kanidmd_lib/repl/ruv.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351
use crate::be::dbrepl::DbReplMeta;
use std::cmp::Ordering;
use std::collections::{BTreeMap, BTreeSet};
use std::ops::Bound::*;
use std::sync::Arc;
use std::time::Duration;
use concread::bptree::{BptreeMap, BptreeMapReadSnapshot, BptreeMapReadTxn, BptreeMapWriteTxn};
use idlset::v2::IDLBitRange;
use crate::prelude::*;
use crate::repl::cid::Cid;
use crate::repl::proto::{ReplAnchoredCidRange, ReplCidRange};
use std::fmt;
#[derive(Default)]
pub struct ReplicationUpdateVector {
// This sorts by time. We store the set of entry id's that are affected in an operation.
// Due to how replication state works, it is possibly that id's in these sets *may* not
// exist anymore, so these bit ranges likely need intersection with allids before use.
data: BptreeMap<Cid, IDLBitRange>,
// This sorts by Server ID. It's used for the RUV to build ranges for you ... guessed it
// range queries. These are used to build the set of differences that need to be sent in
// a replication operation.
//
// we need a way to invert the cid, but without duplication? Maybe an invert cid type?
// This way it still orders things in the right order by time stamp just searches by cid
// first.
ranged: BptreeMap<Uuid, BTreeSet<Duration>>,
}
/// The status of replication after investigating the RUV states.
#[derive(Debug, PartialEq, Eq)]
pub(crate) enum RangeDiffStatus {
/// Ok - can proceed with replication, supplying the following
/// ranges of changes to the consumer.
Ok(BTreeMap<Uuid, ReplCidRange>),
/// Refresh - The consumer is lagging and is missing a set of changes
/// that are required to proceed. The consumer *MUST* be refreshed
/// immediately.
Refresh {
lag_range: BTreeMap<Uuid, ReplCidRange>,
},
/// Unwilling - The consumer is advanced beyond our state, and supplying
/// changes to them may introduce inconsistency in replication. This
/// server should be investigated immediately.
Unwilling {
adv_range: BTreeMap<Uuid, ReplCidRange>,
},
/// Critical - The consumer is lagging and missing changes, but also is
/// in possession of changes advancing it beyond our current state. This
/// is a critical fault in replication and the topology must be
/// investigated immediately.
Critical {
lag_range: BTreeMap<Uuid, ReplCidRange>,
adv_range: BTreeMap<Uuid, ReplCidRange>,
},
/// No RUV Overlap - The consumer has likely desynchronised and no longer has
/// common overlap with it's RUV to ours. This can indicate it has trimmed
/// content we may have, or may have become part of a split brain situation.
/// For replication to proceed, there must be *at least* one common overlapping
/// point in the RUV.
NoRUVOverlap,
}
impl ReplicationUpdateVector {
pub fn write(&self) -> ReplicationUpdateVectorWriteTransaction<'_> {
ReplicationUpdateVectorWriteTransaction {
// Need to take the write first, then the read to guarantee ordering.
added: Some(BTreeSet::default()),
data: self.data.write(),
data_pre: self.data.read(),
ranged: self.ranged.write(),
}
}
pub fn read(&self) -> ReplicationUpdateVectorReadTransaction<'_> {
ReplicationUpdateVectorReadTransaction {
data: self.data.read(),
ranged: self.ranged.read(),
}
}
pub(crate) fn range_diff(
consumer_range: &BTreeMap<Uuid, ReplCidRange>,
supplier_range: &BTreeMap<Uuid, ReplCidRange>,
) -> RangeDiffStatus {
// We need to build a new set of ranges that express the difference between
// these two states.
let mut diff_range = BTreeMap::default();
let mut lag_range = BTreeMap::default();
let mut adv_range = BTreeMap::default();
let mut consumer_lagging = false;
let mut supplier_lagging = false;
let mut valid_content_overlap = false;
// We need to look at each uuid in the *supplier* and assert if they are present
// on the *consumer*.
//
// If there are s_uuids with the same max, we don't add it to the
// diff
for (supplier_s_uuid, supplier_cid_range) in supplier_range.iter() {
match consumer_range.get(supplier_s_uuid) {
Some(consumer_cid_range) => {
// We have the same server uuid in our RUV's so some content overlap
// must exist (or has existed);
valid_content_overlap = true;
// The two windows just have to overlap. If they over lap
// meaning that consumer max > supplier min, then if supplier
// max > consumer max, then the range between consumer max
// and supplier max must be supplied.
//
// [ consumer min ... consumer max ]
// <-- [ supplier min .. supplier max ] -->
//
// In other words if we have:
//
// [ consumer min ... consumer max ]
// [ supplier min ... supplier max ]
// ^
// \-- no overlap of the range windows!
