+170 -61

Cargo.lock

··· 167 167 checksum = "46c5e41b57b8bba42a04676d81cb89e9ee8e859a1a66f80a5a72e1cb76b34d43" 168 168 169 169 [[package]] 170 170 + name = "byteorder-lite" 171 171 + version = "0.1.0" 172 172 + source = "registry+https://github.com/rust-lang/crates.io-index" 173 173 + checksum = "8f1fe948ff07f4bd06c30984e69f5b4899c516a3ef74f34df92a2df2ab535495" 174 174 + 175 175 + [[package]] 170 176 name = "bytes" 171 177 version = "1.10.1" 172 178 source = "registry+https://github.com/rust-lang/crates.io-index" 173 179 checksum = "d71b6127be86fdcfddb610f7182ac57211d4b18a3e9c82eb2d17662f2227ad6a" 180 180 + 181 181 + [[package]] 182 182 + name = "byteview" 183 183 + version = "0.10.0" 184 184 + source = "registry+https://github.com/rust-lang/crates.io-index" 185 185 + checksum = "dda4398f387cc6395a3e93b3867cd9abda914c97a0b344d1eefb2e5c51785fca" 174 186 175 187 [[package]] 176 188 name = "cast" ··· 281 293 checksum = "b05b61dc5112cbb17e4b6cd61790d9845d13888356391624cbe7e41efeac1e75" 282 294 283 295 [[package]] 296 296 + name = "compare" 297 297 + version = "0.0.6" 298 298 + source = "registry+https://github.com/rust-lang/crates.io-index" 299 299 + checksum = "ea0095f6103c2a8b44acd6fd15960c801dafebf02e21940360833e0673f48ba7" 300 300 + 301 301 + [[package]] 284 302 name = "const-str" 285 303 version = "0.4.3" 286 304 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 358 376 ] 359 377 360 378 [[package]] 379 379 + name = "crossbeam-skiplist" 380 380 + version = "0.1.3" 381 381 + source = "registry+https://github.com/rust-lang/crates.io-index" 382 382 + checksum = "df29de440c58ca2cc6e587ec3d22347551a32435fbde9d2bff64e78a9ffa151b" 383 383 + dependencies = [ 384 384 + "crossbeam-epoch", 385 385 + "crossbeam-utils", 386 386 + ] 387 387 + 388 388 + [[package]] 361 389 name = "crossbeam-utils" 362 390 version = "0.8.21" 363 391 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 380 408 ] 381 409 382 410 [[package]] 411 411 + name = "dashmap" 412 412 + version = "6.1.0" 413 413 + source = "registry+https://github.com/rust-lang/crates.io-index" 414 414 + checksum = "5041cc499144891f3790297212f32a74fb938e5136a14943f338ef9e0ae276cf" 415 415 + dependencies = [ 416 416 + "cfg-if", 417 417 + "crossbeam-utils", 418 418 + "hashbrown 0.14.5", 419 419 + "lock_api", 420 420 + "once_cell", 421 421 + "parking_lot_core", 422 422 + ] 423 423 + 424 424 + [[package]] 383 425 name = "data-encoding" 384 426 version = "2.9.0" 385 427 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 422 464 checksum = "48c757948c5ede0e46177b7add2e67155f70e33c07fea8284df6576da70b3719" 423 465 424 466 [[package]] 467 467 + name = "enum_dispatch" 468 468 + version = "0.3.13" 469 469 + source = "registry+https://github.com/rust-lang/crates.io-index" 470 470 + checksum = "aa18ce2bc66555b3218614519ac839ddb759a7d6720732f979ef8d13be147ecd" 471 471 + dependencies = [ 472 472 + "once_cell", 473 473 + "proc-macro2", 474 474 + "quote", 475 475 + "syn 2.0.106", 476 476 + ] 477 477 + 478 478 + [[package]] 425 479 name = "env_filter" 426 480 version = "0.1.3" 427 481 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 445 499 ] 446 500 447 501 [[package]] 502 502 + name = "equivalent" 503 503 + version = "1.0.2" 504 504 + source = "registry+https://github.com/rust-lang/crates.io-index" 505 505 + checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f" 506 506 + 507 507 + [[package]] 448 508 name = "errno" 449 509 version = "0.3.14" 450 510 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 455 515 ] 456 516 457 517 [[package]] 458 458 - name = "fallible-iterator" 459 459 - version = "0.3.0" 518 518 + name = "fastrand" 519 519 + version = "2.3.0" 460 520 source = "registry+https://github.com/rust-lang/crates.io-index" 461 461 - checksum = "2acce4a10f12dc2fb14a218589d4f1f62ef011b2d0cc4b3cb1bba8e94da14649" 521 521 + checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be" 462 522 463 523 [[package]] 464 464 - name = "fallible-streaming-iterator" 465 465 - version = "0.1.9" 524 524 + name = "fjall" 525 525 + version = "3.0.1" 466 526 source = "registry+https://github.com/rust-lang/crates.io-index" 467 467 - checksum = "7360491ce676a36bf9bb3c56c1aa791658183a54d2744120f27285738d90465a" 527 527 + checksum = "4f69637c02d38ad1b0f003101d0195a60368130aa17d9ef78b1557d265a22093" 528 528 + dependencies = [ 529 529 + "byteorder-lite", 530 530 + "byteview", 531 531 + "dashmap", 532 532 + "flume", 533 533 + "log", 534 534 + "lsm-tree", 535 535 + "tempfile", 536 536 + "xxhash-rust", 537 537 + ] 468 538 469 539 [[package]] 470 470 - name = "fastrand" 471 471 - version = "2.3.0" 540 540 + name = "flume" 541 541 + version = "0.12.0" 472 542 source = "registry+https://github.com/rust-lang/crates.io-index" 473 473 - checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be" 474 474 - 475 475 - [[package]] 476 476 - name = "foldhash" 477 477 - version = "0.1.5" 478 478 - source = "registry+https://github.com/rust-lang/crates.io-index" 479 479 - checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2" 543 543 + checksum = "5e139bc46ca777eb5efaf62df0ab8cc5fd400866427e56c68b22e414e53bd3be" 544 544 + dependencies = [ 545 545 + "spin", 546 546 + ] 480 547 481 548 [[package]] 482 549 name = "futures" ··· 608 675 609 676 [[package]] 610 677 name = "hashbrown" 611 611 - version = "0.15.5" 678 678 + version = "0.14.5" 612 679 source = "registry+https://github.com/rust-lang/crates.io-index" 613 613 - checksum = "9229cfe53dfd69f0609a49f65461bd93001ea1ef889cd5529dd176593f5338a1" 614 614 - dependencies = [ 615 615 - "foldhash", 616 616 - ] 680 680 + checksum = "e5274423e17b7c9fc20b6e7e208532f9b19825d82dfd615708b70edd83df41f1" 617 681 618 682 [[package]] 619 619 - name = "hashlink" 620 620 - version = "0.10.0" 683 683 + name = "hashbrown" 684 684 + version = "0.16.1" 621 685 source = "registry+https://github.com/rust-lang/crates.io-index" 622 622 - checksum = "7382cf6263419f2d8df38c55d7da83da5c18aef87fc7a7fc1fb1e344edfe14c1" 623 623 - dependencies = [ 624 624 - "hashbrown", 625 625 - ] 686 686 + checksum = "841d1cc9bed7f9236f321df977030373f4a4163ae1a7dbfe1a51a2c1a51d9100" 626 687 627 688 [[package]] 628 689 name = "heck" ··· 631 692 checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea" 632 693 633 694 [[package]] 695 695 + name = "interval-heap" 696 696 + version = "0.0.5" 697 697 + source = "registry+https://github.com/rust-lang/crates.io-index" 698 698 + checksum = "11274e5e8e89b8607cfedc2910b6626e998779b48a019151c7604d0adcb86ac6" 699 699 + dependencies = [ 700 700 + "compare", 701 701 + ] 702 702 + 703 703 + [[package]] 634 704 name = "io-uring" 635 705 version = "0.7.10" 636 706 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 730 800 checksum = "58f929b4d672ea937a23a1ab494143d968337a5f47e56d0815df1e0890ddf174" 731 801 732 802 [[package]] 733 733 - name = "libsqlite3-sys" 734 734 - version = "0.35.0" 735 735 - source = "registry+https://github.com/rust-lang/crates.io-index" 736 736 - checksum = "133c182a6a2c87864fe97778797e46c7e999672690dc9fa3ee8e241aa4a9c13f" 737 737 - dependencies = [ 738 738 - "pkg-config", 739 739 - "vcpkg", 740 740 - ] 741 741 - 742 742 - [[package]] 743 803 name = "linux-raw-sys" 744 804 version = "0.11.0" 745 805 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 761 821 checksum = "34080505efa8e45a4b816c349525ebe327ceaa8559756f0356cba97ef3bf7432" 762 822 763 823 [[package]] 824 824 + name = "lsm-tree" 825 825 + version = "3.0.1" 826 826 + source = "registry+https://github.com/rust-lang/crates.io-index" 827 827 + checksum = "b875f1dfe14f557f805b167fb9b0fc54c5560c7a4bd6ae02535b2846f276a8cb" 828 828 + dependencies = [ 829 829 + "byteorder-lite", 830 830 + "byteview", 831 831 + "crossbeam-skiplist", 832 832 + "enum_dispatch", 833 833 + "interval-heap", 834 834 + "log", 835 835 + "quick_cache", 836 836 + "rustc-hash", 837 837 + "self_cell", 838 838 + "sfa", 839 839 + "tempfile", 840 840 + "varint-rs", 841 841 + "xxhash-rust", 842 842 + ] 843 843 + 844 844 + [[package]] 764 845 name = "match-lookup" 765 846 version = "0.1.1" 766 847 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 892 973 checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184" 893 974 894 975 [[package]] 895 895 - name = "pkg-config" 896 896 - version = "0.3.32" 897 897 - source = "registry+https://github.com/rust-lang/crates.io-index" 898 898 - checksum = "7edddbd0b52d732b21ad9a5fab5c704c14cd949e5e9a1ec5929a24fded1b904c" 899 899 - 900 900 - [[package]] 901 976 name = "plotters" 902 977 version = "0.3.7" 903 978 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 950 1025 ] 951 1026 952 1027 [[package]] 1028 1028 + name = "quick_cache" 1029 1029 + version = "0.6.18" 1030 1030 + source = "registry+https://github.com/rust-lang/crates.io-index" 1031 1031 + checksum = "7ada44a88ef953a3294f6eb55d2007ba44646015e18613d2f213016379203ef3" 1032 1032 + dependencies = [ 1033 1033 + "equivalent", 1034 1034 + "hashbrown 0.16.1", 1035 1035 + ] 1036 1036 + 1037 1037 + [[package]] 953 1038 name = "quote" 954 1039 version = "1.0.41" 955 1040 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 1024 1109 1025 1110 [[package]] 1026 1111 name = "repo-stream" 1027 1027 - version = "0.1.1" 1112 1112 + version = "0.2.2" 1028 1113 dependencies = [ 1029 1114 "bincode", 1030 1115 "clap", 1031 1116 "criterion", 1032 1117 "env_logger", 1118 1118 + "fjall", 1033 1119 "futures", 1034 1120 "futures-core", 1035 1121 "ipld-core", 1036 1122 "iroh-car", 1037 1123 "log", 1038 1124 "multibase", 1039 1039 - "rusqlite", 1040 1125 "serde", 1041 1126 "serde_bytes", 1042 1127 "serde_ipld_dagcbor", ··· 1047 1132 ] 1048 1133 1049 1134 [[package]] 1050 1050 - name = "rusqlite" 1051 1051 - version = "0.37.0" 1135 1135 + name = "rustc-demangle" 1136 1136 + version = "0.1.26" 1052 1137 source = "registry+https://github.com/rust-lang/crates.io-index" 1053 1053 - checksum = "165ca6e57b20e1351573e3729b958bc62f0e48025386970b6e4d29e7a7e71f3f" 1054 1054 - dependencies = [ 1055 1055 - "bitflags", 1056 1056 - "fallible-iterator", 1057 1057 - "fallible-streaming-iterator", 1058 1058 - "hashlink", 1059 1059 - "libsqlite3-sys", 1060 1060 - "smallvec", 1061 1061 - ] 1138 1138 + checksum = "56f7d92ca342cea22a06f2121d944b4fd82af56988c270852495420f961d4ace" 1062 1139 1063 1140 [[package]] 1064 1064 - name = "rustc-demangle" 1065 1065 - version = "0.1.26" 1141 1141 + name = "rustc-hash" 1142 1142 + version = "2.1.1" 1066 1143 source = "registry+https://github.com/rust-lang/crates.io-index" 1067 1067 - checksum = "56f7d92ca342cea22a06f2121d944b4fd82af56988c270852495420f961d4ace" 1144 1144 + checksum = "357703d41365b4b27c590e3ed91eabb1b663f07c4c084095e60cbed4362dff0d" 1068 1145 1069 1146 [[package]] 1070 1147 name = "rustix" ··· 1105 1182 version = "1.2.0" 1106 1183 source = "registry+https://github.com/rust-lang/crates.io-index" 1107 1184 checksum = "94143f37725109f92c262ed2cf5e59bce7498c01bcc1502d7b9afe439a4e9f49" 1185 1185 + 1186 1186 + [[package]] 1187 1187 + name = "self_cell" 1188 1188 + version = "1.2.2" 1189 1189 + source = "registry+https://github.com/rust-lang/crates.io-index" 1190 1190 + checksum = "b12e76d157a900eb52e81bc6e9f3069344290341720e9178cde2407113ac8d89" 1108 1191 1109 1192 [[package]] 1110 1193 name = "serde" ··· 1172 1255 ] 1173 1256 1174 1257 [[package]] 1258 1258 + name = "sfa" 1259 1259 + version = "1.0.0" 1260 1260 + source = "registry+https://github.com/rust-lang/crates.io-index" 1261 1261 + checksum = "a1296838937cab56cd6c4eeeb8718ec777383700c33f060e2869867bd01d1175" 1262 1262 + dependencies = [ 1263 1263 + "byteorder-lite", 1264 1264 + "log", 1265 1265 + "xxhash-rust", 1266 1266 + ] 1267 1267 + 1268 1268 + [[package]] 1175 1269 name = "sha2" 1176 1270 version = "0.10.9" 1177 1271 source = "registry+https://github.com/rust-lang/crates.io-index" ··· 1211 1305 dependencies = [ 1212 1306 "libc", 1213 1307 "windows-sys 0.59.0", 1308 1308 + ] 1309 1309 + 1310 1310 + [[package]] 1311 1311 + name = "spin" 1312 1312 + version = "0.9.8" 1313 1313 + source = "registry+https://github.com/rust-lang/crates.io-index" 1314 1314 + checksum = "6980e8d7511241f8acf4aebddbb1ff938df5eebe98691418c4468d0b72a96a67" 1315 1315 + dependencies = [ 1316 1316 + "lock_api", 1214 1317 ] 1215 1318 1216 1319 [[package]] ··· 1372 1475 checksum = "06abde3611657adf66d383f00b093d7faecc7fa57071cce2578660c9f1010821" 1373 1476 1374 1477 [[package]] 1375 1375 - name = "vcpkg" 1376 1376 - version = "0.2.15" 1478 1478 + name = "varint-rs" 1479 1479 + version = "2.2.0" 1377 1480 source = "registry+https://github.com/rust-lang/crates.io-index" 1378 1378 - checksum = "accd4ea62f7bb7a82fe23066fb0957d48ef677f6eeb8215f372f52e48bb32426" 1481 1481 + checksum = "8f54a172d0620933a27a4360d3db3e2ae0dd6cceae9730751a036bbf182c4b23" 1379 1482 1380 1483 [[package]] 1381 1484 name = "version_check" ··· 1659 1762 version = "0.46.0" 1660 1763 source = "registry+https://github.com/rust-lang/crates.io-index" 1661 1764 checksum = "f17a85883d4e6d00e8a97c586de764dabcc06133f7f1d55dce5cdc070ad7fe59" 1765 1765 + 1766 1766 + [[package]] 1767 1767 + name = "xxhash-rust" 1768 1768 + version = "0.8.15" 1769 1769 + source = "registry+https://github.com/rust-lang/crates.io-index" 1770 1770 + checksum = "fdd20c5420375476fbd4394763288da7eb0cc0b8c11deed431a91562af7335d3" 1662 1771 1663 1772 [[package]] 1664 1773 name = "zerocopy"

