threads 104 | extern void 105 | wormhole_clean_th(struct wormhole * const map, const u32 nr_threads); 106 | 107 | extern void 108 | wormhole_clean(struct wormhole * const map); 109 | 110 | extern void 111 | wormhole_destroy(struct wormhole * const map); 112 | 113 | // safe API (no need to refresh qstate) 114 | 115 | extern struct kv * 116 | whsafe_get(struct wormref * const ref, const struct kref * const key, struct kv * const out); 117 | 118 | extern bool 119 | whsafe_probe(struct wormref * const ref, const struct kref * const key); 120 | 121 | extern bool 122 | whsafe_put(struct wormref * const ref, struct kv * const kv); 123 | 124 | extern bool 125 | whsafe_merge(struct wormref * const ref, const struct kref * const kref, 126 | kv_merge_func uf, void * const priv); 127 | 128 | extern bool 129 | whsafe_inpr(struct wormref * const ref, const struct kref * const key, 130 | kv_inp_func uf, void * const priv); 131 | 132 | extern bool 133 | whsafe_inpw(struct wormref * const ref, const struct kref * const key, 134 | kv_inp_func uf, void * const priv); 135 | 136 | extern bool 137 | whsafe_del(struct wormref * const ref, const struct kref * const key); 138 | 139 | extern u64 140 | whsafe_delr(struct wormref * const ref, const struct kref * const start, 141 | const struct kref * const end); 142 | 143 | // use wormhole_iter_create 144 | extern void 145 | whsafe_iter_seek(struct wormhole_iter * const iter, const struct kref * const key); 146 | 147 | extern struct kv * 148 | whsafe_iter_peek(struct wormhole_iter * const iter, struct kv * const out); 149 | 150 | // use wormhole_iter_valid 151 | // use wormhole_iter_peek 152 | // use wormhole_iter_kref 153 | // use wormhole_iter_kvref 154 | // use wormhole_iter_skip1 155 | // use wormhole_iter_skip 156 | // use wormhole_iter_next 157 | // use wormhole_iter_inp 158 | 159 | extern void 160 | whsafe_iter_park(struct wormhole_iter * const iter); 161 | 162 | extern void 163 | whsafe_iter_destroy(struct wormhole_iter * const iter); 164 | 165 | extern struct wormref * 166 | whsafe_ref(struct wormhole * const map); 167 | 168 | // use wormhole_unref 169 | 170 | // unsafe API 171 | 172 | extern struct kv * 173 | whunsafe_get(struct wormhole * const map, const struct kref * const key, struct kv * const out); 174 | 175 | extern bool 176 | whunsafe_probe(struct wormhole * const map, const struct kref * const key); 177 | 178 | extern bool 179 | whunsafe_put(struct wormhole * const map, struct kv * const kv); 180 | 181 | extern bool 182 | whunsafe_merge(struct wormhole * const map, const struct kref * const kref, 183 | kv_merge_func uf, void * const priv); 184 | 185 | extern bool 186 | whunsafe_inp(struct wormhole * const map, const struct kref * const key, 187 | kv_inp_func uf, void * const priv); 188 | 189 | extern bool 190 | whunsafe_del(struct wormhole * const map, const struct kref * const key); 191 | 192 | extern u64 193 | whunsafe_delr(struct wormhole * const map, const struct kref * const start, 194 | const struct kref * const end); 195 | 196 | extern struct wormhole_iter * 197 | whunsafe_iter_create(struct wormhole * const map); 198 | 199 | extern void 200 | whunsafe_iter_seek(struct wormhole_iter * const iter, const struct kref * const key); 201 | 202 | // unsafe iter_valid: use wormhole_iter_valid 203 | // unsafe iter_peek: use wormhole_iter_peek 204 | // unsafe iter_kref: use wormhole_iter_kref 205 | 206 | extern void 207 | whunsafe_iter_skip1(struct wormhole_iter * const iter); 208 | 209 | extern