//
// then because there has been too much lag between consumer and
// the supplier then there is a risk of changes being dropped or
// missing. In the future we could alter this to force the resend
// of zero -> supplier max, but I think thought is needed to
// ensure no corruption in this case.
if consumer_cid_range.ts_max < supplier_cid_range.ts_min {
//
// [ consumer min ... consumer max ]
// [ supplier min ... supplier max ]
// ^
// \-- no overlap of the range windows!
//
consumer_lagging = true;
lag_range.insert(
*supplier_s_uuid,
ReplCidRange {
ts_min: supplier_cid_range.ts_min,
ts_max: consumer_cid_range.ts_max,
},
);
} else if supplier_cid_range.ts_max < consumer_cid_range.ts_min {
//
// [ consumer min ... consumer max ]
// [ supplier min ... supplier max ]
// ^
// \-- no overlap of the range windows!
//
// This means we can't supply because we are missing changes that the consumer
// has. *we* are lagging.
supplier_lagging = true;
adv_range.insert(
*supplier_s_uuid,
ReplCidRange {
ts_min: supplier_cid_range.ts_max,
ts_max: consumer_cid_range.ts_min,
},
);
} else if consumer_cid_range.ts_max < supplier_cid_range.ts_max {
//
// /-- consumer needs these changes
// v
// [ consumer min ... consumer max ] --> ]
// [ supplier min ... supplier max ]
// ^
// \-- overlap of the range windows
//
// We require the changes from consumer max -> supplier max.
diff_range.insert(
*supplier_s_uuid,
ReplCidRange {
ts_min: consumer_cid_range.ts_max,
ts_max: supplier_cid_range.ts_max,
},
);
}
//
// /-- The consumer has changes we don't have.
// | So we don't need to supply
// v
// [ consumer min ... consumer max ]
// [ supplier min ... supplier max ]
// ^
// \-- overlap of the range windows
//
// OR
//
// [ consumer min ... consumer max ]
// [ supplier min ... supplier max ]
// ^
// \-- the windows max is identical
// no actions needed
//
// In this case there is no action required since consumer_cid_range.ts_max
// must be greater than or equal to supplier max.
}
None => {
// The consumer does not have any content from this
// server. Select from Zero -> max of the supplier.
diff_range.insert(
*supplier_s_uuid,
ReplCidRange {
ts_min: Duration::ZERO,
ts_max: supplier_cid_range.ts_max,
},
);
}
}
}
if !valid_content_overlap {
return RangeDiffStatus::NoRUVOverlap;
}
match (consumer_lagging, supplier_lagging) {
(false, false) => RangeDiffStatus::Ok(diff_range),
(true, false) => RangeDiffStatus::Refresh { lag_range },
(false, true) => RangeDiffStatus::Unwilling { adv_range },
(true, true) => RangeDiffStatus::Critical {
lag_range,
adv_range,
},
}
}
}
pub struct ReplicationUpdateVectorWriteTransaction<'a> {
added: Option<BTreeSet<Cid>>,
data: BptreeMapWriteTxn<'a, Cid, IDLBitRange>,
data_pre: BptreeMapReadTxn<'a, Cid, IDLBitRange>,
ranged: BptreeMapWriteTxn<'a, Uuid, BTreeSet<Duration>>,
}
impl fmt::Debug for ReplicationUpdateVectorWriteTransaction<'_> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
writeln!(f, "RUV DATA DUMP")?;
self.data
.iter()
.try_for_each(|(cid, idl)| writeln!(f, "* [{cid} {idl:?}]"))?;
writeln!(f, "RUV RANGE DUMP")?;
self.ranged
.iter()
.flat_map(|(s_uuid, ts_set)| ts_set.iter().map(|ts| Cid::new(*s_uuid, *ts)))
.try_for_each(|cid| writeln!(f, "[{cid}]"))
}
}
pub struct ReplicationUpdateVectorReadTransaction<'a> {
data: BptreeMapReadTxn<'a, Cid, IDLBitRange>,
ranged: BptreeMapReadTxn<'a, Uuid, BTreeSet<Duration>>,
}
impl fmt::Debug for ReplicationUpdateVectorReadTransaction<'_> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
writeln!(f, "RUV DATA DUMP")?;
self.data
.iter()
.try_for_each(|(cid, idl)| writeln!(f, "* [{cid} {idl:?}]"))?;
writeln!(f, "RUV RANGE DUMP")?;
self.ranged
.iter()
.try_for_each(|(s_uuid, ts)| writeln!(f, "* [{s_uuid} {ts:?}]"))
}
}
pub trait ReplicationUpdateVectorTransaction {
fn ruv_snapshot(&self) -> BptreeMapReadSnapshot<'_, Cid, IDLBitRange>;
fn range_snapshot(&self) -> BptreeMapReadSnapshot<'_, Uuid, BTreeSet<Duration>>;
fn to_db_backup_ruv(&self) -> DbReplMeta {
DbReplMeta::V1 {
ruv: self.ruv_snapshot().keys().map(|cid| cid.into()).collect(),
}
}
/// Return a filtered view of our RUV ranges. This acts similar to "trim" where any s_uuid
/// where the max cid is less than trim_cid will be excluded from the view.