+3 -3

Cargo.toml

··· 1 1 [package] 2 2 name = "repo-stream" 3 3 - version = "0.1.1" 3 3 + version = "0.2.2" 4 4 edition = "2024" 5 5 license = "MIT OR Apache-2.0" 6 6 - description = "Fast and robust atproto CAR file processing in rust" 6 6 + description = "A robust CAR file -> MST walker for atproto" 7 7 repository = "https://tangled.org/@microcosm.blue/repo-stream" 8 8 9 9 [dependencies] 10 10 bincode = { version = "2.0.1", features = ["serde"] } 11 11 + fjall = { version = "3.0.1", default-features = false } 11 12 futures = "0.3.31" 12 13 futures-core = "0.3.31" 13 14 ipld-core = { version = "0.4.2", features = ["serde"] } 14 15 iroh-car = "0.5.1" 15 16 log = "0.4.28" 16 17 multibase = "0.9.2" 17 17 - rusqlite = "0.37.0" 18 18 serde = { version = "1.0.228", features = ["derive"] } 19 19 serde_bytes = "0.11.19" 20 20 serde_ipld_dagcbor = "0.6.4"

+12 -21

benches/huge-car.rs

··· 1 1 extern crate repo_stream; 2 2 - use futures::TryStreamExt; 3 3 - use iroh_car::CarReader; 4 4 - use std::convert::Infallible; 2 2 + use repo_stream::Driver; 5 3 use std::path::{Path, PathBuf}; 6 4 7 5 use criterion::{Criterion, criterion_group, criterion_main}; ··· 20 18 }); 21 19 } 22 20 23 23 - async fn drive_car(filename: impl AsRef<Path>) { 21 21 + async fn drive_car(filename: impl AsRef<Path>) -> usize { 24 22 let reader = tokio::fs::File::open(filename).await.unwrap(); 25 23 let reader = tokio::io::BufReader::new(reader); 26 26 - let reader = CarReader::new(reader).await.unwrap(); 27 24 28 28 - let root = reader 29 29 - .header() 30 30 - .roots() 31 31 - .first() 32 32 - .ok_or("missing root") 25 25 + let mut driver = match Driver::load_car(reader, |block| block.len(), 1024) 26 26 + .await 33 27 .unwrap() 34 34 - .clone(); 35 35 - 36 36 - let stream = std::pin::pin!(reader.stream()); 37 37 - 38 38 - let (_commit, v) = 39 39 - repo_stream::drive::Vehicle::init(root, stream, |block| Ok::<_, Infallible>(block.len())) 40 40 - .await 41 41 - .unwrap(); 42 42 - let mut record_stream = std::pin::pin!(v.stream()); 28 28 + { 29 29 + Driver::Memory(_, mem_driver) => mem_driver, 30 30 + Driver::Disk(_) => panic!("not doing disk for benchmark"), 31 31 + }; 43 32 44 44 - while let Some(_) = record_stream.try_next().await.unwrap() { 45 45 - // just here for the drive 33 33 + let mut n = 0; 34 34 + while let Some(pairs) = driver.next_chunk(256).await.unwrap() { 35 35 + n += pairs.len(); 46 36 } 37 37 + n 47 38 } 48 39 49 40 criterion_group!(benches, criterion_benchmark);

+16 -22

benches/non-huge-cars.rs

··· 1 1 extern crate repo_stream; 2 2 - use futures::TryStreamExt; 3 3 - use iroh_car::CarReader; 4 4 - use std::convert::Infallible; 2 2 + use repo_stream::Driver; 5 3 6 4 use criterion::{Criterion, criterion_group, criterion_main}; 7 5 6 6 + const EMPTY_CAR: &'static [u8] = include_bytes!("../car-samples/empty.car"); 8 7 const TINY_CAR: &'static [u8] = include_bytes!("../car-samples/tiny.car"); 9 8 const LITTLE_CAR: &'static [u8] = include_bytes!("../car-samples/little.car"); 10 9 const MIDSIZE_CAR: &'static [u8] = include_bytes!("../car-samples/midsize.car"); ··· 15 14 .build() 16 15 .expect("Creating runtime failed"); 17 16 17 17 + c.bench_function("empty-car", |b| { 18 18 + b.to_async(&rt).iter(async || drive_car(EMPTY_CAR).await) 19 19 + }); 18 20 c.bench_function("tiny-car", |b| { 19 21 b.to_async(&rt).iter(async || drive_car(TINY_CAR).await) 20 22 }); ··· 26 28 }); 27 29 } 28 30 29 29 - async fn drive_car(bytes: &[u8]) { 30 30 - let reader = CarReader::new(bytes).await.unwrap(); 31 31 - 32 32 - let root = reader 33 33 - .header() 34 34 - .roots() 35 35 - .first() 36 36 - .ok_or("missing root") 31 31 + async fn drive_car(bytes: &[u8]) -> usize { 32 32 + let mut driver = match Driver::load_car(bytes, |block| block.len(), 32) 33 33 + .await 37 34 .unwrap() 38 38 - .clone(); 39 39 - 40 40 - let stream = std::pin::pin!(reader.stream()); 35 35 + { 36 36 + Driver::Memory(_, mem_driver) => mem_driver, 37 37 + Driver::Disk(_) => panic!("not benching big cars here"), 38 38 + }; 41 39 42 42 - let (_commit, v) = 43 43 - repo_stream::drive::Vehicle::init(root, stream, |block| Ok::<_, Infallible>(block.len())) 44 44 - .await 45 45 - .unwrap(); 46 46 - let mut record_stream = std::pin::pin!(v.stream()); 47 47 - 48 48 - while let Some(_) = record_stream.try_next().await.unwrap() { 49 49 - // just here for the drive 40 40 + let mut n = 0; 41 41 + while let Some(pairs) = driver.next_chunk(256).await.unwrap() { 42 42 + n += pairs.len(); 50 43 } 44 44 + n 51 45 } 52 46 53 47 criterion_group!(benches, criterion_benchmark);

car-samples/empty.car

This is a binary file and will not be displayed.

+54 -34

examples/disk-read-file/main.rs

··· 1 1 + /*! 2 2 + Read a CAR file by spilling to disk 3 3 + */ 4 4 + 1 5 extern crate repo_stream; 2 6 use clap::Parser; 3 3 - use repo_stream::drive::Processable; 4 4 - use serde::{Deserialize, Serialize}; 7 7 + use repo_stream::{DiskBuilder, Driver, DriverBuilder}; 5 8 use std::path::PathBuf; 6 6 - 7 7 - type Result<T> = std::result::Result<T, Box<dyn std::error::Error>>; 9 9 + use std::time::Instant; 8 10 9 11 #[derive(Debug, Parser)] 10 12 struct Args { ··· 14 16 tmpfile: PathBuf, 15 17 } 16 18 17 17 - #[derive(Clone, Serialize, Deserialize)] 18 18 - struct S(usize); 19 19 - 20 20 - impl Processable for S { 21 21 - fn get_size(&self) -> usize { 22 22 - 0 // no additional space taken, just its stack size (newtype is free) 23 23 - } 24 24 - } 25 25 - 26 19 #[tokio::main] 27 27 - async fn main() -> Result<()> { 20 20 + async fn main() -> Result<(), Box<dyn std::error::Error>> { 28 21 env_logger::init(); 29 22 30 23 let Args { car, tmpfile } = Args::parse(); 24 24 + 25 25 + // repo-stream takes an AsyncRead as input. wrapping a filesystem read in 26 26 + // BufReader can provide a really significant performance win. 31 27 let reader = tokio::fs::File::open(car).await?; 32 28 let reader = tokio::io::BufReader::new(reader); 33 29 34 34 - // let kb = 2_usize.pow(10); 35 35 - let mb = 2_usize.pow(20); 30 30 + log::info!("hello! reading the car..."); 31 31 + let t0 = Instant::now(); 36 32 37 37 - let limit_mb = 32; 38 38 - 39 39 - let driver = match repo_stream::drive::load_car(reader, |block| S(block.len()), 10 * mb).await? 33 33 + // in this example we only bother handling CARs that are too big for memory 34 34 + // `noop` helper means: do no block processing, store the raw blocks 35 35 + let driver = match DriverBuilder::new() 36 36 + .with_mem_limit_mb(32) // how much memory can be used before disk spill 37 37 + .load_car(reader) 38 38 + .await? 40 39 { 41 41 - repo_stream::drive::Vehicle::Lil(_, _) => panic!("try this on a bigger car"), 42 42 - repo_stream::drive::Vehicle::Big(big_stuff) => { 43 43 - let disk_store = repo_stream::disk::SqliteStore::new(tmpfile.clone(), limit_mb); 40 40 + Driver::Memory(_, _) => panic!("try this on a bigger car"), 41 41 + Driver::Disk(big_stuff) => { 42 42 + // we reach here if the repo was too big and needs to be spilled to 43 43 + // disk to continue 44 44 + 45 45 + // set up a disk store we can spill to 46 46 + let disk_store = DiskBuilder::new().open(tmpfile).await?; 47 47 + 48 48 + // do the spilling, get back a (similar) driver 44 49 let (commit, driver) = big_stuff.finish_loading(disk_store).await?; 45 45 - log::warn!("big: {:?}", commit); 50 50 + 51 51 + // at this point you might want to fetch the account's signing key 52 52 + // via the DID from the commit, and then verify the signature. 53 53 + log::warn!("big's comit ({:?}): {:?}", t0.elapsed(), commit); 54 54 + 55 55 + // pop the driver back out to get some code indentation relief 46 56 driver 47 57 } 48 58 }; 49 59 60 60 + // collect some random stats about the blocks 50 61 let mut n = 0; 51 51 - let (mut rx, worker) = driver.rx(512).await?; 62 62 + let mut zeros = 0; 63 63 + 64 64 + log::info!("walking..."); 52 65 53 53 - log::debug!("walking..."); 54 54 - while let Some(pairs) = rx.recv().await { 66 66 + // this example uses the disk driver's channel mode: the tree walking is 67 67 + // spawned onto a blocking thread, and we get chunks of rkey+blocks back 68 68 + let (mut rx, join) = driver.to_channel(512); 69 69 + while let Some(r) = rx.recv().await { 70 70 + let pairs = r?; 71 71 + 72 72 + // keep a count of the total number of blocks seen 55 73 n += pairs.len(); 56 56 - } 57 57 - log::debug!("done walking! joining..."); 58 74 59 59 - worker.await.unwrap().unwrap(); 75 75 + for (_, block) in pairs { 76 76 + // for each block, count how many bytes are equal to '0' 77 77 + // (this is just an example, you probably want to do something more 78 78 + // interesting) 79 79 + zeros += block.into_iter().filter(|&b| b == b'0').count() 80 80 + } 81 81 + } 60 82 61 61 - log::debug!("joined."); 83 83 + log::info!("arrived! ({:?}) joining rx...", t0.elapsed()); 62 84 63 63 - // log::info!("now is the time to check mem..."); 64 64 - // tokio::time::sleep(std::time::Duration::from_secs(22)).await; 65 65 - log::info!("bye! {n}"); 85 85 + join.await?; 66 86 67 67 - std::fs::remove_file(tmpfile).unwrap(); // need to also remove -shm -wal 87 87 + log::info!("done. n={n} zeros={zeros}"); 68 88 69 89 Ok(()) 70 90 }