void 210 | whunsafe_iter_skip(struct wormhole_iter * const iter, const u32 nr); 211 | 212 | extern struct kv * 213 | whunsafe_iter_next(struct wormhole_iter * const iter, struct kv * const out); 214 | 215 | // unsafe iter_inp: use wormhole_iter_inp 216 | 217 | extern void 218 | whunsafe_iter_destroy(struct wormhole_iter * const iter); 219 | 220 | extern void 221 | wormhole_fprint(struct wormhole * const map, FILE * const out); 222 | 223 | extern const struct kvmap_api kvmap_api_wormhole; 224 | extern const struct kvmap_api kvmap_api_whsafe; 225 | extern const struct kvmap_api kvmap_api_whunsafe; 226 | // }}} wormhole 227 | 228 | #ifdef __cplusplus 229 | } 230 | #endif 231 | // vim:fdm=marker 232 | -------------------------------------------------------------------------------- /xdb.h: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #pragma once 7 | 8 | #include "lib.h" 9 | #include "kv.h" 10 | 11 | #ifdef __cplusplus 12 | extern "C" { 13 | #endif 14 | 15 | struct xdb; 16 | struct xdb_ref; 17 | struct xdb_iter; 18 | 19 | // xdb {{{ 20 | extern struct xdb * 21 | xdb_open(const char * const dir, const size_t cache_size_mb, const size_t mt_size_mb, const size_t wal_size_mb, 22 | const bool ckeys, const bool tags, const u32 nr_workers, const u32 co_per_worker, const char * const worker_cores); 23 | 24 | extern void 25 | xdb_close(struct xdb * const xdb); 26 | 27 | // kvmap_api 28 | extern struct xdb_ref * 29 | xdb_ref(struct xdb * const xdb); 30 | 31 | extern struct xdb* 32 | xdb_unref(struct xdb_ref * const ref); 33 | 34 | extern struct kv * 35 | xdb_get(struct xdb_ref * const ref, const struct kref * const kref, struct kv * const out); 36 | 37 | extern bool 38 | xdb_probe(struct xdb_ref * const ref, const struct kref * const kref); 39 | 40 | extern bool 41 | xdb_put(struct xdb_ref * const ref, const struct kv * const kv); 42 | 43 | extern bool 44 | xdb_del(struct xdb_ref * const ref, const struct kref * const kref); 45 | 46 | extern void 47 | xdb_sync(struct xdb_ref * const ref); 48 | 49 | // AKA Atomic Read-Modify-Write 50 | // A merge can fail without doing anything due to allocate failures 51 | // uf() can be invoked multiple times due to abort and retry (these are not errors) 52 | // The last invocation will take the actual effect if it is successful 53 | // The returned kvs will be ignored except for the last one (returned by the last call to uf) 54 | // Memory allocated by uf must be freed by the caller after xdb_merge has returned 55 | // The uf can perform in-place update if kv0 is not NULL (just return kv0 from uf()) 56 | // An in-place update may still cause an memtable insertion if kv0 was not from the memtable (loaded from a partition) 57 | extern bool 58 | xdb_merge(struct xdb_ref * const ref, const struct kref * const kref, kv_merge_func uf, void * const priv); 59 | 60 | // iter 61 | extern struct xdb_iter * 62 | xdb_iter_create(struct xdb_ref * const ref); 63 | 64 | extern void 65 | xdb_iter_park(struct xdb_iter * const iter); 66 | 67 | extern void 68 | xdb_iter_seek(struct xdb_iter * const iter, const struct kref * const key); 69 | 70 | extern bool 71 | xdb_iter_valid(struct xdb_iter * const iter); 72 | 73 | extern struct kv * 74 | xdb_iter_peek(struct xdb_iter * const iter, struct kv * const out); 75 | 76 | extern bool 77 | xdb_iter_kref(struct xdb_iter * const iter, struct kref * const kref); 78 | 79 | extern bool 80 | xdb_iter_kvref(struct xdb_iter * const iter, struct kvref * const kvref); 81 | 82 | extern void 