Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
502 lines (443 sloc) 28.5 KB

TiKV Change Log

All notable changes to this project are documented in this file. See also TiDB Changelog and PD Changelog.

[Unreleased]

[3.0.0]

  • Engine

    • Introduce Titan, a key-value plugin that improves write performance for scenarios with value sizes greater than 1KiB, and relieves write amplification in certain degrees
    • Optimize memory management to reduce memory allocation and copying for Iterator Key Bound Option
    • Support block cache sharing among different column families
  • Server

    • Support reversed raw_scan and raw_batch_scan
    • Support batch receiving and sending Raft messages, improving TPS by 7% for write intensive scenarios
    • Support getting monitoring information via HTTP
    • Support Local Reader in RawKV to improve performance
    • Reduce context switch overhead from batch commands
  • Raftstore

    • Support Multi-thread Raftstore and Multi-thread Apply to improve scalabilities, concurrency capacity, and resource usage within a single node. Performance improves by 70% under the same level of pressure
    • Support checking RocksDB Level 0 files before applying snapshots to avoid write stall
    • Support Hibernate Regions to optimize CPU consumption from RaftStore (Experimental)
    • Remove the local reader thread
  • Transaction

    • Support distributed GC and concurrent lock resolving for improved GC performance
    • Support the pessimistic transaction model (Experimental)
    • Modify the semantics of Insert to allow Prewrite to succeed only when there is no Key
    • Remove txn scheduler
    • Add monitoring items related to read index and GC worker
  • Coprocessor

    • Refactor the computation framework to implement vector operators, computation using vector expressions, and vector aggregations to improve performance
    • Support providing operator execution status for the EXPLAIN ANALYZE statement in TiDB
    • Switch to the work-stealing thread pool model to reduce context switch cost
  • Misc

    • Develop a unified log format specification with restructured log system to facilitate collection and analysis by tools
    • Add performance metrics related to configuration information and key bound crossing.

[3.0.0-rc.3]

  • Engine

    • Check iterator status when scanning. 4936
    • Fix the issue that ingested files and directory are not synchronized. 4937
  • Server

    • Sanitize block size configuration. 4928
    • Support replicating the delete_range request without deleting the data when applying. 4490
    • Add read index related metrics. 4830
    • Add GC worker related metrics. 4922
  • Raftstore

    • Fix the issue that local reader cache is not cleared correctly. 4778
    • Fix request latency jetter when transferring leader and conf changes. 4734
    • Remove invalid empty callbacks. 4682
    • Clear stale reads after role change. 4810
    • Synchronize all CF files for the received snapshots. 4807
    • Fix missing fsync calls for snapshots. 4850
  • Coprocessor

    • Improve coprocessor batch executor. 4877
  • Transaction

    • Support ResolveLockLite to allow only resolving specified lock keys. 4882
    • Improve pessimistic lock transaction. 4889
  • Tikv-ctl

    • Improve bad-regions and tombstone subcommands. 4862
  • Misc

    • Add dist_release. 4841

[3.0.0-rc.2]

  • Engine

    • Support multiple column families sharing a block cache #4563
  • Server

    • Remove TxnScheduler #4098
    • Support pessimistic lock transactions #4698
  • Raftstore

    • Support hibernate Regions to reduce the consumption of the raftstore CPU #4591
    • Fix the issue that the leader does not reply to the ReadIndex requests for the learner #4653
    • Fix the issue of transferring leader failure in some cases #4684
    • Fix the possible dirty read issue in some cases #4688
    • Fix the issue that a snapshot lacks data in some cases #4716
  • Coprocessor

    • Add more RPN functions
      • LogicalOr #4691
      • LTReal #4602
      • LEReal #4602
      • GTReal #4602
      • GEReal #4602
      • NEReal #4602
      • EQReal #4602
      • IsNull #4720
      • IsTrue #4720
      • IsFalse #4720
      • Support comparison arithmetic for Int #4625
      • Support comparison arithmetic for Decimal #4625
      • Support comparison arithmetic for String #4625
      • Support comparison arithmetic for Time #4625
      • Support comparison arithmetic for Duration #4625
      • Support comparison arithmetic for Json #4625
      • Support plus arithmetic for Int #4733
      • Support plus arithmetic for Real #4733
      • Support plus arithmetic for Decimal #4733
      • Support MOD functions for Int #4727
      • Support MOD functions for Real #4727
      • Support MOD functions for Decimal #4727
      • Support minus arithmetic for Int #4746
      • Support minus arithmetic for Real #4746
      • Support minus arithmetic for Decimal #4746