+14 -17

examples/read-file/main.rs

··· 1 1 + /*! 2 2 + Read a CAR file with in-memory processing 3 3 + */ 4 4 + 1 5 extern crate repo_stream; 2 6 use clap::Parser; 3 3 - use repo_stream::drive::Processable; 4 4 - use serde::{Deserialize, Serialize}; 7 7 + use repo_stream::{Driver, DriverBuilder}; 5 8 use std::path::PathBuf; 6 9 7 10 type Result<T> = std::result::Result<T, Box<dyn std::error::Error>>; ··· 10 13 struct Args { 11 14 #[arg()] 12 15 file: PathBuf, 13 13 - } 14 14 - 15 15 - #[derive(Clone, Serialize, Deserialize)] 16 16 - struct S(usize); 17 17 - 18 18 - impl Processable for S { 19 19 - fn get_size(&self) -> usize { 20 20 - 0 // no additional space taken, just its stack size (newtype is free) 21 21 - } 22 16 } 23 17 24 18 #[tokio::main] ··· 29 23 let reader = tokio::fs::File::open(file).await?; 30 24 let reader = tokio::io::BufReader::new(reader); 31 25 32 32 - let (commit, mut driver) = 33 33 - match repo_stream::drive::load_car(reader, |block| S(block.len()), 1024 * 1024).await? { 34 34 - repo_stream::drive::Vehicle::Lil(commit, mem_driver) => (commit, mem_driver), 35 35 - repo_stream::drive::Vehicle::Big(_) => panic!("can't handle big cars yet"), 36 36 - }; 26 26 + let (commit, mut driver) = match DriverBuilder::new() 27 27 + .with_block_processor(|block| block.len()) 28 28 + .load_car(reader) 29 29 + .await? 30 30 + { 31 31 + Driver::Memory(commit, mem_driver) => (commit, mem_driver), 32 32 + Driver::Disk(_) => panic!("this example doesn't handle big CARs"), 33 33 + }; 37 34 38 35 log::info!("got commit: {commit:?}"); 39 36 ··· 42 39 n += pairs.len(); 43 40 // log::info!("got {rkey:?}"); 44 41 } 45 45 - log::info!("bye! {n}"); 42 42 + log::info!("bye! total records={n}"); 46 43 47 44 Ok(()) 48 45 }

+67 -2

readme.md

··· 1 1 # repo-stream 2 2 3 3 - Fast and (aspirationally) robust atproto CAR file processing in rust 3 3 + A robust CAR file -> MST walker for atproto 4 4 + 5 5 + [![Crates.io][crates-badge]](https://crates.io/crates/repo-stream) 6 6 + [![Documentation][docs-badge]](https://docs.rs/repo-stream) 7 7 + [![Sponsor][sponsor-badge]](https://github.com/sponsors/uniphil) 8 8 + 9 9 + [crates-badge]: https://img.shields.io/crates/v/repo-stream.svg 10 10 + [docs-badge]: https://docs.rs/repo-stream/badge.svg 11 11 + [sponsor-badge]: https://img.shields.io/badge/at-microcosm-b820f9?labelColor=b820f9&logo=githubsponsors&logoColor=fff 12 12 + 13 13 + ```rust 14 14 + use repo_stream::{Driver, DriverBuilder, DriveError, DiskBuilder}; 15 15 + 16 16 + #[tokio::main] 17 17 + async fn main() -> Result<(), DriveError> { 18 18 + // repo-stream takes any AsyncRead as input, like a tokio::fs::File 19 19 + let reader = tokio::fs::File::open("repo.car".into()).await?; 20 20 + let reader = tokio::io::BufReader::new(reader); 21 21 + 22 22 + // example repo workload is simply counting the total record bytes 23 23 + let mut total_size = 0; 24 24 + 25 25 + match DriverBuilder::new() 26 26 + .with_mem_limit_mb(10) 27 27 + .with_block_processor(|rec| rec.len()) // block processing: just extract the raw record size 28 28 + .load_car(reader) 29 29 + .await? 30 30 + { 31 31 + 32 32 + // if all blocks fit within memory 33 33 + Driver::Memory(_commit, mut driver) => { 34 34 + while let Some(chunk) = driver.next_chunk(256).await? { 35 35 + for (_rkey, size) in chunk { 36 36 + total_size += size; 37 37 + } 38 38 + } 39 39 + }, 40 40 + 41 41 + // if the CAR was too big for in-memory processing 42 42 + Driver::Disk(paused) => { 43 43 + // set up a disk store we can spill to 44 44 + let store = DiskBuilder::new().open("some/path.db".into()).await?; 45 45 + // do the spilling, get back a (similar) driver 46 46 + let (_commit, mut driver) = paused.finish_loading(store).await?; 47 47 + 48 48 + while let Some(chunk) = driver.next_chunk(256).await? { 49 49 + for (_rkey, size) in chunk { 50 50 + total_size += size; 51 51 + } 52 52 + } 53 53 + } 54 54 + }; 55 55 + println!("sum of size of all records: {total_size}"); 56 56 + Ok(()) 57 57 + } 58 58 + ``` 59 59 + 60 60 + more recent todo 61 61 + 62 62 + - [ ] get an *emtpy* car for the test suite 63 63 + - [x] implement a max size on disk limit 64 64 + 65 65 + 66 66 + ----- 67 67 + 68 68 + older stuff (to clean up): 4 69 5 70 6 71 current car processing times (records processed into their length usize, phil's dev machine): ··· 27 92 -> yeah the commit is returned from init 28 93 - [ ] spec compliance todos 29 94 - [x] assert that keys are ordered and fail if not 30 30 - - [ ] verify node mst depth from key (possibly pending [interop test fixes](https://github.com/bluesky-social/atproto-interop-tests/issues/5)) 95 95 + - [x] verify node mst depth from key (possibly pending [interop test fixes](https://github.com/bluesky-social/atproto-interop-tests/issues/5)) 31 96 - [ ] performance todos 32 97 - [x] consume the serialized nodes into a mutable efficient format 33 98 - [ ] maybe customize the deserialize impl to do that directly?

+138 -81

src/disk.rs

··· 1 1 - use rusqlite::OptionalExtension; 2 2 - use std::path::PathBuf; 1 1 + /*! 2 2 + Disk storage for blocks on disk 3 3 4 4 - pub struct SqliteStore { 5 5 - path: PathBuf, 6 6 - limit_mb: usize, 7 7 - } 4 4 + Currently this uses sqlite. In testing sqlite wasn't the fastest, but it seemed 5 5 + to be the best behaved in terms of both on-disk space usage and memory usage. 8 6 9 9 - impl SqliteStore { 10 10 - pub fn new(path: PathBuf, limit_mb: usize) -> Self { 11 11 - Self { path, limit_mb } 12 12 - } 13 13 - } 7 7 + ```no_run 8 8 + # use repo_stream::{DiskBuilder, DiskError}; 9 9 + # #[tokio::main] 10 10 + # async fn main() -> Result<(), DiskError> { 11 11 + let store = DiskBuilder::new() 12 12 + .with_cache_size_mb(32) 13 13 + .with_max_stored_mb(1024) // errors when >1GiB of processed blocks are inserted 14 14 + .open("/some/path.db".into()).await?; 15 15 + # Ok(()) 16 16 + # } 17 17 + ``` 18 18 + */ 14 19 15 15 - impl SqliteStore { 16 16 - pub async fn get_access(&mut self) -> Result<SqliteAccess, rusqlite::Error> { 17 17 - let path = self.path.clone(); 18 18 - let limit_mb = self.limit_mb; 19 19 - let conn = tokio::task::spawn_blocking(move || { 20 20 - let conn = rusqlite::Connection::open(path)?; 21 21 - 22 22 - let sq_mb = -(2_i64.pow(10)); // negative is kibibytes for sqlite cache_size 20 20 + use crate::drive::DriveError; 21 21 + use fjall::config::{CompressionPolicy, PinningPolicy, RestartIntervalPolicy}; 22 22 + use fjall::{CompressionType, Database, Error as FjallError, Keyspace, KeyspaceCreateOptions}; 23 23 + use std::path::PathBuf; 23 24 24 24 - // conn.pragma_update(None, "journal_mode", "OFF")?; 25 25 - // conn.pragma_update(None, "journal_mode", "MEMORY")?; 26 26 - conn.pragma_update(None, "journal_mode", "WAL")?; 27 27 - // conn.pragma_update(None, "wal_autocheckpoint", "0")?; // this lets things get a bit big on disk 28 28 - conn.pragma_update(None, "synchronous", "OFF")?; 29 29 - conn.pragma_update(None, "cache_size", (limit_mb as i64 * sq_mb).to_string())?; 30 30 - conn.execute( 31 31 - "CREATE TABLE blocks ( 32 32 - key BLOB PRIMARY KEY NOT NULL, 33 33 - val BLOB NOT NULL 34 34 - ) WITHOUT ROWID", 35 35 - (), 36 36 - )?; 25 25 + #[derive(Debug, thiserror::Error)] 26 26 + pub enum DiskError { 27 27 + /// A wrapped database error 28 28 + /// 29 29 + /// (The wrapped err should probably be obscured to remove public-facing 30 30 + /// sqlite bits) 31 31 + #[error(transparent)] 32 32 + DbError(#[from] FjallError), 33 33 + /// A tokio blocking task failed to join 34 34 + #[error("Failed to join a tokio blocking task: {0}")] 35 35 + JoinError(#[from] tokio::task::JoinError), 36 36 + /// The total size of stored blocks exceeded the allowed size 37 37 + /// 38 38 + /// If you need to process *really* big CARs, you can configure a higher 39 39 + /// limit. 40 40 + #[error("Maximum disk size reached")] 41 41 + MaxSizeExceeded, 42 42 + } 37 43 38 38 - Ok::<_, rusqlite::Error>(conn) 39 39 - }) 40 40 - .await 41 41 - .expect("join error")?; 44 44 + /// Builder-style disk store setup 45 45 + #[derive(Debug, Clone)] 46 46 + pub struct DiskBuilder { 47 47 + /// Database in-memory cache allowance 48 48 + /// 49 49 + /// Default: 32 MiB 50 50 + pub cache_size_mb: usize, 51 51 + /// Database stored block size limit 52 52 + /// 53 53 + /// Default: 10 GiB 54 54 + /// 55 55 + /// Note: actual size on disk may be more, but should approximately scale 56 56 + /// with this limit 57 57 + pub max_stored_mb: usize, 58 58 + } 42 59 43 43 - Ok(SqliteAccess { conn }) 60 60 + impl Default for DiskBuilder { 61 61 + fn default() -> Self { 62 62 + Self { 63 63 + cache_size_mb: 64, 64 64 + max_stored_mb: 10 * 1024, // 10 GiB 65 65 + } 44 66 } 45 67 } 46 68 47 47 - pub struct SqliteAccess { 48 48 - conn: rusqlite::Connection, 49 49 - } 50 50 - 51 51 - impl SqliteAccess { 52 52 - pub fn get_writer(&'_ mut self) -> Result<SqliteWriter<'_>, rusqlite::Error> { 53 53 - let tx = self.conn.transaction()?; 54 54 - // let insert_stmt = tx.prepare("INSERT INTO blocks (key, val) VALUES (?1, ?2)")?; 55 55 - Ok(SqliteWriter { tx: Some(tx) }) 69 69 + impl DiskBuilder { 70 70 + /// Begin configuring the storage with defaults 71 71 + pub fn new() -> Self { 72 72 + Default::default() 73 73 + } 74 74 + /// Set the in-memory cache allowance for the database 75 75 + /// 76 76 + /// Default: 64 MiB 77 77 + pub fn with_cache_size_mb(mut self, size: usize) -> Self { 78 78 + self.cache_size_mb = size; 79 79 + self 80 80 + } 81 81 + /// Set the approximate stored block size limit 82 82 + /// 83 83 + /// Default: 10 GiB 84 84 + pub fn with_max_stored_mb(mut self, max: usize) -> Self { 85 85 + self.max_stored_mb = max; 86 86 + self 56 87 } 57 57 - pub fn get_reader(&'_ self) -> Result<SqliteReader<'_>, rusqlite::Error> { 58 58 - let select_stmt = self.conn.prepare("SELECT val FROM blocks WHERE key = ?1")?; 59 59 - Ok(SqliteReader { select_stmt }) 88 88 + /// Open and initialize the actual disk storage 89 89 + pub async fn open(&self, path: PathBuf) -> Result<DiskStore, DiskError> { 90 90 + DiskStore::new(path, self.cache_size_mb, self.max_stored_mb).await 60 91 } 61 92 } 62 93 63 63 - pub struct SqliteWriter<'conn> { 64 64 - tx: Option<rusqlite::Transaction<'conn>>, 94 94 + /// On-disk block storage 95 95 + pub struct DiskStore { 96 96 + #[allow(unused)] 97 97 + db: Database, 98 98 + partition: Keyspace, 99 99 + max_stored: usize, 100 100 + stored: usize, 65 101 } 66 102 67 67 - /// oops careful in async 68 68 - impl Drop for SqliteWriter<'_> { 69 69 - fn drop(&mut self) { 70 70 - let tx = self.tx.take(); 71 71 - tx.unwrap().commit().unwrap(); 103 103 + impl DiskStore { 104 104 + /// Initialize a new disk store 105 105 + pub async fn new( 106 106 + path: PathBuf, 107 107 + cache_mb: usize, 108 108 + max_stored_mb: usize, 109 109 + ) -> Result<Self, DiskError> { 110 110 + let max_stored = max_stored_mb * 2_usize.pow(20); 111 111 + let (db, partition) = tokio::task::spawn_blocking(move || { 112 112 + let db = Database::builder(path) 113 113 + // .manual_journal_persist(true) 114 114 + // .flush_workers(1) 115 115 + // .compaction_workers(1) 116 116 + .journal_compression(CompressionType::None) 117 117 + .cache_size(cache_mb as u64 * 2_u64.pow(20)) 118 118 + .temporary(true) 119 119 + .open()?; 120 120 + let opts = KeyspaceCreateOptions::default() 121 121 + .data_block_restart_interval_policy(RestartIntervalPolicy::all(8)) 122 122 + .filter_block_pinning_policy(PinningPolicy::disabled()) 123 123 + .expect_point_read_hits(true) 124 124 + .data_block_compression_policy(CompressionPolicy::disabled()) 125 125 + .manual_journal_persist(true) 126 126 + .max_memtable_size(32 * 2_u64.pow(20)); 127 127 + let partition = db.keyspace("z", || opts)?; 128 128 + 129 129 + Ok::<_, DiskError>((db, partition)) 130 130 + }) 131 131 + .await??; 132 132 + 133 133 + Ok(Self { 134 134 + db, 135 135 + partition, 136 136 + max_stored, 137 137 + stored: 0, 138 138 + }) 72 139 } 73 73 - } 74 140 75 75 - impl SqliteWriter<'_> { 76 76 - pub fn put(&mut self, key: Vec<u8>, val: Vec<u8>) -> rusqlite::Result<()> { 77 77 - let tx = self.tx.as_ref().unwrap(); 78 78 - let mut insert_stmt = tx.prepare_cached("INSERT INTO blocks (key, val) VALUES (?1, ?2)")?; 79 79 - insert_stmt.execute((key, val))?; 80 80 - Ok(()) 81 81 - } 82 82 - pub fn put_many( 141 141 + pub(crate) fn put_many( 83 142 &mut self, 84 84 - kv: impl Iterator<Item = (Vec<u8>, Vec<u8>)>, 85 85 - ) -> rusqlite::Result<()> { 86 86 - let tx = self.tx.as_ref().unwrap(); 87 87 - let mut insert_stmt = tx.prepare_cached("INSERT INTO blocks (key, val) VALUES (?1, ?2)")?; 88 88 - for (k, v) in kv { 89 89 - insert_stmt.execute((k, v))?; 143 143 + kv: impl Iterator<Item = Result<(Vec<u8>, Vec<u8>), DriveError>>, 144 144 + ) -> Result<(), DriveError> { 145 145 + let mut batch = self.db.batch(); 146 146 + for pair in kv { 147 147 + let (k, v) = pair?; 148 148 + self.stored += v.len(); 149 149 + if self.stored > self.max_stored { 150 150 + return Err(DiskError::MaxSizeExceeded.into()); 151 151 + } 152 152 + batch.insert(&self.partition, k, v); 90 153 } 154 154 + batch.commit().map_err(DiskError::DbError)?; 91 155 Ok(()) 92 156 } 93 93 - } 94 157 95 95 - pub struct SqliteReader<'conn> { 96 96 - select_stmt: rusqlite::Statement<'conn>, 97 97 - } 98 98 - 99 99 - impl SqliteReader<'_> { 100 100 - pub fn get(&mut self, key: Vec<u8>) -> rusqlite::Result<Option<Vec<u8>>> { 101 101 - self.select_stmt 102 102 - .query_one((&key,), |row| row.get(0)) 103 103 - .optional() 158 158 + #[inline] 159 159 + pub(crate) fn get(&mut self, key: &[u8]) -> Result<Option<fjall::Slice>, FjallError> { 160 160 + self.partition.get(key) 104 161 } 105 162 }

