Commit Graph
98 Commits
Author SHA1 Message Date
leventov b2cd5af700 Rework modification iteration and bootstrapping 2015-12-30 20:38:55 +03:00
leventov d060fd0c2d HCOLL-412 fix a bug in HashLookupSearch.putNewVolatile() 2015-12-24 18:31:32 +03:00
leventov 263e366ee8 Update delayed check sum on WriteLock.unlock(), and close stages only if unlocking actually done 2015-12-23 18:37:14 +03:00
leventov 9ff159987e Release writeLock after visiting each entry in HashSegmentIteration.forEachTierEntryWhile() 2015-12-23 16:30:20 +03:00
leventov 246eec8b77 Fix a bug in SegmentStages.unlinkFromSegmentContextsChain() -- didn't account for exceptional exit from nested context 2015-12-23 16:29:08 +03:00
leventov 3773b3d60a Add methods for Engine support 2015-12-18 03:00:38 +03:00
leventov 27c8148007 Fix the prev commit e78e070 2015-12-17 18:16:51 +03:00
leventov a11f79f798 Remove map persistence file locking, as unnecessary 2015-12-14 18:49:47 +03:00
leventov 4907af142b Prefix Read/Update/WriteLock methods with checkOnEachLockOperation() 2015-12-11 18:50:01 +03:00
leventov 7a5c799a28 Remove throws IORuntimeException everywhere as redundant and deprecated 2015-12-11 16:23:45 +03:00
leventov c79b078c36 Apply timed tryLock() heuristic from commit 6c56813b5e to tryWriteLock() and tryUpgradeUpdateToWriteLock(), was used only in tryReadLock() and tryUpdateLock() 2015-12-10 17:43:50 +03:00
leventov 9390aa87a5 Replace RWU lock implementation in BigSegmentHeader with identical from VanillaReadWriteUpdateWithWaitsLockingStrategy (DRY), remove IllegalInterProcessLockStateException in favor of java.lang.IllegalMonitorStateException 2015-12-10 17:43:47 +03:00
leventov 8abe53912a Make max hashLookup load factor (for tiering) greater than max 'normal'
load factor (0.66), to avoid tiering due to natural variance in segment
load. Account non-tiered segments percentile
2015-12-10 17:43:47 +03:00
leventov b711ca8d96 Remove unused and incomplete support of keeping exclusive lock holder thread id in segment header 2015-12-09 08:24:02 +03:00
leventov c8483dfb7c HCOLL-408 Trigger new segment tier allocation on hashLookup overflow 2015-12-07 12:31:28 +03:00
leventov ce45903af7 Make entries and deleted counters per-tier 2015-12-07 11:43:37 +03:00
leventov 2f58bfb399 Replace nextPosToSearchFrom with lowestPossiblyFreeChunk semantically, that allows to avoid searching free chunks though a full tier, when segment is tiered, again and again. Also, make chunk block reallocation in innerDefaultReplaceValue() and putValueDeletedEntry() check it doesn't overflow actualChunksPerSegmentTier 2015-12-07 07:20:47 +03:00
leventov 82988e7435 Replace outdated (pre-tiers) implementation of ReplicatedMapSegmentIteration.forEachSegmentReplicableEntryWhile() by shared code with HashSegmentIteration.forEachSegmentEntryWhile() 2015-12-06 15:19:49 +03:00
leventov 23b3eb7c7a Rename a VanillaChronicleMap.zeroOutNewlyMappedTier parameter name 2015-12-05 05:40:37 +03:00
leventov d5ee1f9aae Make checksum computation delayed, on lock release 2015-12-05 05:37:17 +03:00
leventov 8ea5f2dfac Make searchStatePresent() call in HashLookupSearch.putNewVolatile() independent on assertions enabled. Similar to HashLookupPos.initHashLookupPos() and segmentTier() initialization. Possible fix for #60 2015-12-03 09:58:07 +03:00
leventov aad6c115c2 Replace Crc32 checksum computation with xxHash, because it is faster according to https://github.com/benalexau/hash-bench 2015-11-30 00:34:22 +03:00
leventov a0c982374d Fix a bug in HashSegmentIteration.forEachTierWhile() 2015-11-29 01:52:57 +03:00
leventov 2d7e25985a Replace CityHash 1.1 with xxHash as the default hash function 2015-11-28 23:31:25 +03:00
leventov bfa9988551 Add ChronicleSetBuilder.entryOperations() and .remoteOperations() interception configuration methods; Rename SetRemoteOperations.insert() to put(), for consistency with MapRemoteOperations 2015-11-28 23:20:29 +03:00