[3.0.0-rc.1]

  • Engine

    • Fix the issue that may cause incorrect statistics on read traffic #4436
    • Fix the issue that may cause prefix extractor panic when deleting a range #4503
    • Optimize memory management to reduce memory allocation and copying for Iterator Key Bound Option #4537
    • Fix the issue that failing to consider learner log gap may in some cases cause panic #4559
    • Support block cache sharing among different column families#4612
  • Server

    • Reduce context switch overhead of batch commands #4473
    • Check the validity of seek iterator status #4470
  • RaftStore

    • Support configurable properties index distance #4517
  • Coprocessor

    • Add batch index scan executor #4419
    • Add vectorized evaluation framework #4322
    • Add execution summary framework for batch executors #4433
    • Check the maximum column when constructing the RPN expression to avoid invalid column offset that may cause evaluation panic #4481
    • Add BatchLimitExecutor #4469
    • Replace the original futures-cpupool with tokio-threadpool in ReadPool to reduce context switch #4486
    • Add batch aggregation framework #4533
    • Add BatchSelectionExecutor #4562
    • Add batch aggression function AVG #4570
    • Add RPN function LogicalAnd#4575
  • Misc

    • Support tcmalloc as a memory allocator #4370

[3.0.0-beta.1]

  • Optimize the Coprocessor calculation execution framework and implement the TableScan section, with the Single TableScan performance improved by 5% ~ 30%
    • Implement the definition of the BatchRows row and the BatchColumn column #3660
    • Implement VectorLike to support accessing encoded and decoded data in the same way #4242
    • Define the BatchExecutor to interface and implement the way of converting requests to BatchExecutor #4243
    • Implement transforming the expression tree into the RPN format #4329
    • Implement the BatchTableScanExecutor vectorization calculation operator #4351
  • Unify the log format for easy collection and analysis by tools
  • Support using the Local Reader to read in the Raw Read interface #4222
  • Add metrics about configuration information #4206
  • Add metrics about key exceeding bound #4255
  • Add an option to control panic or return an error when encountering the key exceeding bound error #4254
  • Add support for the INSERT operation, make prewrite succeed only when keys do not exist, and eliminate Batch Get #4085
  • Use more fair batch strategy in the Batch System #4200
  • Support Raw scan in tikv-ctl #3825
  • Support hibernating regions #4591

[3.0.0-beta]

  • Support distributed GC #3179
  • Check RocksDB Level 0 files before applying snapshots to avoid Write Stall #3606
  • Support reverse raw_scan and raw_batch_scan #3724
  • Support using HTTP to obtain monitoring information #3855
  • Support DST better #3786
  • Support receiving and sending Raft messages in batch #3913
  • Introduce a new storage engine Titan #3985
  • Upgrade gRPC to v1.17.2 #4023
  • Support receiving the client requests and sending replies in batch #4043
  • Support multi-thread Apply #4044
  • Support multi-thread Raftstore #4066

[2.1.2]

  • Support the configuration format in the unit of DAY (d) and fix the configuration compatibility issue #3931
  • Fix the possible panic issue caused by Approximate Size Split #3942
  • Fix two issues about Region merge #3822, #3873

[2.1.1]

  • Avoid transferring the leader to a newly created peer, to optimize the possible delay #3878

[2.1.0]

[2.1.0-rc.5]

  • Improve the error message of WriteConflict #3750
  • Add the panic mark file #3746
  • Downgrade grpcio to avoid the segment fault issue caused by the new version of gRPC #3650
  • Add the upper limit to the kv_scan interface #3749

[2.1.0-rc.4]

  • Optimize the RocksDB Write stall issue caused by applying snapshots #3606
  • Add raftstore tick metrics #3657
  • Upgrade RocksDB and fix the Write block issue and that the source file might be damaged by the Write operation when performing IngestExternalFile #3661
  • Upgrade grpcio and fix the issue that “too many pings” is wrongly reported #3650