+390 -230

src/drive.rs

··· 1 1 - //! Consume an MST block stream, producing an ordered stream of records 1 1 + //! Consume a CAR from an AsyncRead, producing an ordered stream of records 2 2 3 3 - use crate::disk::{SqliteAccess, SqliteStore}; 3 3 + use crate::disk::{DiskError, DiskStore}; 4 4 + use crate::process::Processable; 4 5 use ipld_core::cid::Cid; 5 6 use iroh_car::CarReader; 6 6 - use serde::de::DeserializeOwned; 7 7 use serde::{Deserialize, Serialize}; 8 8 use std::collections::HashMap; 9 9 use std::convert::Infallible; 10 10 - use tokio::io::AsyncRead; 10 10 + use tokio::{io::AsyncRead, sync::mpsc}; 11 11 12 12 use crate::mst::{Commit, Node}; 13 13 - use crate::walk::{DiskTrip, Step, Trip, Walker}; 13 13 + use crate::walk::{Step, WalkError, Walker}; 14 14 15 15 /// Errors that can happen while consuming and emitting blocks and records 16 16 #[derive(Debug, thiserror::Error)] ··· 24 24 #[error("The MST block {0} could not be found")] 25 25 MissingBlock(Cid), 26 26 #[error("Failed to walk the mst tree: {0}")] 27 27 - Tripped(#[from] Trip), 27 27 + WalkError(#[from] WalkError), 28 28 #[error("CAR file had no roots")] 29 29 MissingRoot, 30 30 - } 31 31 - 32 32 - #[derive(Debug, thiserror::Error)] 33 33 - pub enum DiskDriveError { 34 34 - #[error("Error from iroh_car: {0}")] 35 35 - CarReader(#[from] iroh_car::Error), 36 36 - #[error("Failed to decode commit block: {0}")] 37 37 - BadBlock(#[from] serde_ipld_dagcbor::DecodeError<Infallible>), 38 30 #[error("Storage error")] 39 39 - StorageError(#[from] rusqlite::Error), 40 40 - #[error("The Commit block reference by the root was not found")] 41 41 - MissingCommit, 42 42 - #[error("The MST block {0} could not be found")] 43 43 - MissingBlock(Cid), 31 31 + StorageError(#[from] DiskError), 44 32 #[error("Encode error: {0}")] 45 33 BincodeEncodeError(#[from] bincode::error::EncodeError), 46 46 - #[error("Decode error: {0}")] 47 47 - BincodeDecodeError(#[from] bincode::error::DecodeError), 48 48 - #[error("disk tripped: {0}")] 49 49 - DiskTripped(#[from] DiskTrip), 34 34 + #[error("Tried to send on a closed channel")] 35 35 + ChannelSendError, // SendError takes <T> which we don't need 36 36 + #[error("Failed to join a task: {0}")] 37 37 + JoinError(#[from] tokio::task::JoinError), 50 38 } 51 39 52 52 - pub trait Processable: Clone + Serialize + DeserializeOwned { 53 53 - /// the additional size taken up (not including its mem::size_of) 54 54 - fn get_size(&self) -> usize; 40 40 + #[derive(Debug, thiserror::Error)] 41 41 + pub enum DecodeError { 42 42 + #[error(transparent)] 43 43 + BincodeDecodeError(#[from] bincode::error::DecodeError), 44 44 + #[error("extra bytes remained after decoding")] 45 45 + ExtraGarbage, 55 46 } 56 47 48 48 + /// An in-order chunk of Rkey + (processed) Block pairs 49 49 + pub type BlockChunk<T> = Vec<(String, T)>; 50 50 + 57 51 #[derive(Debug, Clone, Serialize, Deserialize)] 58 58 - pub enum MaybeProcessedBlock<T> { 52 52 + pub(crate) enum MaybeProcessedBlock<T> { 59 53 /// A block that's *probably* a Node (but we can't know yet) 60 54 /// 61 55 /// It *can be* a record that suspiciously looks a lot like a node, so we ··· 97 91 } 98 92 } 99 93 100 100 - pub enum Vehicle<R: AsyncRead + Unpin, T: Processable> { 101 101 - Lil(Commit, MemDriver<T>), 102 102 - Big(BigCar<R, T>), 94 94 + impl<T> MaybeProcessedBlock<T> { 95 95 + fn maybe(process: fn(Vec<u8>) -> T, data: Vec<u8>) -> Self { 96 96 + if Node::could_be(&data) { 97 97 + MaybeProcessedBlock::Raw(data) 98 98 + } else { 99 99 + MaybeProcessedBlock::Processed(process(data)) 100 100 + } 101 101 + } 103 102 } 104 103 105 105 - pub async fn load_car<R: AsyncRead + Unpin, T: Processable>( 106 106 - reader: R, 107 107 - process: fn(Vec<u8>) -> T, 108 108 - max_size: usize, 109 109 - ) -> Result<Vehicle<R, T>, DriveError> { 110 110 - let mut mem_blocks = HashMap::new(); 104 104 + /// Read a CAR file, buffering blocks in memory or to disk 105 105 + pub enum Driver<R: AsyncRead + Unpin, T: Processable> { 106 106 + /// All blocks fit within the memory limit 107 107 + /// 108 108 + /// You probably want to check the commit's signature. You can go ahead and 109 109 + /// walk the MST right away. 110 110 + Memory(Commit, MemDriver<T>), 111 111 + /// Blocks exceed the memory limit 112 112 + /// 113 113 + /// You'll need to provide a disk storage to continue. The commit will be 114 114 + /// returned and can be validated only once all blocks are loaded. 115 115 + Disk(NeedDisk<R, T>), 116 116 + } 117 117 + 118 118 + /// Builder-style driver setup 119 119 + #[derive(Debug, Clone)] 120 120 + pub struct DriverBuilder { 121 121 + pub mem_limit_mb: usize, 122 122 + } 111 123 112 112 - let mut car = CarReader::new(reader).await?; 124 124 + impl Default for DriverBuilder { 125 125 + fn default() -> Self { 126 126 + Self { mem_limit_mb: 16 } 127 127 + } 128 128 + } 113 129 114 114 - let root = *car 115 115 - .header() 116 116 - .roots() 117 117 - .first() 118 118 - .ok_or(DriveError::MissingRoot)?; 119 119 - log::debug!("root: {root:?}"); 130 130 + impl DriverBuilder { 131 131 + /// Begin configuring the driver with defaults 132 132 + pub fn new() -> Self { 133 133 + Default::default() 134 134 + } 135 135 + /// Set the in-memory size limit, in MiB 136 136 + /// 137 137 + /// Default: 16 MiB 138 138 + pub fn with_mem_limit_mb(self, new_limit: usize) -> Self { 139 139 + Self { 140 140 + mem_limit_mb: new_limit, 141 141 + } 142 142 + } 143 143 + /// Set the block processor 144 144 + /// 145 145 + /// Default: noop, raw blocks will be emitted 146 146 + pub fn with_block_processor<T: Processable>( 147 147 + self, 148 148 + p: fn(Vec<u8>) -> T, 149 149 + ) -> DriverBuilderWithProcessor<T> { 150 150 + DriverBuilderWithProcessor { 151 151 + mem_limit_mb: self.mem_limit_mb, 152 152 + block_processor: p, 153 153 + } 154 154 + } 155 155 + /// Begin processing an atproto MST from a CAR file 156 156 + pub async fn load_car<R: AsyncRead + Unpin>( 157 157 + &self, 158 158 + reader: R, 159 159 + ) -> Result<Driver<R, Vec<u8>>, DriveError> { 160 160 + Driver::load_car(reader, crate::process::noop, self.mem_limit_mb).await 161 161 + } 162 162 + } 120 163 121 121 - let mut commit = None; 164 164 + /// Builder-style driver intermediate step 165 165 + /// 166 166 + /// start from `DriverBuilder` 167 167 + #[derive(Debug, Clone)] 168 168 + pub struct DriverBuilderWithProcessor<T: Processable> { 169 169 + pub mem_limit_mb: usize, 170 170 + pub block_processor: fn(Vec<u8>) -> T, 171 171 + } 122 172 123 123 - // try to load all the blocks into memory 124 124 - let mut mem_size = 0; 125 125 - while let Some((cid, data)) = car.next_block().await? { 126 126 - // the root commit is a Special Third Kind of block that we need to make 127 127 - // sure not to optimistically send to the processing function 128 128 - if cid == root { 129 129 - let c: Commit = serde_ipld_dagcbor::from_slice(&data)?; 130 130 - commit = Some(c); 131 131 - continue; 173 173 + impl<T: Processable> DriverBuilderWithProcessor<T> { 174 174 + /// Set the in-memory size limit, in MiB 175 175 + /// 176 176 + /// Default: 16 MiB 177 177 + pub fn with_mem_limit_mb(mut self, new_limit: usize) -> Self { 178 178 + self.mem_limit_mb = new_limit; 179 179 + self 180 180 + } 181 181 + /// Begin processing an atproto MST from a CAR file 182 182 + pub async fn load_car<R: AsyncRead + Unpin>( 183 183 + &self, 184 184 + reader: R, 185 185 + ) -> Result<Driver<R, T>, DriveError> { 186 186 + Driver::load_car(reader, self.block_processor, self.mem_limit_mb).await 187 187 + } 188 188 + } 189 189 + 190 190 + impl<R: AsyncRead + Unpin, T: Processable> Driver<R, T> { 191 191 + /// Begin processing an atproto MST from a CAR file 192 192 + /// 193 193 + /// Blocks will be loaded, processed, and buffered in memory. If the entire 194 194 + /// processed size is under the `mem_limit_mb` limit, a `Driver::Memory` 195 195 + /// will be returned along with a `Commit` ready for validation. 196 196 + /// 197 197 + /// If the `mem_limit_mb` limit is reached before loading all blocks, the 198 198 + /// partial state will be returned as `Driver::Disk(needed)`, which can be 199 199 + /// resumed by providing a `SqliteStorage` for on-disk block storage. 200 200 + pub async fn load_car( 201 201 + reader: R, 202 202 + process: fn(Vec<u8>) -> T, 203 203 + mem_limit_mb: usize, 204 204 + ) -> Result<Driver<R, T>, DriveError> { 205 205 + let max_size = mem_limit_mb * 2_usize.pow(20); 206 206 + let mut mem_blocks = HashMap::new(); 207 207 + 208 208 + let mut car = CarReader::new(reader).await?; 209 209 + 210 210 + let root = *car 211 211 + .header() 212 212 + .roots() 213 213 + .first() 214 214 + .ok_or(DriveError::MissingRoot)?; 215 215 + log::debug!("root: {root:?}"); 216 216 + 217 217 + let mut commit = None; 218 218 + 219 219 + // try to load all the blocks into memory 220 220 + let mut mem_size = 0; 221 221 + while let Some((cid, data)) = car.next_block().await? { 222 222 + // the root commit is a Special Third Kind of block that we need to make 223 223 + // sure not to optimistically send to the processing function 224 224 + if cid == root { 225 225 + let c: Commit = serde_ipld_dagcbor::from_slice(&data)?; 226 226 + commit = Some(c); 227 227 + continue; 228 228 + } 229 229 + 230 230 + // remaining possible types: node, record, other. optimistically process 231 231 + let maybe_processed = MaybeProcessedBlock::maybe(process, data); 232 232 + 233 233 + // stash (maybe processed) blocks in memory as long as we have room 234 234 + mem_size += std::mem::size_of::<Cid>() + maybe_processed.get_size(); 235 235 + mem_blocks.insert(cid, maybe_processed); 236 236 + if mem_size >= max_size { 237 237 + return Ok(Driver::Disk(NeedDisk { 238 238 + car, 239 239 + root, 240 240 + process, 241 241 + max_size, 242 242 + mem_blocks, 243 243 + commit, 244 244 + })); 245 245 + } 132 246 } 133 247 134 134 - // remaining possible types: node, record, other. optimistically process 135 135 - // TODO: get the actual in-memory size to compute disk spill 136 136 - let maybe_processed = if Node::could_be(&data) { 137 137 - MaybeProcessedBlock::Raw(data) 138 138 - } else { 139 139 - MaybeProcessedBlock::Processed(process(data)) 140 140 - }; 248 248 + // all blocks loaded and we fit in memory! hopefully we found the commit... 249 249 + let commit = commit.ok_or(DriveError::MissingCommit)?; 250 250 + 251 251 + let walker = Walker::new(commit.data); 141 252 142 142 - // stash (maybe processed) blocks in memory as long as we have room 143 143 - mem_size += std::mem::size_of::<Cid>() + maybe_processed.get_size(); 144 144 - mem_blocks.insert(cid, maybe_processed); 145 145 - if mem_size >= max_size { 146 146 - return Ok(Vehicle::Big(BigCar { 147 147 - car, 148 148 - root, 253 253 + Ok(Driver::Memory( 254 254 + commit, 255 255 + MemDriver { 256 256 + blocks: mem_blocks, 257 257 + walker, 149 258 process, 150 150 - max_size, 151 151 - mem_blocks, 152 152 - commit, 153 153 - })); 154 154 - } 259 259 + }, 260 260 + )) 155 261 } 262 262 + } 156 263 157 157 - // all blocks loaded and we fit in memory! hopefully we found the commit... 158 158 - let commit = commit.ok_or(DriveError::MissingCommit)?; 264 264 + /// The core driver between the block stream and MST walker 265 265 + /// 266 266 + /// In the future, PDSs will export CARs in a stream-friendly order that will 267 267 + /// enable processing them with tiny memory overhead. But that future is not 268 268 + /// here yet. 269 269 + /// 270 270 + /// CARs are almost always in a stream-unfriendly order, so I'm reverting the 271 271 + /// optimistic stream features: we load all block first, then walk the MST. 272 272 + /// 273 273 + /// This makes things much simpler: we only need to worry about spilling to disk 274 274 + /// in one place, and we always have a reasonable expecatation about how much 275 275 + /// work the init function will do. We can drop the CAR reader before walking, 276 276 + /// so the sync/async boundaries become a little easier to work around. 277 277 + #[derive(Debug)] 278 278 + pub struct MemDriver<T: Processable> { 279 279 + blocks: HashMap<Cid, MaybeProcessedBlock<T>>, 280 280 + walker: Walker, 281 281 + process: fn(Vec<u8>) -> T, 282 282 + } 159 283 160 160 - let walker = Walker::new(commit.data); 284 284 + impl<T: Processable> MemDriver<T> { 285 285 + /// Step through the record outputs, in rkey order 286 286 + pub async fn next_chunk(&mut self, n: usize) -> Result<Option<BlockChunk<T>>, DriveError> { 287 287 + let mut out = Vec::with_capacity(n); 288 288 + for _ in 0..n { 289 289 + // walk as far as we can until we run out of blocks or find a record 290 290 + match self.walker.step(&mut self.blocks, self.process)? { 291 291 + Step::Missing(cid) => return Err(DriveError::MissingBlock(cid)), 292 292 + Step::Finish => break, 293 293 + Step::Found { rkey, data } => { 294 294 + out.push((rkey, data)); 295 295 + continue; 296 296 + } 297 297 + }; 298 298 + } 161 299 162 162 - Ok(Vehicle::Lil( 163 163 - commit, 164 164 - MemDriver { 165 165 - blocks: mem_blocks, 166 166 - walker, 167 167 - process, 168 168 - }, 169 169 - )) 300 300 + if out.is_empty() { 301 301 + Ok(None) 302 302 + } else { 303 303 + Ok(Some(out)) 304 304 + } 305 305 + } 170 306 } 171 307 172 172 - /// a paritally memory-loaded car file that needs disk spillover to continue 173 173 - pub struct BigCar<R: AsyncRead + Unpin, T: Processable> { 308 308 + /// A partially memory-loaded car file that needs disk spillover to continue 309 309 + pub struct NeedDisk<R: AsyncRead + Unpin, T: Processable> { 174 310 car: CarReader<R>, 175 311 root: Cid, 176 312 process: fn(Vec<u8>) -> T, ··· 183 319 bincode::serde::encode_to_vec(v, bincode::config::standard()) 184 320 } 185 321 186 186 - pub fn decode<T: Processable>(bytes: &[u8]) -> Result<T, bincode::error::DecodeError> { 322 322 + pub(crate) fn decode<T: Processable>(bytes: &[u8]) -> Result<T, DecodeError> { 187 323 let (t, n) = bincode::serde::decode_from_slice(bytes, bincode::config::standard())?; 188 188 - assert_eq!