fn filter_ruv_range(
&self,
trim_cid: &Cid,
) -> Result<BTreeMap<Uuid, ReplCidRange>, OperationError> {
self.range_snapshot()
.iter()
.filter_map(|(s_uuid, range)| match (range.first(), range.last()) {
(Some(first), Some(last)) => {
if last < &trim_cid.ts {
None
} else {
Some(Ok((
*s_uuid,
ReplCidRange {
ts_min: *first,
ts_max: *last,
},
)))
}
}
_ => {
error!(
"invalid state for server uuid {:?}, no ranges present",
s_uuid
);
Some(Err(OperationError::InvalidState))
}
})
.collect::<Result<BTreeMap<_, _>, _>>()
}
/// Return the complete set of RUV ranges present on this replica
fn current_ruv_range(&self) -> Result<BTreeMap<Uuid, ReplCidRange>, OperationError> {
self.range_snapshot()
.iter()
.map(|(s_uuid, range)| match (range.first(), range.last()) {
(Some(first), Some(last)) => Ok((
*s_uuid,
ReplCidRange {
ts_min: *first,
ts_max: *last,
},
)),
_ => {
error!(
"invalid state for server uuid {:?}, no ranges present",
s_uuid
);
Err(OperationError::InvalidState)
}
})
.collect::<Result<BTreeMap<_, _>, _>>()
}
fn range_to_idl(&self, ctx_ranges: &BTreeMap<Uuid, ReplCidRange>) -> IDLBitRange {
let mut idl = IDLBitRange::new();
// Force the set to be compressed, saves on seeks during inserts.
idl.compress();
let range = self.range_snapshot();
let ruv = self.ruv_snapshot();
// The range we have has a collection of s_uuid containing low -> high ranges.
// We need to convert this to absolute ranges of all the idlbitranges that
// relate to the entries we have.
for (s_uuid, ctx_range) in ctx_ranges {
// For each server and range low to high, iterate over
// the list of CID's in the main RUV.
let Some(ruv_range) = range.get(s_uuid) else {
// This is valid because if we clean up a server range on
// this node, but the other server isn't aware yet, so we
// just no-op this. The changes we have will still be
// correctly found and sent.
debug!(?s_uuid, "range not found in ruv.");
continue;
};
// Get from the min to the max. Unbounded and
// Included(ctx_range.ts_max) are the same in
// this context.
for ts in ruv_range.range((Excluded(ctx_range.ts_min), Unbounded)) {
let cid = Cid {
ts: *ts,
s_uuid: *s_uuid,
};
if let Some(ruv_idl) = ruv.get(&cid) {
ruv_idl.into_iter().for_each(|id| idl.insert_id(id))
}
// If the cid isn't found, it may have been trimmed, but that's okay. A cid in
// a range can be trimmed if all entries of that cid have since tombstoned so
// no longer need to be applied in change ranges.
}
}
idl
}
fn verify(
&self,
entries: &[Arc<EntrySealedCommitted>],
results: &mut Vec<Result<(), ConsistencyError>>,
) {
// Okay rebuild the RUV in parallel.
let mut check_ruv: BTreeMap<Cid, IDLBitRange> = BTreeMap::default();
for entry in entries {
// The DB id we need.
let eid = entry.get_id();
let ecstate = entry.get_changestate();
// We don't need the details of the change - only the cid of the
// change that this entry was involved in.
for cid in ecstate.cid_iter() {
// Add to the main ruv data.
if let Some(idl) = check_ruv.get_mut(cid) {
// We can't guarantee id order, so we have to do this properly.
idl.insert_id(eid);
} else {
let mut idl = IDLBitRange::new();
idl.insert_id(eid);
check_ruv.insert(cid.clone(), idl);
}
}
}
trace!(?check_ruv);
// Get the current state
let snapshot_ruv = self.ruv_snapshot();
// Now compare. We want to do this checking for each CID in each, and then asserting
// the content is the same.
let mut check_iter = check_ruv.iter();
let mut snap_iter = snapshot_ruv.iter();
let mut check_next = check_iter.next();
let mut snap_next = snap_iter.next();
while let (Some((ck, cv)), Some((sk, sv))) = (&check_next, &snap_next) {
match ck.cmp(sk) {
Ordering::Equal => {
// Counter intuitive, but here we check that the check set is a *subset*
// of the ruv snapshot. This is because when we have an entry that is
// tombstoned, all it's CID interactions are "lost" and it's cid becomes
// that of when it was tombstoned. So the "rebuilt" ruv will miss that
// entry.