[2.1.0-rc.3]

Performance

  • Optimize the concurrency for coprocessor requests #3515

New features

  • Add the support for Log functions #3603
  • Add the support for the sha1 function #3612
  • Add the support for the truncate_int function #3532
  • Add the support for the year function #3622
  • Add the support for the truncate_real function #3633

Bug fixes

  • Fix the reporting error behavior related to time functions #3487, #3615
  • Fix the issue that the time parsed from string is inconsistent with that in TiDB #3589

[2.1.0-rc.2]

Performance

  • Support splitting Regions based on statistics estimation to reduce the I/O cost #3511
  • Reduce clone in the transaction scheduler #3530

Improvements

  • Add the pushdown support for a large number of built-in functions
  • Add the leader-transfer-max-log-lag configuration to fix the failure issue of leader scheduling in specific scenarios #3507
  • Add the max-open-engines configuration to limit the number of engines opened by tikv-importer simultaneously #3496
  • Limit the cleanup speed of garbage data to reduce the impact on snapshot apply #3547
  • Broadcast the commit message for crucial Raft messages to avoid unnecessary delay #3592

Bug fixes

  • Fix the leader election issue caused by discarding the PreVote message of the newly split Region #3557
  • Fix follower related statistics after merging Regions #3573
  • Fix the issue that the local reader uses obsolete Region information #3565
  • Support UnsafeDestroyRange API to speedup garbage data cleaning after table/index has been truncated/dropped #3560

[2.1.0-rc.1]

Features

  • Support batch split to avoid too large Regions caused by the Write operation on hot Regions
  • Support splitting Regions based on the number of rows to improve the index scan efficiency

Performance

  • Use LocalReader to separate the Read operation from the raftstore thread to lower the Read latency
  • Refactor the MVCC framework, optimize the memory usage and improve the scan Read performance
  • Support splitting Regions based on statistics estimation to reduce the I/O usage
  • Optimize the issue that the Read performance is affected by continuous Write operations on the rollback record
  • Reduce the memory usage of pushdown aggregation computing

Improvements

  • Add the pushdown support for a large number of built-in functions and better charset support
  • Optimize the GC workflow, improve the GC speed and decrease the impact of GC on the system
  • Enable prevote to speed up service recovery when the network is abnormal
  • Add the related configuration items of RocksDB log files
  • Adjust the default configuration of scheduler_latch
  • Support setting whether to compact the data in the bottom layer of RocksDB when using tikv-ctl to compact data manually
  • Add the check for environment variables when starting TiKV
  • Support dynamically configuring the dynamic_level_bytes parameter based on the existing data
  • Support customizing the log format
  • Integrate tikv-fail in tikv-ctl
  • Add I/O metrics of threads

Bug fixes

  • Fix decimal related issues
  • Fix the issue that gRPC max_send_message_len is set mistakenly
  • Fix the issue caused by misconfiguration of region_size

[2.1.0-beta]

Features

  • Upgrade Rust to the nightly-2018-06-14 version
  • Provide a Raft PreVote configuration to avoid leader reelection generated when network recovers after network isolation
  • Add a metric to display the number of files and ingest related information in each layer of RocksDB
  • Print key with too many versions when GC works

Performance

  • Use static metric to optimize multi-label metric performance (YCSB raw get is improved by 3%)
  • Remove box in multiple modules and use patterns to improve the operating performance (YCSB raw get is improved by 3%)
  • Use asynchronous log to improve the performance of writing logs
  • Add a metric to collect the thread status
  • Decease memory copy times by decreasing box used in the application to improve the performance

[2.0.4]

Features

  • Add the RocksDB PerfContext interface for debugging
  • Add the region-properties command for tikv-ctl

Improvements

  • Make GC record the log when GC encounters many versions of data
  • Remove the import-mode parameter

Bug Fixes

  • Fix the issue that reverse-seek is slow when many RocksDB tombstones exist
  • Fix the crash issue caused by do_sub

[2.0.3]