(n, bytes.len(), "expected to decode all bytes"); // TODO 324 324 + if n != bytes.len() { 325 325 + return Err(DecodeError::ExtraGarbage); 326 326 + } 189 327 Ok(t) 190 328 } 191 329 192 192 - impl<R: AsyncRead + Unpin, T: Processable + Send + 'static> BigCar<R, T> { 330 330 + impl<R: AsyncRead + Unpin, T: Processable + Send + 'static> NeedDisk<R, T> { 193 331 pub async fn finish_loading( 194 332 mut self, 195 195 - mut store: SqliteStore, 196 196 - ) -> Result<(Commit, BigCarReady<T>), DiskDriveError> { 197 197 - // set up access for real 198 198 - let mut access = store.get_access().await?; 199 199 - 200 200 - // move access in and back out so we can manage lifetimes 333 333 + mut store: DiskStore, 334 334 + ) -> Result<(Commit, DiskDriver<T>), DriveError> { 335 335 + // move store in and back out so we can manage lifetimes 201 336 // dump mem blocks into the store 202 202 - access = tokio::task::spawn(async move { 203 203 - let mut writer = access.get_writer()?; 204 204 - 337 337 + store = tokio::task::spawn(async move { 205 338 let kvs = self 206 339 .mem_blocks 207 340 .into_iter() 208 208 - .map(|(k, v)| (k.to_bytes(), encode(v).unwrap())); 341 341 + .map(|(k, v)| Ok(encode(v).map(|v| (k.to_bytes(), v))?)); 209 342 210 210 - writer.put_many(kvs)?; 211 211 - 212 212 - drop(writer); // cannot outlive access 213 213 - Ok::<_, DiskDriveError>(access) 343 343 + store.put_many(kvs)?; 344 344 + Ok::<_, DriveError>(store) 214 345 }) 215 215 - .await 216 216 - .unwrap()?; 346 346 + .await??; 217 347 218 218 - let (tx, mut rx) = tokio::sync::mpsc::channel::<Vec<(Cid, MaybeProcessedBlock<T>)>>(2); 348 348 + let (tx, mut rx) = mpsc::channel::<Vec<(Cid, MaybeProcessedBlock<T>)>>(1); 219 349 220 220 - let access_worker = tokio::task::spawn_blocking(move || { 221 221 - let mut writer = access.get_writer()?; 222 222 - 350 350 + let store_worker = tokio::task::spawn_blocking(move || { 223 351 while let Some(chunk) = rx.blocking_recv() { 224 352 let kvs = chunk 225 353 .into_iter() 226 226 - .map(|(k, v)| (k.to_bytes(), encode(v).unwrap())); 227 227 - writer.put_many(kvs)?; 354 354 + .map(|(k, v)| Ok(encode(v).map(|v| (k.to_bytes(), v))?)); 355 355 + store.put_many(kvs)?; 228 356 } 229 229 - 230 230 - drop(writer); // cannot outlive access 231 231 - Ok::<_, DiskDriveError>(access) 357 357 + Ok::<_, DriveError>(store) 232 358 }); // await later 233 359 234 360 // dump the rest to disk (in chunks) ··· 248 374 } 249 375 // remaining possible types: node, record, other. optimistically process 250 376 // TODO: get the actual in-memory size to compute disk spill 251 251 - let maybe_processed = if Node::could_be(&data) { 252 252 - MaybeProcessedBlock::Raw(data) 253 253 - } else { 254 254 - MaybeProcessedBlock::Processed((self.process)(data)) 255 255 - }; 377 377 + let maybe_processed = MaybeProcessedBlock::maybe(self.process, data); 256 378 mem_size += std::mem::size_of::<Cid>() + maybe_processed.get_size(); 257 379 chunk.push((cid, maybe_processed)); 258 380 if mem_size >= self.max_size { ··· 265 387 if chunk.is_empty() { 266 388 break; 267 389 } 268 268 - tx.send(chunk).await.unwrap(); 390 390 + tx.send(chunk) 391 391 + .await 392 392 + .map_err(|_| DriveError::ChannelSendError)?; 269 393 } 270 394 drop(tx); 271 395 log::debug!("done. waiting for worker to finish..."); 272 396 273 273 - access = access_worker.await.unwrap()?; 397 397 + store = store_worker.await??; 274 398 275 399 log::debug!("worker finished."); 276 400 277 277 - let commit = self.commit.ok_or(DiskDriveError::MissingCommit)?; 401 401 + let commit = self.commit.ok_or(DriveError::MissingCommit)?; 278 402 279 403 let walker = Walker::new(commit.data); 280 404 281 405 Ok(( 282 406 commit, 283 283 - BigCarReady { 407 407 + DiskDriver { 284 408 process: self.process, 285 285 - access, 286 286 - walker, 409 409 + state: Some(BigState { store, walker }), 287 410 }, 288 411 )) 289 412 } 290 413 } 291 414 292 292 - pub struct BigCarReady<T: Clone> { 293 293 - process: fn(Vec<u8>) -> T, 294 294 - access: SqliteAccess, 415 415 + struct BigState { 416 416 + store: DiskStore, 295 417 walker: Walker, 296 418 } 297 419 298 298 - impl<T: Processable + Send + 'static> BigCarReady<T> { 299 299 - pub async fn next_chunk( 300 300 - mut self, 301 301 - n: usize, 302 302 - ) -> Result<(Self, Option<Vec<(String, T)>>), DiskDriveError> { 303 303 - let mut out = Vec::with_capacity(n); 304 304 - (self, out) = tokio::task::spawn_blocking(move || { 305 305 - let access = self.access; 306 306 - let mut reader = access.get_reader()?; 420 420 + /// MST walker that reads from disk instead of an in-memory hashmap 421 421 + pub struct DiskDriver<T: Clone> { 422 422 + process: fn(Vec<u8>) -> T, 423 423 + state: Option<BigState>, 424 424 + } 307 425 308 308 - for _ in 0..n { 309 309 - // walk as far as we can until we run out of blocks or find a record 310 310 - match self.walker.disk_step(&mut reader, self.process)? { 311 311 - Step::Missing(cid) => return Err(DiskDriveError::MissingBlock(cid)), 312 312 - Step::Finish => break, 313 313 - Step::Step { rkey, data } => { 314 314 - out.push((rkey, data)); 315 315 - continue; 316 316 - } 317 317 - }; 318 318 - } 319 319 - 320 320 - drop(reader); // cannot outlive access 321 321 - self.access = access; 322 322 - Ok::<_, DiskDriveError>((self, out)) 323 323 - }) 324 324 - .await 325 325 - .unwrap()?; // TODO 326 326 - 327 327 - if out.is_empty() { 328 328 - Ok((self, None)) 329 329 - } else { 330 330 - Ok((self, Some(out))) 331 331 - } 426 426 + // for doctests only 427 427 + #[doc(hidden)] 428 428 + pub fn _get_fake_disk_driver() -> DiskDriver<Vec<u8>> { 429 429 + use crate::process::noop; 430 430 + DiskDriver { 431 431 + process: noop, 432 432 + state: None, 332 433 } 434 434 + } 333 435 334 334 - pub async fn rx( 335 335 - mut self, 336 336 - n: usize, 337 337 - ) -> Result< 338 338 - ( 339 339 - tokio::sync::mpsc::Receiver<Vec<(String, T)>>, 340 340 - tokio::task::JoinHandle<Result<(), DiskDriveError>>, 341 341 - ), 342 342 - DiskDriveError, 343 343 - > { 344 344 - let (tx, rx) = tokio::sync::mpsc::channel::<Vec<(String, T)>>(1); 436 436 + impl<T: Processable + Send + 'static> DiskDriver<T> { 437 437 + /// Walk the MST returning up to `n` rkey + record pairs 438 438 + /// 439 439 + /// ```no_run 440 440 + /// # use repo_stream::{drive::{DiskDriver, DriveError, _get_fake_disk_driver}, process::noop}; 441 441 + /// # #[tokio::main] 442 442 + /// # async fn main() -> Result<(), DriveError> { 443 443 + /// # let mut disk_driver = _get_fake_disk_driver(); 444 444 + /// while let Some(pairs) = disk_driver.next_chunk(256).await? { 445 445 + /// for (rkey, record) in pairs { 446 446 + /// println!("{rkey}: size={}", record.len()); 447 447 + /// } 448 448 + /// } 449 449 + /// # Ok(()) 450 450 + /// # } 451 451 + /// ``` 452 452 + pub async fn next_chunk(&mut self, n: usize) -> Result<Option<BlockChunk<T>>, DriveError> { 453 453 + let process = self.process; 345 454 346 346 - // sketch: this worker is going to be allowed to execute without a join handle 347 347 - // ...should we return the join handle here so the caller at least knows about it? 348 348 - // yes probably for error handling?? (orrr put errors in the channel) 349 349 - let worker = tokio::task::spawn_blocking(move || { 350 350 - let mut reader = self.access.get_reader()?; 455 455 + // state should only *ever* be None transiently while inside here 456 456 + let mut state = self.state.take().expect("DiskDriver must have Some(state)"); 351 457 352 352 - loop { 458 458 + // the big pain here is that we don't want to leave self.state in an 459 459 + // invalid state (None), so all the error paths have to make sure it 460 460 + // comes out again. 461 461 + let (state, res) = tokio::task::spawn_blocking( 462 462 + move || -> (BigState, Result<BlockChunk<T>, DriveError>) { 353 463 let mut out = Vec::with_capacity(n); 354 464 355 465 for _ in 0..n { 356 466 // walk as far as we can until we run out of blocks or find a record 357 357 - match self.walker.disk_step(&mut reader, self.process)? { 358 358 - Step::Missing(cid) => return Err(DiskDriveError::MissingBlock(cid)), 359 359 - Step::Finish => break, 360 360 - Step::Step { rkey, data } => { 361 361 - out.push((rkey, data)); 362 362 - continue; 467 467 + let step = match state.walker.disk_step(&mut state.store, process) { 468 468 + Ok(s) => s, 469 469 + Err(e) => { 470 470 + return (state, Err(e.into())); 363 471 } 364 472 }; 473 473 + match step { 474 474 + Step::Missing(cid) => { 475 475 + return (state, Err(DriveError::MissingBlock(cid))); 476 476 + } 477 477 + Step::Finish => break, 478 478 + Step::Found { rkey, data } => out.push((rkey, data)), 479 479 + }; 365 480 } 366 481 367 367 - if out.is_empty() { 368 368 - break; 369 369 - } 370 370 - tx.blocking_send(out).unwrap(); 371 371 - } 482 482 + (state, Ok::<_, DriveError>(out)) 483 483 + }, 484 484 + ) 485 485 + .await?; // on tokio JoinError, we'll be left with invalid state :( 372 486 373 373 - drop(reader); // cannot outlive access 374 374 - Ok(()) 375 375 - }); // await later 376 376 - 377 377 - Ok((rx, worker)) 378 378 - } 379 379 - } 380 380 - 381 381 - /// The core driver between the block stream and MST walker 382 382 - /// 383 383 - /// In the future, PDSs will export CARs in a stream-friendly order that will 384 384 - /// enable processing them with tiny memory overhead. But that future is not 385 385 - /// here yet. 386 386 - /// 387 387 - /// CARs are almost always in a stream-unfriendly order, so I'm reverting the 388 388 - /// optimistic stream features: we load all block first, then walk the MST. 389 389 - /// 390 390 - /// This makes things much simpler: we only need to worry about spilling to disk 391 391 - /// in one place, and we always have a reasonable expecatation about how much 392 392 - /// work the init function will do. We can drop the CAR reader before walking, 393 393 - /// so the sync/async boundaries become a little easier to work around. 394 394 - #[derive(Debug)] 395 395 - pub struct MemDriver<T: Processable> { 396 396 - blocks: HashMap<Cid, MaybeProcessedBlock<T>>, 397 397 - walker: Walker, 398 398 - process: fn(Vec<u8>) -> T, 399 399 - } 487 487 + // *must* restore state before dealing with the actual result 488 488 + self.state = Some(state); 400 489 401 401 - impl<T: Processable> MemDriver<T> { 402 402 - /// Manually step through the record outputs 403 403 - pub async fn next_chunk(&mut self, n: usize) -> Result<Option<Vec<(String, T)>>, DriveError> { 404 404 - let mut out = Vec::with_capacity(n); 405 405 - for _ in 0..n { 406 406 - // walk as far as we can until we run out of blocks or find a record 407 407 - match self.walker.step(&mut self.blocks, self.process)? { 408 408 - Step::Missing(cid) => return Err(DriveError::MissingBlock(cid)), 409 409 - Step::Finish => break, 410 410 - Step::Step { rkey, data } => { 411 411 - out.push((rkey, data)); 412 412 - continue; 413 413 - } 414 414 - }; 415 415 - } 490 490 + let out = res?; 416 491 417 492 if out.is_empty() { 418 493 Ok(None) 419 494 } else { 420 495 Ok(Some(out)) 421 496 } 497 497 + } 498 498 + 499 499 + fn read_tx_blocking( 500 500 + &mut self, 501 501 + n: usize, 502 502 + tx: mpsc::Sender<Result<BlockChunk<T>, DriveError>>, 503 503 + ) -> Result<(), mpsc::error::SendError<Result<BlockChunk<T>, DriveError>>> { 504 504 + let BigState { store, walker } = self.state.as_mut().expect("valid state"); 505 505 + 506 506 + loop { 507 507 + let mut out: BlockChunk<T> = Vec::with_capacity(n); 508 508 + 509 509 + for _ in 0..n { 510 510 + // walk as far as we can until we run out of blocks or find a record 511 511 + 512 512 + let step = match walker.disk_step(store, self.process) { 513 513 + Ok(s) => s, 514 514 + Err(e) => return tx.blocking_send(Err(e.into())), 515 515 + }; 516 516 + 517 517 + match step { 518 518 + Step::Missing(cid) => { 519 519 + return tx.blocking_send(Err(DriveError::MissingBlock(cid))); 520 520 + } 521 521 + Step::Finish => return Ok(()), 522 522 + Step::Found { rkey, data } => { 523 523 + out.push((rkey, data)); 524 524 + continue; 525 525 + } 526 526 + }; 527 527 + } 528 528 + 529 529 + if out.is_empty() { 530 530 + break; 531 531 + } 532 532 + tx.blocking_send(Ok(out))?; 533 533 + } 534 534 + 535 535 + Ok(()) 536 536 + } 537 537 + 538 538 + /// Spawn the disk reading task into a tokio blocking thread 539 539 + /// 540 540 + /// The idea is to avoid so much sending back and forth to the blocking 541 541 + /// thread, letting a blocking task do all the disk reading work and sending 542 542 + /// records and rkeys back through an `mpsc` channel instead. 543 543 + /// 544 544 + /// This might also allow the disk work to continue while processing the 545 545 + /// records. It's still not yet clear if this method actually has much 546 546 + /// benefit over just using `.next_chunk(n)`. 547 547 + /// 548 548 + /// ```no_run 549 549 + /// # use repo_stream::{drive::{DiskDriver, DriveError, _get_fake_disk_driver}, process::noop}; 550 550 + /// # #[tokio::main] 551 551 + /// # async fn main() -> Result<(), DriveError> { 552 552 + /// # let mut disk_driver = _get_fake_disk_driver(); 553 553 + /// let (mut rx, join) = disk_driver.to_channel(512); 554 554 + /// while let Some(recvd) = rx.recv().await { 555 555 + /// let pairs = recvd?; 556 556 + /// for (rkey, record) in pairs { 557 557 + /// println!("{rkey}: size={}", record.len()); 558 558 + /// } 559 559 + /// 560 560 + /// } 561 561 + /// # Ok(()) 562 562 + /// # } 563 563 + /// ``` 564 564 + pub fn to_channel( 565 565 + mut self, 566 566 + n: usize, 567 567 + ) -> ( 568 568 + mpsc::Receiver<Result<BlockChunk<T>, DriveError>>, 569 569 + tokio::task::JoinHandle<Self>, 570 570 + ) { 571 571 + let (tx, rx) = mpsc::channel::<Result<BlockChunk<T>, DriveError>>(1); 572 572 + 573 573 + // sketch: this worker is going to be allowed to execute without a join handle 574 574 + let chan_task = tokio::task::spawn_blocking(move || { 575 575 + if let Err(mpsc::error::SendError(_)) = self.read_tx_blocking(n, tx) { 576 576 + log::debug!("big car reader exited early due to dropped receiver channel"); 577 577 + } 578 578 + self 579 579 + }); 580 580 + 581 581 + (rx, chan_task) 422 582 } 423 583 }