//
// In the future the RUV concept may be ditched entirely anyway, thoughts needed.
let intersect = *cv & *sv;
if *cv == &intersect {
trace!("{:?} is consistent!", ck);
} else {
error!("{:?} is NOT consistent! IDL's differ", ck);
debug_assert!(false);
results.push(Err(ConsistencyError::RuvInconsistent(ck.to_string())));
}
check_next = check_iter.next();
snap_next = snap_iter.next();
}
// Because we are zipping between these two sets, we only need to compare when
// the CID's are equal. Otherwise we need the other iter to "catch up"
Ordering::Less => {
check_next = check_iter.next();
}
Ordering::Greater => {
snap_next = snap_iter.next();
}
}
}
while let Some((ck, _cv)) = &check_next {
debug!("{:?} may not be consistent! CID missing from RUV", ck);
// debug_assert!(false);
// results.push(Err(ConsistencyError::RuvInconsistent(ck.to_string())));
check_next = check_iter.next();
}
while let Some((sk, _sv)) = &snap_next {
debug!(
"{:?} may not be consistent! CID should not exist in RUV",
sk
);
// debug_assert!(false);
// results.push(Err(ConsistencyError::RuvInconsistent(sk.to_string())));
snap_next = snap_iter.next();
}
// Assert that the content of the ranged set matches the data set and has the
// correct set of values.
let snapshot_range = self.range_snapshot();
for cid in snapshot_ruv.keys() {
if let Some(server_range) = snapshot_range.get(&cid.s_uuid) {
if !server_range.contains(&cid.ts) {
warn!(
"{:?} is NOT consistent! server range is missing cid in index",
cid
);
debug_assert!(false);
results.push(Err(ConsistencyError::RuvInconsistent(
cid.s_uuid.to_string(),
)));
}
} else {
warn!(
"{:?} is NOT consistent! server range is not present",
cid.s_uuid
);
debug_assert!(false);
results.push(Err(ConsistencyError::RuvInconsistent(
cid.s_uuid.to_string(),
)));
}
}
// Done!
}
fn get_anchored_ranges(
&self,
ranges: BTreeMap<Uuid, ReplCidRange>,
) -> Result<BTreeMap<Uuid, ReplAnchoredCidRange>, OperationError> {
let self_range_snapshot = self.range_snapshot();
ranges
.into_iter()
.map(|(s_uuid, ReplCidRange { ts_min, ts_max })| {
let ts_range = self_range_snapshot.get(&s_uuid).ok_or_else(|| {
error!(
?s_uuid,
"expected cid range for server in ruv, was not present"
);
OperationError::InvalidState
})?;
// If these are equal and excluded, btreeset panics
let anchors = if ts_max > ts_min {
// We exclude the ends because these are already in the ts_min/max
ts_range
.range((Excluded(ts_min), Excluded(ts_max)))
.copied()
.collect::<Vec<_>>()
} else {
Vec::with_capacity(0)
};
Ok((
s_uuid,
ReplAnchoredCidRange {
ts_min,
anchors,
ts_max,
},
))
})
.collect()
}
}
impl ReplicationUpdateVectorTransaction for ReplicationUpdateVectorWriteTransaction<'_> {
fn ruv_snapshot(&self) -> BptreeMapReadSnapshot<'_, Cid, IDLBitRange> {
self.data.to_snapshot()
}
fn range_snapshot(&self) -> BptreeMapReadSnapshot<'_, Uuid, BTreeSet<Duration>> {
self.ranged.to_snapshot()
}
}
impl ReplicationUpdateVectorTransaction for ReplicationUpdateVectorReadTransaction<'_> {
fn ruv_snapshot(&self) -> BptreeMapReadSnapshot<'_, Cid, IDLBitRange> {
self.data.to_snapshot()
}
fn range_snapshot(&self) -> BptreeMapReadSnapshot<'_, Uuid, BTreeSet<Duration>> {
self.ranged.to_snapshot()
}
}
impl ReplicationUpdateVectorWriteTransaction<'_> {
pub fn clear(&mut self) {
self.added = None;
self.data.clear();
self.ranged.clear();
}
pub(crate) fn incremental_preflight_validate_ruv(
&self,
ctx_ranges: &BTreeMap<Uuid, ReplAnchoredCidRange>,
txn_cid: &Cid,
) -> Result<(), OperationError> {
// Check that the incoming ranges, for our servers id, do not exceed
// our servers max state. This can occur if we restore from a backup
// where the replication state is older than what our partners have,
// meaning that the context may have diverged in a way we can't then
// resolve.