83 | xdb_iter_skip1(struct xdb_iter * const iter); 84 | 85 | extern void 86 | xdb_iter_skip(struct xdb_iter * const iter, u32 n); 87 | 88 | extern struct kv* 89 | xdb_iter_next(struct xdb_iter * const iter, struct kv * const out); 90 | 91 | extern void 92 | xdb_iter_destroy(struct xdb_iter * const iter); 93 | 94 | extern const struct kvmap_api kvmap_api_xdb; 95 | // }}} xdb 96 | 97 | // remixdb {{{ 98 | extern struct xdb * 99 | remixdb_open(const char * const dir, const size_t cache_size_mb, const size_t mt_size_mb, const bool tags); 100 | 101 | extern struct xdb * 102 | remixdb_open_compact(const char * const dir, const size_t cache_size_mb, const size_t mt_size_mb); 103 | 104 | extern struct xdb_ref * 105 | remixdb_ref(struct xdb * const xdb); 106 | 107 | extern void 108 | remixdb_unref(struct xdb_ref * const ref); 109 | 110 | extern void 111 | remixdb_close(struct xdb * const xdb); 112 | 113 | extern bool 114 | remixdb_put(struct xdb_ref * const ref, const void * const kbuf, const u32 klen, 115 | const void * const vbuf, const u32 vlen); 116 | 117 | extern bool 118 | remixdb_del(struct xdb_ref * const ref, const void * const kbuf, const u32 klen); 119 | 120 | extern bool 121 | remixdb_probe(struct xdb_ref * const ref, const void * const kbuf, const u32 klen); 122 | 123 | extern bool 124 | remixdb_get(struct xdb_ref * const ref, const void * const kbuf, const u32 klen, 125 | void * const vbuf_out, u32 * const vlen_out); 126 | 127 | extern void 128 | remixdb_sync(struct xdb_ref * const ref); 129 | 130 | extern struct xdb_iter * 131 | remixdb_iter_create(struct xdb_ref * const ref); 132 | 133 | extern void 134 | remixdb_iter_seek(struct xdb_iter * const iter, const void * const kbuf, const u32 klen); 135 | 136 | extern bool 137 | remixdb_iter_valid(struct xdb_iter * const iter); 138 | 139 | extern bool 140 | remixdb_iter_peek(struct xdb_iter * const iter, 141 | void * const kbuf_out, u32 * const klen_out, 142 | void * const vbuf_out, u32 * const vlen_out); 143 | 144 | extern void 145 | remixdb_iter_skip1(struct xdb_iter * const iter); 146 | 147 | extern void 148 | remixdb_iter_skip(struct xdb_iter * const iter, const u32 nr); 149 | 150 | extern void 151 | remixdb_iter_park(struct xdb_iter * const iter); 152 | 153 | extern void 154 | remixdb_iter_destroy(struct xdb_iter * const iter); 155 | // }}} remixdb 156 | 157 | #ifdef __cplusplus 158 | } 159 | #endif 160 | // vim:fdm=marker 161 | -------------------------------------------------------------------------------- /xdb.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | # 4 | # Copyright (c) 2016--2021 Wu, Xingbo 5 | # 6 | # All rights reserved. No warranty, explicit or implicit, provided. 7 | # 8 | 9 | import msgpack 10 | from ctypes import * # CDLL and c_xxx types 11 | 12 | # libxdb {{{ 13 | # Change this path when necessary 14 | libxdb = CDLL("./libremixdb.so") 15 | 16 | # open 17 | # dir, cachesz, mtsz, tags -> xdbptr 18 | libxdb.remixdb_open.argtypes = [c_char_p, c_uint, c_uint, c_bool] 19 | libxdb.remixdb_open.restype = c_void_p 20 | 21 | # close (no return value) 22 | libxdb.remixdb_close.argtypes = [c_void_p] 23 | 24 | # ref 25 | libxdb.remixdb_ref.argtypes = [c_void_p] 26 | libxdb.remixdb_ref.restype = c_void_p 27 | 28 | # unref 29 | libxdb.remixdb_unref.argtypes = [c_void_p] 30 | 31 | # put 32 | # xdbptr, keyptr, keylen, vptr, vlen -> bool 33 | libxdb.remixdb_put.argtypes = [c_void_p, c_char_p, c_uint, c_char_p, c_uint] 34 | libxdb.remixdb_put.restype = c_bool 35 | 36 | # get 37 | # xdbptr, keyptr, keylen, vptr_out, vlen_out -> bool 38 | libxdb.remixdb_get.argtypes = [c_void_p, c_char_p, c_uint, c_char_p, c_void_p] 39 | libxdb.remixdb_get.restype = c_bool 40 | 41 | # probe 42 | libxdb.remixdb_probe.argtypes = [c_void_p, c_char_p, c_uint] 43 | libxdb.remixdb_probe.restype = c_bool 44 | 45 | # del 46 | libxdb.remixdb_del.argtypes = [c_void_p, c_char_p, c_uint] 47 | libxdb.remixdb_del.restype = c_bool 48 | 49 | # sync 50 | libxdb.remixdb_sync.argtypes = [c_void_p] 51 | 52 | # iter_create 53 | libxdb.remixdb_iter_create.argtypes = [c_void_p] 54 | libxdb.remixdb_iter_create.restype = c_void_p 55 | 56 | # iter_seek 57 | libxdb.remixdb_iter_seek.argtypes = [c_void_p, c_char_p, c_uint] 58 | 59 | # iter_valid 60 | libxdb.remixdb_iter_valid.argtypes = [c_void_p] 61 | libxdb.remixdb_iter_valid.restype = c_bool 62 | 63 | # iter_skip1 64 | libxdb.remixdb_iter_skip1.argtypes = [c_void_p] 65 | 66 | # iter_skip 67 | libxdb.remixdb_iter_skip.argtypes = [c_void_p, c_uint] 68 | 69 | # iter_peek 70 | libxdb.remixdb_iter_peek.argtypes = [c_void_p, c_char_p, c_void_p, c_char_p, c_void_p] 71 | libxdb.remixdb_iter_peek.restype = c_bool 72 | 73 | # iter_destroy 74 | libxdb.remixdb_iter_destroy.argtypes = [c_void_p] 75 | # }}} libxdb 76 | 77 | # class {{{ 78 | class Xdb: 79 | def __init__(self, dirname, cachesz=256, mtsz=256, tags=True): 80 | self.xdbptr = libxdb.remixdb_open(dirname.encode('ascii'), c_uint(cachesz), c_uint(mtsz), c_bool(tags)) 81 | 82 | # user must call explicitly 83 | def close(self): 84 | libxdb.remixdb_close(self.xdbptr) 85 | 86 | def ref(self): 87 | return XdbRef(self.xdbptr) 88 | 89 | class XdbRef: 90 | # use xdb.ref() 91 | def __init__(self, xdbptr): 92 | self.refptr = libxdb.remixdb_ref(xdbptr) 93 | self.vbuf = create_string_buffer(65500) 94 | 95 | # user must call explicitly 96 | def unref(self): 97 | libxdb.remixdb_unref(self.refptr) 98 | 99 | def iter(self): 100 | return XdbIter(self.refptr) 101 | 102 | # key: python string; value: any (hierarchical) python object 103 | def put(self, key, value): 104 | binkey = key.encode() 105 | binvalue = msgpack.packb(value) 106 | print(key, msgpack.unpackb(binvalue), len(binvalue)) 107 | 108 | return libxdb.remixdb_put(self.refptr, binkey, c_uint(len(binkey)), binvalue, c_uint(len(binvalue))) 109 | 110 | 111 | # return the value as a python object 112 | def get(self, key): 113 | binkey = key.encode() 114 | vlen = c_uint() 115 | ret = libxdb.remixdb_get(self.refptr, binkey, len(binkey), self.vbuf, byref(vlen)) 116 | if ret: 117 | #vbuf[vlen.value] = b'\x00' 118 | return msgpack.unpackb(self.vbuf.value) 119 | else: 120 | return None 121 | 122 | def delete(self, key): 123 | binkey = key.encode() 124 | return libxdb.remixdb_del(self.refptr, binkey, c_uint(len(binkey))) 125 | 126 | def probe(self, key): 127 | binkey = key.encode() 128 | return libxdb.remixdb_probe(self.refptr, binkey, c_uint(len(binkey))) 129 | 130 | def sync(self): 131 | return libxdb.remixdb_sync(self.refptr) 132 | 133 | class XdbIter: 134 | def __init__(self, refptr): 135 | self.iptr = libxdb.remixdb_iter_create(refptr) 136 | self.kbuf = create_string_buffer(65500) 137 | self.vbuf = create_string_buffer(65500) 138 | 139 | # user must call explicitly 140 | def destroy(self): 141 | libxdb.remixdb_iter_destroy(self.iptr) 142 | 143 | def seek(self, key): 144 | if key is None: 145 | libxdb.remixdb_iter_seek(self.iptr, None, c_uint(0)) 146 | else: 147 | binkey = key.encode() 148 | libxdb.remixdb_iter_seek(self.iptr, binkey, c_uint(len(binkey))) 149 | 150 | def valid(self): 151 | return libxdb.remixdb_iter_valid(self.iptr) 152 | 153 | def skip1(self): 154 | libxdb.remixdb_iter_skip1(self.iptr) 155 | 156 | def skip(self, nr): 157 | libxdb.remixdb_iter_skip(self.iptr, c_uint(nr)) 158 | 159 | # return (key, value) pair or None 160 | def peek(self): 161 | klen = c_uint() 162 | vlen = c_uint() 163 | if libxdb.remixdb_iter_peek(self.iptr, self.kbuf, byref(klen), self.vbuf, byref(vlen)): 164 | self.kbuf[klen.value] = b'\x00' 165 | #vbuf[vlen.value] = b'\x00' 166 | return (self.kbuf.value.decode(), klen.value, msgpack.unpackb(self.vbuf.value), vlen.value) 167 | else: 168 | return None 169 | 170 | # }}} class 171 | 172 | # examples 173 | xdb1 = Xdb("/tmp/pyxdb") # change this path when necessary 174 | ref1 = xdb1.ref() # take a ref for kv operations 175 | 176 | ref1.put("Hello", "pyxdb") 177 | ref1.put("key1", "value1") 178 | ref1.put("key2", "value2") 179 | ref1.put("key3", {"xxx":"valuex", "yyy":"valuey"}) 180 | ref1.delete("key2") 181 | 182 | rget = ref1.get("Hello") 183 | print(rget) 184 | 185 | # don't use ref when iterating 186 | iter1 = ref1.iter() 187 | iter1.seek(None) 188 | while iter1.valid(): 189 | r = iter1.peek() 190 | print(r) 191 | iter1.skip1() 192 | 193 | iter1.destroy() # must destroy all iters before unref 194 | 195 | ref1.sync() 196 | ref1.unref() # must unref all refs before close() 197 | xdb1.close() 198 | 199 | # vim:fdm=marker 200 | -------------------------------------------------------------------------------- /xdbcow.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | # This script performs a naive COW copy of an xdb directory 4 | # orig and dest must be in the same file system that supports hard links 5 | # The table files will be duplicated using hard links without copying 6 | # the WAL files are really copied because they are mutable 7 | 8 | if [[ $# -lt 2 ]]; then 9 | echo "usage: " 10 | exit 0 11 | fi 12 | 13 | orig=${1} dest=${2} 14 | 15 | if [[ ! -d ${orig} || ! -h ${orig}/HEAD ]]; then 16 | echo "${orig}/HEAD is not a symbolic link" 17 | exit 0 18 | fi 19 | 20 | if [[ -d ${dest} ]]; then 21 | echo "${dest} already exists; must use a non-existing path" 22 | exit 0 23 | fi 24 | 25 | mkdir -p ${dest} 26 | if [[ ! -d ${dest} ]]; then 27 | echo "creating ${dest} failed" 28 | exit 0 29 | fi 30 | 31 | # hardlinks for immutable files 32 | cp -l ${orig}/*.sstx ${orig}/*.ssty ${orig}/*.ver ${dest}/ 33 | 34 | # duplicate softlinks HEAD and HEAD1 (pointing to a *.ver) 35 | cp -a ${orig}/HEAD ${dest}/HEAD 36 | cp -a ${orig}/HEAD1 ${dest}/HEAD1 37 | 38 | # really copy wals 39 | cp ${orig}/wal1 ${orig}/wal2 ${dest}/ 40 | -------------------------------------------------------------------------------- /xdbdemo.c: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #define _GNU_SOURCE 7 | 8 | #include 9 | 10 | #include "lib.h" 11 | #include "kv.h" 12 | #include "xdb.h" 13 | 14 | int 15 | main(int argc, char ** argv) 16 | { 17 | (void)argc; 18 | (void)argv; 19 | // Use a small config for demo 20 | // In a moderate setup the recommended numbers are 4096 for both 21 | struct xdb * const xdb = remixdb_open("./xdbdemo", 256, 256, true); // blockcache=256MB, MemTable=256MB, use_tags=true 22 | 23 | // A ref is required to perform the following DB operations. 