+81 -5

src/lib.rs

··· 1 1 - //! Fast and robust atproto CAR file processing in rust 2 2 - //! 3 3 - //! For now see the [examples](https://tangled.org/@microcosm.blue/repo-stream/tree/main/examples) 1 1 + /*! 2 2 + A robust CAR file -> MST walker for atproto 3 3 + 4 4 + Small CARs have their blocks buffered in memory. If a configurable memory limit 5 5 + is reached while reading blocks, CAR reading is suspended, and can be continued 6 6 + by providing disk storage to buffer the CAR blocks instead. 7 7 + 8 8 + A `process` function can be provided for tasks where records are transformed 9 9 + into a smaller representation, to save memory (and disk) during block reading. 10 10 + 11 11 + Once blocks are loaded, the MST is walked and emitted as chunks of pairs of 12 12 + `(rkey, processed_block)` pairs, in order (depth first, left-to-right). 13 13 + 14 14 + Some MST validations are applied 15 15 + - Keys must appear in order 16 16 + - Keys must be at the correct MST tree depth 17 17 + 18 18 + `iroh_car` additionally applies a block size limit of `2MiB`. 19 19 + 20 20 + ``` 21 21 + use repo_stream::{Driver, DriverBuilder, DiskBuilder}; 22 22 + 23 23 + # #[tokio::main] 24 24 + # async fn main() -> Result<(), Box<dyn std::error::Error>> { 25 25 + # let reader = include_bytes!("../car-samples/tiny.car").as_slice(); 26 26 + let mut total_size = 0; 27 27 + 28 28 + match DriverBuilder::new() 29 29 + .with_mem_limit_mb(10) 30 30 + .with_block_processor(|rec| rec.len()) // block processing: just extract the raw record size 31 31 + .load_car(reader) 32 32 + .await? 33 33 + { 34 34 + 35 35 + // if all blocks fit within memory 36 36 + Driver::Memory(_commit, mut driver) => { 37 37 + while let Some(chunk) = driver.next_chunk(256).await? { 38 38 + for (_rkey, size) in chunk { 39 39 + total_size += size; 40 40 + } 41 41 + } 42 42 + }, 43 43 + 44 44 + // if the CAR was too big for in-memory processing 45 45 + Driver::Disk(paused) => { 46 46 + // set up a disk store we can spill to 47 47 + let store = DiskBuilder::new().open("some/path.db".into()).await?; 48 48 + // do the spilling, get back a (similar) driver 49 49 + let (_commit, mut driver) = paused.finish_loading(store).await?; 50 50 + 51 51 + while let Some(chunk) = driver.next_chunk(256).await? { 52 52 + for (_rkey, size) in chunk { 53 53 + total_size += size; 54 54 + } 55 55 + } 56 56 + } 57 57 + }; 58 58 + println!("sum of size of all records: {total_size}"); 59 59 + # Ok(()) 60 60 + # } 61 61 + ``` 62 62 + 63 63 + Disk spilling suspends and returns a `Driver::Disk(paused)` instead of going 64 64 + ahead and eagerly using disk I/O. This means you have to write a bit more code 65 65 + to handle both cases, but it allows you to have finer control over resource 66 66 + usage. For example, you can drive a number of parallel memory CAR workers, and 67 67 + separately have a different number of disk workers picking up suspended disk 68 68 + tasks from a queue. 69 69 + 70 70 + Find more [examples in the repo](https://tangled.org/@microcosm.blue/repo-stream/tree/main/examples). 71 71 + 72 72 + */ 73 73 + 74 74 + pub mod mst; 75 75 + mod walk; 4 76 5 77 pub mod disk; 6 78 pub mod drive; 7 7 - pub mod mst; 8 8 - pub mod walk; 79 79 + pub mod process; 80 80 + 81 81 + pub use disk::{DiskBuilder, DiskError, DiskStore}; 82 82 + pub use drive::{DriveError, Driver, DriverBuilder, NeedDisk}; 83 83 + pub use mst::Commit; 84 84 + pub use process::Processable;