if let Some(our_cid_range_max) = self
.ranged
.get(&txn_cid.s_uuid)
.and_then(|range| range.last().copied())
{
if let Some(incoming_cid_range) = ctx_ranges.get(&txn_cid.s_uuid) {
if incoming_cid_range.ts_max > our_cid_range_max {
error!("The incoming data contains changes matching this server's UUID, and those changes are newer than the local version. This can occur if the server was restored from a backup which was taken before sending out changes. Replication is unable to proceed as data corruption may occur. You must refresh this consumer immediately to continue.");
return Err(OperationError::ReplServerUuidSplitDataState);
}
}
}
let warn_time = txn_cid.ts + REPL_SUPPLIER_ADVANCE_WINDOW;
for (s_uuid, incoming_cid_range) in ctx_ranges.iter() {
if incoming_cid_range.ts_max > warn_time {
// TODO: This would be a great place for fault management to pick up this warning
warn!(
"Incoming changes from {:?} are further than {} seconds in the future.",
s_uuid,
REPL_SUPPLIER_ADVANCE_WINDOW.as_secs()
);
}
}
Ok(())
}
pub(crate) fn refresh_validate_ruv(
&self,
ctx_ranges: &BTreeMap<Uuid, ReplAnchoredCidRange>,
) -> Result<(), OperationError> {
// Assert that the ruv that currently exists, is a valid data set of
// the supplied consumer range - especially check that when a uuid exists in
// our ruv, that it's maximum matches the ctx ruv.
//
// Since the ctx range comes from the supplier, when we rebuild due to the
// state machine then some values may not exist since they were replaced
// or updated. It's also possible that the imported range maximums *may not*
// exist especially in three way replication scenarios where S1:A was the S1
// maximum but is replaced by S2:B. This would make S1:A still it's valid
// maximum but no entry reflects that in it's change state.
let mut valid = true;
for (ctx_server_uuid, ctx_server_range) in ctx_ranges.iter() {
match self.ranged.get(ctx_server_uuid) {
Some(server_range) => {
let ctx_ts = &ctx_server_range.ts_max;
match server_range.last() {
Some(s_ts) if s_ts <= ctx_ts => {
// Ok - our entries reflect maximum or earlier.
trace!(?ctx_server_uuid, ?ctx_ts, ?s_ts, "valid");
}
Some(s_ts) => {
valid = false;
warn!(?ctx_server_uuid, ?ctx_ts, ?s_ts, "inconsistent s_uuid in ruv, consumer ruv is advanced past supplier");
}
None => {
valid = false;
warn!(
?ctx_server_uuid,
?ctx_ts,
"inconsistent server range in ruv, no maximum ts found for s_uuid"
);
}
}
}
None => {
// valid = false;
trace!(
?ctx_server_uuid,
"s_uuid absent from ranged ruv, possible that changes have been expired"
);
}
}
}
if valid {
Ok(())
} else {
Err(OperationError::ReplInvalidRUVState)
}
}
#[instrument(level = "trace", name = "ruv::refresh_update_ruv", skip_all)]
pub(crate) fn refresh_update_ruv(
&mut self,
ctx_ranges: &BTreeMap<Uuid, ReplAnchoredCidRange>,
) -> Result<(), OperationError> {
// Previously this would just add in the ranges, and then the actual entries
// from the changestate would populate the data/ranges. Now we add empty idls
// to each of these so that they are db persisted allowing ruv reload.
for (ctx_s_uuid, ctx_range) in ctx_ranges.iter() {
let cid_iter = std::iter::once(&ctx_range.ts_min)
.chain(ctx_range.anchors.iter())
.chain(std::iter::once(&ctx_range.ts_max))
.map(|ts| Cid::new(*ctx_s_uuid, *ts));
for cid in cid_iter {
self.insert_change(&cid, IDLBitRange::default())?;
}
}
Ok(())
}
/// Restore the ruv from a DB backup. It's important to note here that
/// we don't actually need to restore and of the IDL's in the process. we only
/// needs the CID's of the changes/points in time. This is because when the
/// db entries are restored, their changesets will re-populate the data that we
/// need in the RUV at these points. The reason we need these ranges without IDL
/// is so that trim and replication works properly.