leventov 1f3fc3f785 Don't rely on default field values, when VanillaChronicleMap instance is unsafe.allocateInstance()-ed for deserialization 2015-11-27 15:25:13 +03:00
leventov 4faefe8ca9 Improve FileChannel read/write for persisted ChronicleMaps 2015-11-26 18:49:54 +03:00
leventov e29a473e05 Flush persistence file contents on ChronicleMap.close() 2015-11-26 15:31:48 +03:00
leventov 12989df576 Remove potentially dangerous freeList bits clearing before allocation of a new block, on relocation, because overlapping range copy is platform dependent 2015-11-25 22:29:55 +03:00
leventov 25edcf9b68 Little simplification in VanillaChronicleHash.createMappedStoreAndSegments(File) 2015-11-25 22:08:53 +03:00
leventov fe77012381 Read/Write ChronicleMap's boostrapping header with Wire 2015-11-25 20:37:19 +03:00
leventov f575218da9 Extend persistence file on mapping 2015-11-23 17:42:36 +03:00
leventov 701c1c8e52 Rename VanillaChronicleHash.msBytesSegmentOffset -> segmentOffset 2015-11-23 17:16:19 +03:00
leventov ade2196e72 Java-Lang -> Chronicle-Bytes, Chronicle-Values 2015-11-22 22:10:39 +03:00
leventov 4c04a942f7 Update to the latest chronicle-bom 2015-10-22 00:18:31 +03:00
leventov a13bf449d0 Make value and value size configuration obligatory in ChronicleMapBuilder 2015-10-14 17:47:22 +03:00
leventov 7c4b0ed10b Remove tier bulk alignment (as actually not needed) 2015-10-11 17:04:20 +03:00
leventov f1a13a4dfd Rename rootContextOnThisSegment -> rootContextLockedOnThisSegment and concurrentSameThreadContexts -> nestedContextsLockedOnSameSegment -- clearer names 2015-10-09 19:17:59 +03:00
leventov cad3d3abfe HCOLL-400 forbid nested update and write locks inside iteration contexts; rename IterationContextInterface to IterationContext; update stage-compiler dependency 2015-10-09 15:09:32 +03:00
leventov 96cee73f26 Unconditional zeroing out of newly allocated memory, because 1) it is very slow in Java-Lang 6.7.3 due to a bug 2) not clear which way is better, see HCOLL-401 2015-10-07 16:51:21 +03:00
leventov 81535163b2 Make IterationContexts' hashLookupPos depend on segmentTier as well as QueryContexts'; add missed hashLookupPos step back in iteration, if hash lookup slot is removed, because entry is relocated to a chained tier; add missed tierIndex initialization in initSegmentTier_WithBaseAddr() 2015-10-07 16:43:37 +03:00
leventov 03a761ebbf Zero out last allocated tier in VanillaChronicleHash.allocateTierBulk() 2015-10-07 16:14:28 +03:00
leventov 70fb0dd2fa Refactor: move initUsed(true) inside ChainingInterface.getContext() 2015-10-06 18:59:54 +03:00
leventov b967ad14d4 Factor Alloc stage abstraction, make alloc which fails to allocate in the current tier search from the first tier in chain 2015-10-06 18:59:45 +03:00
leventov 8cb556a9c1 Change allocatedExtraTierBulks counter type from long to int 2015-10-05 10:57:42 +03:00
leventov 0adaaa935d Store segmentHeadersOffset in global Mutable state (according to spec; to make ChronicleMap file independent from current mapping page size) 2015-10-05 10:15:31 +03:00
leventov 1384a56768 Align tier bulks to mapping granularity, instead of page boundary 2015-10-05 10:12:39 +03:00
leventov 21d5342048 Optimize memory in non-persisted case -- don't zero out all memory eagery; remove global mutable state duplication code in VanillaChronicleHash and ReplicatedChronicleMap; make sleep interruptible in OldEntriesCleanup thread; remove old entries cleanup duplicatation tracking; fix bugs 2015-10-04 07:57:12 +03:00
leventov 01c81ca05b Add debug logging of extra tier allocation 2015-10-03 05:18:08 +03:00
leventov 901ce5b55c Fix a bug: tierIndexOffsetWithinBulk extracted from position in tier bulk changes bitset was used as tierIndexMinusOne; refactorings 2015-10-03 05:14:38 +03:00