+4 -4

src/mst.rs

··· 39 39 /// MST node data schema 40 40 #[derive(Debug, Deserialize, PartialEq)] 41 41 #[serde(deny_unknown_fields)] 42 42 - pub struct Node { 42 42 + pub(crate) struct Node { 43 43 /// link to sub-tree Node on a lower level and with all keys sorting before 44 44 /// keys at this node 45 45 #[serde(rename = "l")] ··· 62 62 /// so if a block *could be* a node, any record converter must postpone 63 63 /// processing. if it turns out it happens to be a very node-looking record, 64 64 /// well, sorry, it just has to only be processed later when that's known. 65 65 - pub fn could_be(bytes: impl AsRef<[u8]>) -> bool { 65 65 + pub(crate) fn could_be(bytes: impl AsRef<[u8]>) -> bool { 66 66 const NODE_FINGERPRINT: [u8; 3] = [ 67 67 0xA2, // map length 2 (for "l" and "e" keys) 68 68 0x61, // text length 1 ··· 83 83 /// with an empty array of entries. This is the only situation in which a 84 84 /// tree may contain an empty leaf node which does not either contain keys 85 85 /// ("entries") or point to a sub-tree containing entries. 86 86 - pub fn is_empty(&self) -> bool { 86 86 + pub(crate) fn is_empty(&self) -> bool { 87 87 self.left.is_none() && self.entries.is_empty() 88 88 } 89 89 } ··· 91 91 /// TreeEntry object 92 92 #[derive(Debug, Deserialize, PartialEq)] 93 93 #[serde(deny_unknown_fields)] 94 94 - pub struct Entry { 94 94 + pub(crate) struct Entry { 95 95 /// count of bytes shared with previous TreeEntry in this Node (if any) 96 96 #[serde(rename = "p")] 97 97 pub prefix_len: usize,

+108

src/process.rs

··· 1 1 + /*! 2 2 + Record processor function output trait 3 3 + 4 4 + The return type must satisfy the `Processable` trait, which requires: 5 5 + 6 6 + - `Clone` because two rkeys can refer to the same record by CID, which may 7 7 + only appear once in the CAR file. 8 8 + - `Serialize + DeserializeOwned` so it can be spilled to disk. 9 9 + 10 10 + One required function must be implemented, `get_size()`: this should return the 11 11 + approximate total off-stack size of the type. (the on-stack size will be added 12 12 + automatically via `std::mem::get_size`). 13 13 + 14 14 + Note that it is **not guaranteed** that the `process` function will run on a 15 15 + block before storing it in memory or on disk: it's not possible to know if a 16 16 + block is a record without actually walking the MST, so the best we can do is 17 17 + apply `process` to any block that we know *cannot* be an MST node, and otherwise 18 18 + store the raw block bytes. 19 19 + 20 20 + Here's a silly processing function that just collects 'eyy's found in the raw 21 21 + record bytes 22 22 + 23 23 + ``` 24 24 + # use repo_stream::Processable; 25 25 + # use serde::{Serialize, Deserialize}; 26 26 + #[derive(Debug, Clone, Serialize, Deserialize)] 27 27 + struct Eyy(usize, String); 28 28 + 29 29 + impl Processable for Eyy { 30 30 + fn get_size(&self) -> usize { 31 31 + // don't need to compute the usize, it's on the stack 32 32 + self.1.capacity() // in-mem size from the string's capacity, in bytes 33 33 + } 34 34 + } 35 35 + 36 36 + fn process(raw: Vec<u8>) -> Vec<Eyy> { 37 37 + let mut out = Vec::new(); 38 38 + let to_find = "eyy".as_bytes(); 39 39 + for i in 0..(raw.len() - 3) { 40 40 + if &raw[i..(i+3)] == to_find { 41 41 + out.push(Eyy(i, "eyy".to_string())); 42 42 + } 43 43 + } 44 44 + out 45 45 + } 46 46 + ``` 47 47 + 48 48 + The memory sizing stuff is a little sketch but probably at least approximately 49 49 + works. 50 50 + */ 51 51 + 52 52 + use serde::{Serialize, de::DeserializeOwned}; 53 53 + 54 54 + /// Output trait for record processing 55 55 + pub trait Processable: Clone + Serialize + DeserializeOwned { 56 56 + /// Any additional in-memory size taken by the processed type 57 57 + /// 58 58 + /// Do not include stack size (`std::mem::size_of`) 59 59 + fn get_size(&self) -> usize; 60 60 + } 61 61 + 62 62 + /// Processor that just returns the raw blocks 63 63 + #[inline] 64 64 + pub fn noop(block: Vec<u8>) -> Vec<u8> { 65 65 + block 66 66 + } 67 67 + 68 68 + impl Processable for u8 { 69 69 + fn get_size(&self) -> usize { 70 70 + 0 71 71 + } 72 72 + } 73 73 + 74 74 + impl Processable for usize { 75 75 + fn get_size(&self) -> usize { 76 76 + 0 // no additional space taken, just its stack size (newtype is free) 77 77 + } 78 78 + } 79 79 + 80 80 + impl Processable for String { 81 81 + fn get_size(&self) -> usize { 82 82 + self.capacity() 83 83 + } 84 84 + } 85 85 + 86 86 + impl<Item: Sized + Processable> Processable for Vec<Item> { 87 87 + fn get_size(&self) -> usize { 88 88 + let slot_size = std::mem::size_of::<Item>(); 89 89 + let direct_size = slot_size * self.capacity(); 90 90 + let items_referenced_size: usize = self.iter().map(|item| item.get_size()).sum(); 91 91 + direct_size + items_referenced_size 92 92 + } 93 93 + } 94 94 + 95 95 + impl<Item: Processable> Processable for Option<Item> { 96 96 + fn get_size(&self) -> usize { 97 97 + self.as_ref().map(|item| item.get_size()).unwrap_or(0) 98 98 + } 99 99 + } 100 100 + 101 101 + impl<Item: Processable, Error: Processable> Processable for Result<Item, Error> { 102 102 + fn get_size(&self) -> usize { 103 103 + match self { 104 104 + Ok(item) => item.get_size(), 105 105 + Err(err) => err.get_size(), 106 106 + } 107 107 + } 108 108 + }