#[instrument(level = "debug", name = "ruv::restore", skip_all)]
pub(crate) fn restore<I>(&mut self, iter: I) -> Result<(), OperationError>
where
I: IntoIterator<Item = Cid>,
{
let mut rebuild_ruv: BTreeMap<Cid, IDLBitRange> = BTreeMap::new();
let mut rebuild_range: BTreeMap<Uuid, BTreeSet<Duration>> = BTreeMap::default();
for cid in iter {
if !rebuild_ruv.contains_key(&cid) {
let idl = IDLBitRange::new();
rebuild_ruv.insert(cid.clone(), idl);
}
if let Some(server_range) = rebuild_range.get_mut(&cid.s_uuid) {
server_range.insert(cid.ts);
} else {
let mut ts_range = BTreeSet::default();
ts_range.insert(cid.ts);
rebuild_range.insert(cid.s_uuid, ts_range);
}
}
self.data.extend(rebuild_ruv);
self.ranged.extend(rebuild_range);
Ok(())
}
#[instrument(level = "debug", name = "ruv::rebuild", skip_all)]
pub fn rebuild(&mut self, entries: &[Arc<EntrySealedCommitted>]) -> Result<(), OperationError> {
// Entries and their internal changestates are the "source of truth" for all changes
// that have ever occurred and are stored on this server. So we use them to rebuild our RUV
// here!
//
// We only update RUV items where an anchor exists.
// let mut rebuild_ruv: BTreeMap<Cid, IDLBitRange> = BTreeMap::new();
// let mut rebuild_range: BTreeMap<Uuid, BTreeSet<Duration>> = BTreeMap::default();
for entry in entries {
// The DB id we need.
let eid = entry.get_id();
let ecstate = entry.get_changestate();
// We don't need the details of the change - only the cid of the
// change that this entry was involved in.
for cid in ecstate.cid_iter() {
// if let Some(idl) = rebuild_ruv.get_mut(cid) {
if let Some(idl) = self.data.get_mut(cid) {
// We can't guarantee id order, so we have to do this properly.
idl.insert_id(eid);
/*
} else {
let mut idl = IDLBitRange::new();
idl.insert_id(eid);
rebuild_ruv.insert(cid.clone(), idl);
*/
}
/*
if let Some(server_range) = rebuild_range.get_mut(&cid.s_uuid) {
server_range.insert(cid.ts);
} else {
let mut ts_range = BTreeSet::default();
ts_range.insert(cid.ts);
rebuild_range.insert(cid.s_uuid, ts_range);
}
*/
}
}
// Finally, we need to do a cleanup/compact of the IDL's if possible.
self.data.range_mut(..).for_each(|(_k, idl)| {
idl.maybe_compress();
});
// self.data.extend(rebuild_ruv);
// Warning - you can't extend here because this is keyed by UUID. You need
// to go through each key and then merge the sets.
/*
rebuild_range.into_iter().for_each(|(s_uuid, ts_set)| {
if let Some(ex_ts_set) = self.ranged.get_mut(&s_uuid) {
ex_ts_set.extend(ts_set)
} else {
self.ranged.insert(s_uuid, ts_set);
}
});
*/
Ok(())
}
pub fn insert_change(&mut self, cid: &Cid, idl: IDLBitRange) -> Result<(), OperationError> {
// Remember, in a transaction the changes can be updated multiple times.
if let Some(ex_idl) = self.data.get_mut(cid) {
// This ensures both sets have all the available ids.
let idl = ex_idl as &_ | &idl;
*ex_idl = idl;
} else {
self.data.insert(cid.clone(), idl);
}
if let Some(server_range) = self.ranged.get_mut(&cid.s_uuid) {
server_range.insert(cid.ts);
} else {
let mut range = BTreeSet::default();
range.insert(cid.ts);
self.ranged.insert(cid.s_uuid, range);
}
if let Some(added) = &mut self.added {
added.insert(cid.clone());
}
Ok(())
}
pub fn update_entry_changestate(
&mut self,
entry: &EntrySealedCommitted,
) -> Result<(), OperationError> {
let eid = entry.get_id();
let ecstate = entry.get_changestate();
trace!("Updating ruv state from entry {}", eid);
trace!(?ecstate);
for cid in ecstate.cid_iter() {
if let Some(idl) = self.data.get_mut(cid) {
// We can't guarantee id order, so we have to do this properly.
idl.insert_id(eid);
} else {
let mut idl = IDLBitRange::new();
idl.insert_id(eid);
self.data.insert(cid.clone(), idl);
}
if let Some(server_range) = self.ranged.get_mut(&cid.s_uuid) {
server_range.insert(cid.ts);
} else {
let mut ts_range = BTreeSet::default();
ts_range.insert(cid.ts);
self.ranged.insert(cid.s_uuid, ts_range);
}
}
Ok(())
}
pub fn ruv_idls(&self) -> IDLBitRange {
let mut idl = IDLBitRange::new();
self.data.iter().for_each(|(_cid, ex_idl)| {
idl = ex_idl as &_ | &idl;
});
idl
}
/*
How to handle changelog trimming? If we trim a server out from the RUV as a whole, we
need to be sure we don't oversupply changes the consumer already has. How can we do
this cleanly? Or do we just deal with it because our own local trim will occur soon after?