24 | // A thread should maintain a ref and keep using it. 25 | // Different threads should use different refs. 26 | struct xdb_ref * const ref = remixdb_ref(xdb); 27 | 28 | bool r; 29 | 30 | r = remixdb_put(ref, "remix", 5, "easy", 4); 31 | printf("remixdb_put remix easy %c\n", r?'T':'F'); 32 | 33 | r = remixdb_put(ref, "time_travel", 11, "impossible", 10); 34 | printf("remixdb_put time_travel impossible %c\n", r?'T':'F'); 35 | 36 | r = remixdb_del(ref, "time_travel", 11); 37 | printf("remixdb_del time_travel %c\n", r?'T':'F'); 38 | 39 | r = remixdb_probe(ref, "time_travel", 11); 40 | printf("remixdb_probe time_travel %c\n", r?'T':'F'); 41 | 42 | u32 klen_out = 0; 43 | char kbuf_out[8] = {}; 44 | u32 vlen_out = 0; 45 | char vbuf_out[8] = {}; 46 | r = remixdb_get(ref, "remix", 5, vbuf_out, &vlen_out); 47 | printf("remixdb_get remix %c %u %.*s\n", r?'T':'F', vlen_out, vlen_out, vbuf_out); 48 | 49 | // prepare a few keys for range ops 50 | remixdb_put(ref, "00", 2, "0_value", 7); 51 | remixdb_put(ref, "11", 2, "1_value", 7); 52 | remixdb_put(ref, "22", 2, "2_value", 7); 53 | 54 | // Make all the data persistent in the log. 55 | // Performing sync is expensive. 56 | remixdb_sync(ref); 57 | 58 | // range operations 59 | struct xdb_iter * const iter = remixdb_iter_create(ref); 60 | 61 | printf("remixdb_iter_seek \"\" (zero-length string)\n"); 62 | remixdb_iter_seek(iter, NULL, 0); // seek to the first key 63 | // You can actually insert an zero-size key to the store. (0 <= klen, klen+vlen <= 65500) 64 | 65 | while (remixdb_iter_valid(iter)) { // check whether the iter points to a valid KV pair 66 | r = remixdb_iter_peek(iter, kbuf_out, &klen_out, vbuf_out, &vlen_out); 67 | if (r) { 68 | printf("remixdb_iter_peek klen=%u key=%.*s vlen=%u value=%.*s\n", 69 | klen_out, klen_out, kbuf_out, vlen_out, vlen_out, vbuf_out); 70 | } else { 71 | printf("ERROR!\n"); 72 | } 73 | remixdb_iter_skip1(iter); 74 | } 75 | 76 | // This is OPTIONAL! 77 | // an iter can hold some (reader) locks. 78 | // Other (writer) threads can be blocked by active iters. 79 | // call iter_park to release those resources when you need to go idle 80 | // don't need to call iter_park if you're actively using the iter 81 | remixdb_iter_park(iter); 82 | usleep(10); 83 | 84 | // after calling iter_park, you must perform a seek() to proceed with other operations. 85 | printf("remixdb_iter_seek \"0\" (key_length=1)\n"); 86 | remixdb_iter_seek(iter, "0", 1); 87 | // this time we don't want to copy the value 88 | r = remixdb_iter_peek(iter, kbuf_out, &klen_out, NULL, NULL); 89 | if (r){ 90 | printf("remixdb_iter_peek klen=%u key=%.*s\n", klen_out, klen_out, kbuf_out); 91 | } else { 92 | printf("ERROR: iter_peek failed\n"); 93 | } 94 | 95 | remixdb_iter_destroy(iter); 96 | // there must be no active iters when we call unref() 97 | remixdb_unref(ref); 98 | 99 | // close is not thread-safe 100 | // other threads must have released their references when you call close() 101 | remixdb_close(xdb); 102 | return 0; 103 | } 104 | -------------------------------------------------------------------------------- /xdbexit.c: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #define _GNU_SOURCE 7 | 8 | #include "lib.h" 9 | #include "kv.h" 10 | #include "sst.h" 11 | #include "xdb.h" 12 | 13 | int 14 | main(int argc, char** argv) 15 | { 16 | if (argc < 4) { 17 | printf("Usage: \n"); 18 | return 0; 19 | } 20 | 21 | struct xdb * const xdb = remixdb_open(argv[1], a2u64(argv[2]), a2u64(argv[3]), true); 22 | if (!xdb) { 23 | fprintf(stderr, "xdb_open failed\n"); 24 | return 0; 25 | } 26 | struct xdb_ref * const ref = remixdb_ref(xdb); 27 | 28 | struct xdb_iter * const iter = remixdb_iter_create(ref); 29 | u64 kid = 0; 30 | remixdb_iter_seek(iter, "", 0); 31 | remixdb_iter_skip(iter, 1000); 32 | u8 key[20]; 33 | u8 keycmp[20]; 34 | u32 klen = 0; 35 | while (remixdb_iter_valid(iter)) { 36 | kid += 1000; 37 | remixdb_iter_peek(iter, key, &klen, NULL, NULL); 38 | strdec_64(keycmp, kid); 39 | if (memcmp(key, keycmp, 20)) { 40 | printf("key mismatch at %lu; delete %s and restart the loop\n", kid, argv[1]); 41 | exit(0); 42 | } 43 | remixdb_iter_skip(iter, 1000); 44 | } 45 | 46 | u64 count = kid; 47 | remixdb_iter_seek(iter, "", 0); 48 | remixdb_iter_skip(iter, kid); 49 | while (remixdb_iter_valid(iter)) { 50 | remixdb_iter_peek(iter, key, &klen, NULL, NULL); 51 | remixdb_iter_skip1(iter); 52 | strdec_64(keycmp, count); 53 | count++; 54 | if (memcmp(key, keycmp, 20)) { 55 | printf("key mismatch at %lu; delete %s and restart loop again\n", count, argv[1]); 56 | exit(0); 57 | } 58 | } 59 | printf("found %lu keys, last %.20s OK\n", count, key); 60 | remixdb_iter_destroy(iter); 61 | 62 | u8 value[1024]; 63 | memset(value, 0x11, 1024); 64 | #define NEW ((100000)) 65 | for (u64 i = 0; i < NEW; i++) { 66 | strdec_64(key, count + i); 67 | remixdb_put(ref, key, 20, value, 1024); 68 | } 69 | printf("insert [%lu, %lu]; now exit()\n", count, count + NEW - 1); 70 | remixdb_sync(ref); 71 | exit(0); 72 | return 0; 73 | } 74 | -------------------------------------------------------------------------------- /xdbtest.c: -------------------------------------------------------------------------------- 1 | /* 2 | * Copyright (c) 2016--2021 Wu, Xingbo 3 | * 4 | * All rights reserved. No warranty, explicit or implicit, provided. 5 | */ 6 | #define _GNU_SOURCE 7 | 8 | #include "ctypes.h" 9 | #include "lib.h" 10 | #include "kv.h" 11 | #include "sst.h" 12 | #include "xdb.h" 13 | 14 | struct xdb * xdb; 15 | static u64 nkeys = 0; 16 | static u64 nupdate = 0; 17 | static u64 min_stale = 0; 18 | static u8 * magics = NULL; 19 | u32 nths_update = 0; 20 | u32 nths_getscan = 0; 21 | 22 | au64 all_seq; 23 | au64 all_stale; 24 | au64 all_found; 25 | 26 | static void * 27 | update_worker(void * const ptr) 28 | { 29 | (void)ptr; 30 | srandom_u64(time_nsec()); 31 | const u64 seq = atomic_fetch_add(&all_seq, 1); 32 | const u64 range = nkeys / nths_update; 33 | const u64 mask = range - 1; 34 | const u64 base = seq * range; 35 | struct xdb_ref * const ref = remixdb_ref(xdb); 36 | u8 ktmp[16]; 37 | u8 * const vtmp = calloc(1, 1lu << 16); 38 | memset(vtmp, (int)random_u64(), 1lu << 16); 39 | 40 | //printf("random update [%lu, %lu]\n", base, base+mask); 41 | // set/del 42 | for (u64 i = 0; i < nupdate; i++) { 43 | const u64 r = random_u64(); 44 | const u64 k = base + ((r >> 8) & mask); 45 | const u8 v = r & 0xff; 46 | vtmp[0] = v; 47 | magics[k] = v; 48 | strhex_64(ktmp, k); 49 | 50 | if (v == 0) { // delete 51 | remixdb_del(ref, ktmp, 16); 52 | } else { // update 53 | const u32 vlen = ((i & 0x3fffu) != 0x1357u) ? (((u32)r & 0xf0) + 100) : ((((u32)r & 0xf0) << 6) + 4200); 54 | remixdb_put(ref, ktmp, 16, vtmp, vlen); 55 | } 56 | } 57 | 58 | remixdb_unref(ref); 59 | free(vtmp); 60 | return NULL; 61 | } 62 | 63 | static void * 64 | getscan_worker(void * const ptr) 65 | { 66 | (void)ptr; 67 | const u64 seq = atomic_fetch_add(&all_seq, 1); 68 | const u64 unit = nkeys / nths_getscan + 1; 69 | const u64 min = unit * seq; 70 | const u64 max0 = min + unit; 71 | const u64 max = nkeys < max0 ? nkeys : max0; 72 | 73 | struct xdb_ref * const ref = remixdb_ref(xdb); 74 | 75 | u8 ktmp[16]; 76 | u8 * const out = calloc(1, 1lu << 16); 77 | u32 vlen_out = 0; 78 | 79 | // get seq 80 | u64 stale = 0; 81 | for (u64 i = min; i < max; i++) { 82 | strhex_64(ktmp, i); 83 | const bool r = remixdb_get(ref, ktmp, 16, out, &vlen_out); 84 | if ((r ? out[0] : 0) != magics[i]) 85 | stale++; 86 | } 87 | 88 | u32 klen_out; 89 | u8 kend[16]; 90 | strhex_64(ktmp, min); 91 | strhex_64(kend, max); 92 | struct xdb_iter * const iter = remixdb_iter_create(ref); 93 | remixdb_iter_seek(iter, ktmp, 16); 94 | memset(ktmp, 0, 16); 95 | 96 | // scan 97 | u64 found = 0; 98 | while (remixdb_iter_valid(iter)) { 99 | remixdb_iter_peek(iter, ktmp, &klen_out, NULL, NULL); 100 | debug_assert(klen_out == 16); 101 | if (memcmp(ktmp, kend, 16) < 0) { 102 | found++; 103 | remixdb_iter_skip1(iter); 104 | } else { 105 | break; 106 | } 107 | } 108 | 109 | //printf("get [%lu, %lu] stale %lu found %lu\n", min, max-1, stale, found); 110 | atomic_fetch_add(&all_stale, stale); 111 | atomic_fetch_add(&all_found, found); 112 | 113 | remixdb_iter_destroy(iter); 114 | remixdb_unref(ref); 115 | free(out); 116 | return NULL; 117 | } 118 | 119 | int 120 | main(int argc, char** argv) 121 | { 122 | if (argc < 6) { 123 | printf("Usage: []\n"); 124 | printf(" WAL size = *2\n"); 125 | return 0; 126 | } 127 | 128 | const u32 ncores = process_affinity_count(); 129 | if (ncores < 5) { 130 | fprintf(stderr, "Need at least five cores on the cpu affinity list\n"); 131 | exit(0); 132 | } 133 | 134 | const u64 cachesz = a2u64(argv[2]); 135 | const u64 mtsz = a2u64(argv[3]); 136 | const u64 dpower = a2u64(argv[4]); 137 | const u64 upower = a2u64(argv[5]); 138 | 139 | xdb = remixdb_open(argv[1], cachesz, mtsz, true); 140 | if (!xdb) { 141 | fprintf(stderr, "xdb_open failed\n"); 142 | return 0; 143 | } 144 | 145 | nkeys = 1lu << dpower; 146 | if (nkeys < 1024) 147 | nkeys = 1024; 148 | nupdate = 1lu << upower; 149 | 150 | min_stale = nkeys; 151 | magics = calloc(nkeys, 1); 152 | nths_getscan = ncores - 4; 153 | nths_update = ncores - 4; 154 | 155 | while (__builtin_popcount(nths_update) > 1) 156 | nths_update--; 157 | printf("write threads %u check threads %u\n", nths_update, nths_getscan); 158 | 159 | debug_assert(magics); 160 | const u32 ne = (argc < 7) ? 1000000 : a2u32(argv[6]); 161 | 162 | for (u32 e = 0; e < ne; e++) { 163 | all_seq = 0; 164 | const u64 dt = thread_fork_join(nths_update, update_worker, false, NULL); 165 | if ((e & 0x3u) == 0x3u) { // close/open every 4 epochs 166 | remixdb_close(xdb); 167 | // turn on/off ckeys alternatively, very stressful. 168 | xdb = (e & 4) ? remixdb_open(argv[1], cachesz, mtsz, (e & 8) != 0) : remixdb_open_compact(argv[1], cachesz, mtsz); 169 | if (xdb) { 170 | printf("reopen remixdb ok\n"); 171 | } else { 172 | printf("reopen failed\n"); 173 | exit(0); 174 | } 175 | } 176 | all_stale = 0; 177 | all_found = 0; 178 | all_seq = 0; 179 | (void)thread_fork_join(nths_getscan, getscan_worker, false, NULL); 180 | 181 | char ts[64]; 182 | time_stamp(ts, sizeof(ts)); 183 | const u64 nr = nupdate * nths_update; 184 | printf("[%4u] %s put/del nr %lu mops %.3lf keyrange %lu keycount %lu stale %lu\n", 185 | e, ts, nr, (double)nr / (double)dt * 1e3, nkeys, all_found, all_stale); 186 | 187 | if (all_stale > min_stale) 188 | debug_die(); 189 | min_stale = all_stale; 190 | } 191 | free(magics); 192 | remixdb_close(xdb); 193 | return 0; 194 | } 195 | --------------------------------------------------------------------------------