+33 -239

src/walk.rs

··· 1 1 //! Depth-first MST traversal 2 2 3 3 - use crate::disk::SqliteReader; 4 4 - use crate::drive::{MaybeProcessedBlock, Processable}; 3 3 + use crate::disk::DiskStore; 4 4 + use crate::drive::{DecodeError, MaybeProcessedBlock}; 5 5 use crate::mst::Node; 6 6 + use crate::process::Processable; 6 7 use ipld_core::cid::Cid; 7 8 use sha2::{Digest, Sha256}; 8 9 use std::collections::HashMap; ··· 10 11 11 12 /// Errors that can happen while walking 12 13 #[derive(Debug, thiserror::Error)] 13 13 - pub enum Trip { 14 14 - #[error("empty mst nodes are not allowed")] 15 15 - NodeEmpty, 14 14 + pub enum WalkError { 16 15 #[error("Failed to fingerprint commit block")] 17 16 BadCommitFingerprint, 18 17 #[error("Failed to decode commit block: {0}")] 19 18 BadCommit(#[from] serde_ipld_dagcbor::DecodeError<Infallible>), 20 19 #[error("Action node error: {0}")] 21 20 MstError(#[from] MstError), 22 22 - #[error("Encountered an rkey out of order while walking the MST")] 23 23 - RkeyOutOfOrder, 24 24 - } 25 25 - 26 26 - /// Errors that can happen while walking 27 27 - #[derive(Debug, thiserror::Error)] 28 28 - pub enum DiskTrip { 29 29 - #[error("tripped: {0}")] 30 30 - Trip(#[from] Trip), 31 21 #[error("storage error: {0}")] 32 32 - StorageError(#[from] rusqlite::Error), 22 22 + StorageError(#[from] fjall::Error), 33 23 #[error("Decode error: {0}")] 34 34 - BincodeDecodeError(#[from] bincode::error::DecodeError), 24 24 + DecodeError(#[from] DecodeError), 35 25 } 36 26 37 27 /// Errors from invalid Rkeys ··· 49 39 LostDepth, 50 40 #[error("MST depth underflow: depth-0 node with child trees")] 51 41 DepthUnderflow, 42 42 + #[error("Encountered an rkey out of order while walking the MST")] 43 43 + RkeyOutOfOrder, 52 44 } 53 45 54 46 /// Walker outputs ··· 59 51 /// Reached the end of the MST! yay! 60 52 Finish, 61 53 /// A record was found! 62 62 - Step { rkey: String, data: T }, 54 54 + Found { rkey: String, data: T }, 63 55 } 64 56 65 57 #[derive(Debug, Clone, PartialEq)] ··· 95 87 } 96 88 97 89 fn push_from_node(stack: &mut Vec<Need>, node: &Node, parent_depth: Depth) -> Result<(), MstError> { 98 98 - // empty nodes are not allowed in the MST 99 99 - // ...except for a single one for empty MST, but we wouldn't be pushing that 90 90 + // empty nodes are not allowed in the MST except in an empty MST 100 91 if node.is_empty() { 101 101 - return Err(MstError::EmptyNode); 92 92 + if parent_depth == Depth::Root { 93 93 + return Ok(()); // empty mst, nothing to push 94 94 + } else { 95 95 + return Err(MstError::EmptyNode); 96 96 + } 102 97 } 103 98 104 99 let mut entries = Vec::with_capacity(node.entries.len()); ··· 185 180 &mut self, 186 181 blocks: &mut HashMap<Cid, MaybeProcessedBlock<T>>, 187 182 process: impl Fn(Vec<u8>) -> T, 188 188 - ) -> Result<Step<T>, Trip> { 183 183 + ) -> Result<Step<T>, WalkError> { 189 184 loop { 190 185 let Some(need) = self.stack.last_mut() else { 191 186 log::trace!("tried to walk but we're actually done."); ··· 201 196 }; 202 197 203 198 let MaybeProcessedBlock::Raw(data) = block else { 204 204 - return Err(Trip::BadCommitFingerprint); 199 199 + return Err(WalkError::BadCommitFingerprint); 205 200 }; 206 206 - let node = 207 207 - serde_ipld_dagcbor::from_slice::<Node>(&data).map_err(Trip::BadCommit)?; 201 201 + let node = serde_ipld_dagcbor::from_slice::<Node>(&data) 202 202 + .map_err(WalkError::BadCommit)?; 208 203 209 204 // found node, make sure we remember 210 205 self.stack.pop(); ··· 214 209 } 215 210 Need::Record { rkey, cid } => { 216 211 log::trace!("need record {cid:?}"); 212 212 + // note that we cannot *remove* a record block, sadly, since 213 213 + // there can be multiple rkeys pointing to the same cid. 217 214 let Some(data) = blocks.get_mut(cid) else { 218 218 - log::trace!("record block not found, resting"); 219 215 return Ok(Step::Missing(*cid)); 220 216 }; 221 217 let rkey = rkey.clone(); ··· 227 223 // found node, make sure we remember 228 224 self.stack.pop(); 229 225 230 230 - log::trace!("emitting a block as a step. depth={}", self.stack.len()); 231 231 - 232 226 // rkeys *must* be in order or else the tree is invalid (or 233 227 // we have a bug) 234 228 if rkey <= self.prev { 235 235 - return Err(Trip::RkeyOutOfOrder); 229 229 + return Err(MstError::RkeyOutOfOrder)?; 236 230 } 237 231 self.prev = rkey.clone(); 238 232 239 239 - return Ok(Step::Step { rkey, data }); 233 233 + return Ok(Step::Found { rkey, data }); 240 234 } 241 235 } 242 236 } ··· 245 239 /// blocking!!!!!! 246 240 pub fn disk_step<T: Processable>( 247 241 &mut self, 248 248 - reader: &mut SqliteReader, 242 242 + reader: &mut DiskStore, 249 243 process: impl Fn(Vec<u8>) -> T, 250 250 - ) -> Result<Step<T>, DiskTrip> { 244 244 + ) -> Result<Step<T>, WalkError> { 251 245 loop { 252 246 let Some(need) = self.stack.last_mut() else { 253 247 log::trace!("tried to walk but we're actually done."); ··· 258 252 &mut Need::Node { depth, cid } => { 259 253 let cid_bytes = cid.to_bytes(); 260 254 log::trace!("need node {cid:?}"); 261 261 - let Some(block_bytes) = reader.get(cid_bytes)? else { 255 255 + let Some(block_bytes) = reader.get(&cid_bytes)? else { 262 256 log::trace!("node not found, resting"); 263 257 return Ok(Step::Missing(cid)); 264 258 }; ··· 266 260 let block: MaybeProcessedBlock<T> = crate::drive::decode(&block_bytes)?; 267 261 268 262 let MaybeProcessedBlock::Raw(data) = block else { 269 269 - return Err(Trip::BadCommitFingerprint.into()); 263 263 + return Err(WalkError::BadCommitFingerprint); 270 264 }; 271 271 - let node = 272 272 - serde_ipld_dagcbor::from_slice::<Node>(&data).map_err(Trip::BadCommit)?; 265 265 + let node = serde_ipld_dagcbor::from_slice::<Node>(&data) 266 266 + .map_err(WalkError::BadCommit)?; 273 267 274 268 // found node, make sure we remember 275 269 self.stack.pop(); 276 270 277 271 // queue up work on the found node next 278 278 - push_from_node(&mut self.stack, &node, depth).map_err(Trip::MstError)?; 272 272 + push_from_node(&mut self.stack, &node, depth).map_err(WalkError::MstError)?; 279 273 } 280 274 Need::Record { rkey, cid } => { 281 275 log::trace!("need record {cid:?}"); 282 276 let cid_bytes = cid.to_bytes(); 283 283 - let Some(data_bytes) = reader.get(cid_bytes)? else { 277 277 + let Some(data_bytes) = reader.get(&cid_bytes)? else { 284 278 log::trace!("record block not found, resting"); 285 279 return Ok(Step::Missing(*cid)); 286 280 }; ··· 299 293 // rkeys *must* be in order or else the tree is invalid (or 300 294 // we have a bug) 301 295 if rkey <= self.prev { 302 302 - return Err(DiskTrip::Trip(Trip::RkeyOutOfOrder)); 296 296 + return Err(MstError::RkeyOutOfOrder)?; 303 297 } 304 298 self.prev = rkey.clone(); 305 299 306 306 - return Ok(Step::Step { rkey, data }); 300 300 + return Ok(Step::Found { rkey, data }); 307 301 } 308 302 } 309 303 } ··· 313 307 #[cfg(test)] 314 308 mod test { 315 309 use super::*; 316 316 - // use crate::mst::Entry; 317 310 318 311 fn cid1() -> Cid { 319 312 "bafyreihixenvk3ahqbytas4hk4a26w43bh6eo3w6usjqtxkpzsvi655a3m" 320 313 .parse() 321 314 .unwrap() 322 315 } 323 323 - // fn cid2() -> Cid { 324 324 - // "QmY7Yh4UquoXHLPFo2XbhXkhBvFoPwmQUSa92pxnxjQuPU" 325 325 - // .parse() 326 326 - // .unwrap() 327 327 - // } 328 328 - // fn cid3() -> Cid { 329 329 - // "bafybeigdyrzt5sfp7udm7hu76uh7y26nf3efuylqabf3oclgtqy55fbzdi" 330 330 - // .parse() 331 331 - // .unwrap() 332 332 - // } 333 333 - // fn cid4() -> Cid { 334 334 - // "QmbWqxBEKC3P8tqsKc98xmWNzrzDtRLMiMPL8wBuTGsMnR" 335 335 - // .parse() 336 336 - // .unwrap() 337 337 - // } 338 338 - // fn cid5() -> Cid { 339 339 - // "QmSnuWmxptJZdLJpKRarxBMS2Ju2oANVrgbr2xWbie9b2D" 340 340 - // .parse() 341 341 - // .unwrap() 342 342 - // } 343 343 - // fn cid6() -> Cid { 344 344 - // "QmdmQXB2mzChmMeKY47C43LxUdg1NDJ5MWcKMKxDu7RgQm" 345 345 - // .parse() 346 346 - // .unwrap() 347 347 - // } 348 348 - // fn cid7() -> Cid { 349 349 - // "bafybeiaysi4s6lnjev27ln5icwm6tueaw2vdykrtjkwiphwekaywqhcjze" 350 350 - // .parse() 351 351 - // .unwrap() 352 352 - // } 353 353 - // fn cid8() -> Cid { 354 354 - // "bafyreif3tfdpr5n4jdrbielmcapwvbpcthepfkwq2vwonmlhirbjmotedi" 355 355 - // .parse() 356 356 - // .unwrap() 357 357 - // } 358 358 - // fn cid9() -> Cid { 359 359 - // "bafyreicnokmhmrnlp2wjhyk2haep4tqxiptwfrp2rrs7rzq7uk766chqvq" 360 360 - // .parse() 361 361 - // .unwrap() 362 362 - // } 363 316 364 317 #[test] 365 318 fn test_depth_spec_0() { ··· 450 403 .as_ref() 451 404 ); 452 405 } 453 453 - 454 454 - // #[test] 455 455 - // fn test_needs_from_node_just_one_record() { 456 456 - // let node = Node { 457 457 - // left: None, 458 458 - // entries: vec![Entry { 459 459 - // keysuffix: "asdf".into(), 460 460 - // prefix_len: 0, 461 461 - // value: cid1(), 462 462 - // tree: None, 463 463 - // }], 464 464 - // }; 465 465 - // assert_eq!( 466 466 - // needs_from_node(node).unwrap(), 467 467 - // vec![Need::Record { 468 468 - // rkey: "asdf".into(), 469 469 - // cid: cid1(), 470 470 - // },] 471 471 - // ); 472 472 - // } 473 473 - 474 474 - // #[test] 475 475 - // fn test_needs_from_node_two_records() { 476 476 - // let node = Node { 477 477 - // left: None, 478 478 - // entries: vec![ 479 479 - // Entry { 480 480 - // keysuffix: "asdf".into(), 481 481 - // prefix_len: 0, 482 482 - // value: cid1(), 483 483 - // tree: None, 484 484 - // }, 485 485 - // Entry { 486 486 - // keysuffix: "gh".into(), 487 487 - // prefix_len: 2, 488 488 - // value: cid2(), 489 489 - // tree: None, 490 490 - // }, 491 491 - // ], 492 492 - // }; 493 493 - // assert_eq!( 494 494 - // needs_from_node(node).unwrap(), 495 495 - // vec![ 496 496 - // Need::Record { 497 497 - // rkey: "asdf".into(), 498 498 - // cid: cid1(), 499 499 - // }, 500 500 - // Need::Record { 501 501 - // rkey: "asgh".into(), 502 502 - // cid: cid2(), 503 503 - // }, 504 504 - // ] 505 505 - // ); 506 506 - // } 507 507 - 508 508 - // #[test] 509 509 - // fn test_needs_from_node_with_both() { 510 510 - // let node = Node { 511 511 - // left: None, 512 512 - // entries: vec![Entry { 513 513 - // keysuffix: "asdf".into(), 514 514 - // prefix_len: 0, 515 515 - // value: cid1(), 516 516 - // tree: Some(cid2()), 517 517 - // }], 518 518 - // }; 519 519 - // assert_eq!( 520 520 - // needs_from_node(node).unwrap(), 521 521 - // vec![ 522 522 - // Need::Record { 523 523 - // rkey: "asdf".into(), 524 524 - // cid: cid1(), 525 525 - // }, 526 526 - // Need::Node(cid2()), 527 527 - // ] 528 528 - // ); 529 529 - // } 530 530 - 531 531 - // #[test] 532 532 - // fn test_needs_from_node_left_and_record() { 533 533 - // let node = Node { 534 534 - // left: Some(cid1()), 535 535 - // entries: vec![Entry { 536 536 - // keysuffix: "asdf".into(), 537 537 - // prefix_len: 0, 538 538 - // value: cid2(), 539 539 - // tree: None, 540 540 - // }], 541 541 - // }; 542 542 - // assert_eq!( 543 543 - // needs_from_node(node).unwrap(), 544 544 - // vec![ 545 545 - // Need::Node(cid1()), 546 546 - // Need::Record { 547 547 - // rkey: "asdf".into(), 548 548 - // cid: cid2(), 549 549 - // }, 550 550 - // ] 551 551 - // ); 552 552 - // } 553 553 - 554 554 - // #[test] 555 555 - // fn test_needs_from_full_node() { 556 556 - // let node = Node { 557 557 - // left: Some(cid1()), 558 558 - // entries: vec![ 559 559 - // Entry { 560 560 - // keysuffix: "asdf".into(), 561 561 - // prefix_len: 0, 562 562 - // value: cid2(), 563 563 - // tree: Some(cid3()), 564 564 - // }, 565 565 - // Entry { 566 566 - // keysuffix: "ghi".into(), 567 567 - // prefix_len: 1, 568 568 - // value: cid4(), 569 569 - // tree: Some(cid5()), 570 570 - // }, 571 571 - // Entry { 572 572 - // keysuffix: "jkl".into(), 573 573 - // prefix_len: 2, 574 574 - // value: cid6(), 575 575 - // tree: Some(cid7()), 576 576 - // }, 577 577 - // Entry { 578 578 - // keysuffix: "mno".into(), 579 579 - // prefix_len: 4, 580 580 - // value: cid8(), 581 581 - // tree: Some(cid9()), 582 582 - // }, 583 583 - // ], 584 584 - // }; 585 585 - // assert_eq!( 586 586 - // needs_from_node(node).unwrap(), 587 587 - // vec![ 588 588 - // Need::Node(cid1()), 589 589 - // Need::Record { 590 590 - // rkey: "asdf".into(), 591 591 - // cid: cid2(), 592 592 - // }, 593 593 - // Need::Node(cid3()), 594 594 - // Need::Record { 595 595 - // rkey: "aghi".into(), 596 596 - // cid: cid4(), 597 597 - // }, 598 598 - // Need::Node(cid5()), 599 599 - // Need::Record { 600 600 - // rkey: "agjkl".into(), 601 601 - // cid: cid6(), 602 602 - // }, 603 603 - // Need::Node(cid7()), 604 604 - // Need::Record { 605 605 - // rkey: "agjkmno".into(), 606 606 - // cid: cid8(), 607 607 - // }, 608 608 - // Need::Node(cid9()), 609 609 - // ] 610 610 - // ); 611 611 - // } 612 406 }

+21 -22

tests/non-huge-cars.rs

··· 1 1 extern crate repo_stream; 2 2 - use repo_stream::drive::Processable; 3 3 - use serde::{Deserialize, Serialize}; 2 2 + use repo_stream::Driver; 4 3 4 4 + const EMPTY_CAR: &'static [u8] = include_bytes!("../car-samples/empty.car"); 5 5 const TINY_CAR: &'static [u8] = include_bytes!("../car-samples/tiny.car"); 6 6 const LITTLE_CAR: &'static [u8] = include_bytes!("../car-samples/little.car"); 7 7 const MIDSIZE_CAR: &'static [u8] = include_bytes!("../car-samples/midsize.car"); 8 8 9 9 - #[derive(Clone, Serialize, Deserialize)] 10 10 - struct S(usize); 11 11 - 12 12 - impl Processable for S { 13 13 - fn get_size(&self) -> usize { 14 14 - 0 // no additional space taken, just its stack size (newtype is free) 15 15 - } 16 16 - } 17 17 - 18 18 - async fn test_car(bytes: &[u8], expected_records: usize, expected_sum: usize) { 19 19 - let mb = 2_usize.pow(20); 20 20 - 21 21 - let mut driver = match repo_stream::drive::load_car(bytes, |block| S(block.len()), 10 * mb) 9 9 + async fn test_car( 10 10 + bytes: &[u8], 11 11 + expected_records: usize, 12 12 + expected_sum: usize, 13 13 + expect_profile: bool, 14 14 + ) { 15 15 + let mut driver = match Driver::load_car(bytes, |block| block.len(), 10 /* MiB */) 22 16 .await 23 17 .unwrap() 24 18 { 25 25 - repo_stream::drive::Vehicle::Lil(_commit, mem_driver) => mem_driver, 26 26 - repo_stream::drive::Vehicle::Big(_) => panic!("too big"), 19 19 + Driver::Memory(_commit, mem_driver) => mem_driver, 20 20 + Driver::Disk(_) => panic!("too big"), 27 21 }; 28 22 29 23 let mut records = 0; ··· 32 26 let mut prev_rkey = "".to_string(); 33 27 34 28 while let Some(pairs) = driver.next_chunk(256).await.unwrap() { 35 35 - for (rkey, S(size)) in pairs { 29 29 + for (rkey, size) in pairs { 36 30 records += 1; 37 31 sum += size; 38 32 if rkey == "app.bsky.actor.profile/self" { ··· 45 39 46 40 assert_eq!(records, expected_records); 47 41 assert_eq!(sum, expected_sum); 48 48 - assert!(found_bsky_profile); 42 42 + assert_eq!(found_bsky_profile, expect_profile); 43 43 + } 44 44 + 45 45 + #[tokio::test] 46 46 + async fn test_empty_car() { 47 47 + test_car(EMPTY_CAR, 0, 0, false).await 49 48 } 50 49 51 50 #[tokio::test] 52 51 async fn test_tiny_car() { 53 53 - test_car(TINY_CAR, 8, 2071).await 52 52 + test_car(TINY_CAR, 8, 2071, true).await 54 53 } 55 54 56 55 #[tokio::test] 57 56 async fn test_little_car() { 58 58 - test_car(LITTLE_CAR, 278, 246960).await 57 57 + test_car(LITTLE_CAR, 278, 246960, true).await 59 58 } 60 59 61 60 #[tokio::test] 62 61 async fn test_midsize_car() { 63 63 - test_car(MIDSIZE_CAR, 11585, 3741393).await 62 62 + test_car(MIDSIZE_CAR, 11585, 3741393, true).await 64 63 }