The situation would be
A: 1 -> 3
B: 1 -> 3
Assuming A trims first:
A:
B: 1 -> 3
Then on A <- B, B would try to supply 1->3 to A assuming it is not present. However,
the trim would occur soon after on B causing:
A:
B:
And then the supply would stop. So either A needs to retain the max/min in it's range
to allow the comparison here to continue even if it's ruv is cleaned. Or, we need to
have a delayed trim on the range that is 2x the normal trim range to give a buffer?
Mostly longer ruv/cid ranges aren't an issue for us, so could we just make these ranges
really large?
NOTE: For now we do NOT trim out max CID's of any s_uuid so that we don't have to confront
this edge case yet.
// == RESOLVED: Previously this was a problem as the CID ranges of any node may not be a
// complete view of all CID's that existed on any other node. Now with anchors in replication
// this changes and we have not only a complete view of all CID's that were created, but
// tombstone purge always create an empty anchor so the RUV always advances. This way we
// have a stronger assurance about which servers are live and which are not.
*/
// Problem Cases
/*
What about generations? There is a "live" generation which can be replicated and a
former generation of ranges that previously existed. To replicate:
// The consumer must have content within the current live range.
consumer.live_max < supplier.live_max
consumer.live_max >= supplier.live_min
// The consumer must have all content that was formerly known.
consumer.live_min >= supplier.former_max
// I don't think we care what
// == RESOLVED: Anchors give us the generations that existed previously without us
// needing to worry about this.
*/
pub fn trim_up_to(&mut self, cid: &Cid) -> Result<IDLBitRange, OperationError> {
trace!(trim_up_to_cid = ?cid);
let mut idl = IDLBitRange::new();
let mut remove_suuid = Vec::with_capacity(0);
// Here we can use the for_each here to be trimming the
// range set since that is not ordered by time, we need
// to do fragmented searches over this no matter what we
// try to do.
for (cid, ex_idl) in self.data.range((Unbounded, Excluded(cid))) {
trace!(?cid, "examining for RUV removal");
idl = ex_idl as &_ | &idl;
// Remove the reverse version of the cid from the ranged index.
match self.ranged.get_mut(&cid.s_uuid) {
Some(server_range) => {
// Remove returns a bool if the element WAS present.
if !server_range.remove(&cid.ts) {
error!("Impossible State - The RUV is corrupted due to missing sid:ts pair in ranged index");
error!(ruv = ?self);
error!(?cid);
return Err(OperationError::InvalidState);
}
if server_range.is_empty() {
remove_suuid.push(cid.s_uuid);
warn!(s_uuid = ?cid.s_uuid, "disconnected server detected - this will be removed!");
} else {
trace!(?server_range, "retaining server");
}
}
None => {
error!("Impossible State - The RUV is corrupted due to missing sid in ranged index");
error!(ruv = ?self);
error!(?cid);
return Err(OperationError::InvalidState);
}
}
}
// We can now remove old server id's because we have a reliable liveness check in the
// method of anchors being transmissed during replication. If a server is offline for
// an extended period, it will not have created any anchors, and it will eventually become
// empty in the data range. This allow it to be trimmed out.
for s_uuid in remove_suuid {
let x = self.ranged.remove(&s_uuid);
assert!(x.map(|y| y.is_empty()).unwrap_or(false))
}
// Trim all cid's up to this value, and return the range of IDs
// that are affected.
self.data.split_off_lt(cid);
Ok(idl)
}
pub fn added(&self) -> Box<dyn Iterator<Item = Cid> + '_> {
if let Some(added) = self.added.as_ref() {
// return what was added this txn. We previously would iterate
// from data_pre.max() with data, but if an anchor was added that
// pre-dated data_pre.max() it wouldn't be committed to the db ruv
// (even though it was in the in memory ruv).
Box::new(added.iter().map(|cid| {
debug!(added_cid = ?cid);
cid.clone()
}))
} else {
// We have been cleared during this txn, so everything in data is
// added.
Box::new(self.data.iter().map(|(cid, _)| {
debug!(added_cid = ?cid);
cid.clone()
}))
}
}
pub fn removed(&self) -> impl Iterator<Item = Cid> + '_ {
let prev_bound = if self.added.is_none() {
// We have been cleared during this txn, so everything in pre is
// removed.
Unbounded
} else if let Some((min, _)) = self.data.first_key_value() {
Excluded(min.clone())
} else {
// If empty, assume everything is removed.
Unbounded
};
// iterate through our previous data to find what has been removed given
// the ranges determined above.
self.data_pre
.range((Unbounded, prev_bound))
.map(|(cid, _)| {
debug!(removed_cid = ?cid);
cid.clone()
})
}
pub fn commit(self) {
self.data.commit();
self.ranged.commit();
}
}
#[cfg(test)]
mod tests {
use super::RangeDiffStatus;
use super::ReplCidRange;
use super::ReplicationUpdateVector;
use std::collections::BTreeMap;
use std::time::Duration;
const UUID_A: uuid::Uuid = uuid::uuid!("13b530b0-efdd-4934-8fb7-9c35c8aab79e");
const UUID_B: uuid::Uuid = uuid::uuid!("16327cf8-6a34-4a17-982c-b2eaa6d02d00");
const UUID_C: uuid::Uuid = uuid::uuid!("2ed717e3-15be-41e6-b966-10a1f6d7ea1c");
#[test]
fn test_ruv_range_diff_1() {
let ctx_a = BTreeMap::default();
let ctx_b = BTreeMap::default();
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::NoRUVOverlap;
assert_eq!(result, expect);
// Test the inverse.
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::NoRUVOverlap;
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_2() {
let ctx_a = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(3),
}
));
let ctx_b = BTreeMap::default();
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::NoRUVOverlap;
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::NoRUVOverlap;
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_3() {
let ctx_a = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(3),
}
));
let ctx_b = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(3),
}
));
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::Ok(BTreeMap::default());
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::Ok(BTreeMap::default());
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_4() {
let ctx_a = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(3),
}
));
let ctx_b = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(4),
}
));
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::Ok(btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(3),
ts_max: Duration::from_secs(4),
}
)));
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::Ok(BTreeMap::default());
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_5() {
let ctx_a = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(5),
ts_max: Duration::from_secs(7),
}
));
let ctx_b = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(4),
}
));
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::Unwilling {
adv_range: btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(4),
ts_max: Duration::from_secs(5),
}
)),
};
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::Refresh {
lag_range: btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(5),
ts_max: Duration::from_secs(4),
}
)),
};
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_6() {
let ctx_a = btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(4),
}
));
let ctx_b = btreemap!(
(
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(3),
}
),
(
UUID_B,
ReplCidRange {
ts_min: Duration::from_secs(2),
ts_max: Duration::from_secs(4),
}
)
);
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::Ok(btreemap!((
UUID_B,
ReplCidRange {
ts_min: Duration::ZERO,
ts_max: Duration::from_secs(4),
}
)));
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::Ok(btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(3),
ts_max: Duration::from_secs(4),
}
)));
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_7() {
let ctx_a = btreemap!(
(
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(4),
}
),
(
UUID_C,
ReplCidRange {
ts_min: Duration::from_secs(2),
ts_max: Duration::from_secs(5),
}
)
);
let ctx_b = btreemap!(
(
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(3),
}
),
(
UUID_B,
ReplCidRange {
ts_min: Duration::from_secs(2),
ts_max: Duration::from_secs(4),
}
),
(
UUID_C,
ReplCidRange {
ts_min: Duration::from_secs(3),
ts_max: Duration::from_secs(4),
}
)
);
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::Ok(btreemap!((
UUID_B,
ReplCidRange {
ts_min: Duration::ZERO,
ts_max: Duration::from_secs(4),
}
)));
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::Ok(btreemap!(
(
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(3),
ts_max: Duration::from_secs(4),
}
),
(
UUID_C,
ReplCidRange {
ts_min: Duration::from_secs(4),
ts_max: Duration::from_secs(5),
}
)
));
assert_eq!(result, expect);
}
#[test]
fn test_ruv_range_diff_8() {
let ctx_a = btreemap!(
(
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(4),
ts_max: Duration::from_secs(6),
}
),
(
UUID_B,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(2),
}
)
);
let ctx_b = btreemap!(
(
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(1),
ts_max: Duration::from_secs(2),
}
),
(
UUID_B,
ReplCidRange {
ts_min: Duration::from_secs(4),
ts_max: Duration::from_secs(6),
}
)
);
let result = ReplicationUpdateVector::range_diff(&ctx_a, &ctx_b);
let expect = RangeDiffStatus::Critical {
adv_range: btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(2),
ts_max: Duration::from_secs(4),
}
)),
lag_range: btreemap!((
UUID_B,
ReplCidRange {
ts_min: Duration::from_secs(4),
ts_max: Duration::from_secs(2),
}
)),
};
assert_eq!(result, expect);
let result = ReplicationUpdateVector::range_diff(&ctx_b, &ctx_a);
let expect = RangeDiffStatus::Critical {
adv_range: btreemap!((
UUID_B,
ReplCidRange {
ts_min: Duration::from_secs(2),
ts_max: Duration::from_secs(4),
}
)),
lag_range: btreemap!((
UUID_A,
ReplCidRange {
ts_min: Duration::from_secs(4),
ts_max: Duration::from_secs(2),
}
)),
};
assert_eq!(result, expect);
}
}