Bug Fixes

  • Correct wrong peer meta for learners
  • Report an error instead of getting a result if divisor/dividend is 0 in do_div_mod

[2.0.2]

Improvements

  • Support configuring more gRPC related parameters
  • Support configuring the timeout range of leader election

Bug Fixes

  • Fix the issue that the Raft log is not printed
  • Fix the issue that obsolete learner is not deleted
  • Fix the issue that the snapshot intermediate file is mistakenly deleted

[2.0.1]

Performance

  • Reduced number of thread_yield calls
  • Fix the issue that SELECT FOR UPDATE prevents others from reading

Improvements

  • More verbose logs for slow query
  • Speed up delete range

Bug Fixes

  • Fix the bug that raftstore is accidentally blocked when generating the snapshot
  • Fix the issue that Learner cannot be successfully elected in special conditions
  • Fix the issue that split might cause dirty read in extreme conditions
  • Correct the default value of the read thread pool configuration

[2.0.0] - 2018-04-27

Features

  • Protect critical configuration from incorrect modification
  • Support Region Merge [experimental]
  • Add the Raw DeleteRange API
  • Add the GetMetric API
  • Add Raw Batch Put, Raw Batch Get, Raw Batch Delete and Raw Batch Scan
  • Add Column Family options for the RawKV API and support executing operation on a specific Column Family
  • Support Streaming and Streaming Aggregation in Coprocessor
  • Support configuring the request timeout of Coprocessor
  • Carry timestamps with Region heartbeats
  • Support modifying some RocksDB parameters online, such as block-cache-size
  • Support configuring the behavior of Coprocessor when it encounters some warnings or errors
  • Support starting in the importing data mode to reduce write amplification during the data importing process
  • Support manually splitting Region in halves
  • Improve the data recovery tool tikv-ctl
  • Return more statistics in Coprocessor to guide the behavior of TiDB
  • Support the ImportSST API to import SST files [experimental]
  • Add the TiKV Importer binary to integrate with TiDB Lightning to import data quickly [experimental]

Performance

  • Optimize read performance using ReadPool and increase the raw_get/get/batch_get by 30%
  • Improve metrics performance
  • Inform PD immediately once the Raft snapshot process is completed to speed up balancing
  • Solve performance jitter caused by RocksDB flushing
  • Optimize the space reclaiming mechanism after deleting data
  • Speed up garbage cleaning while starting the server
  • Reduce the I/O overhead during replica migration using DeleteFilesInRanges

Stability

  • Fix the issue that gRPC call does not returned when the PD leader switches
  • Fix the issue that it is slow to offline nodes caused by snapshots
  • Limit the temporary space usage consumed by migrating replicas
  • Report the Regions that cannot elect a leader for a long time
  • Update the Region size information in time according to compaction events
  • Limit the size of scan lock to avoid request timeout
  • Limit the memory usage when receiving snapshots to avoid OOM
  • Increase the speed of CI test
  • Fix the OOM issue caused by too many snapshots
  • Configure keepalive of gRPC
  • Fix the OOM issue caused by an increase of the Region number

[2.0.0-rc6] - 2018-04-19

Improvements

  • Reduce lock contention in Worker
  • Add metrics to the FuturePool

Bug Fixes

  • Fix misused metrics in Coprocessor

[2.0.0-rc.5] - 2018-04-17

New Features

  • Support compacting Regions in tikv-ctl
  • Add raw batch put/get/delete/scan API for TiKV service
  • Add ImportKV service
  • Support eval error in Coprocessor
  • Support dynamic adjustment of RocksDB cache size by tikv-ctl
  • Collect number of rows scanned for each range in Coprocessor
  • Support treating overflow as warning in Coprocessor
  • Support learner in raftstore

Improvements

  • Increase snap GC timeout

[2.0.0-rc.4] - 2018-04-01

New Features

  • Limit the memory usage during receiving snapshots, to avoid OOM in extreme conditions
  • Support configuring the behavior of Coprocessor when it encounters warnings
  • Support importing the data pattern in TiKV
  • Support splitting Region in the middle

Improvements

  • Fix the issue that too many logs are output caused by leader missing when TiKV is isolated
  • Use crossbeam channel in worker

[2.0.0-rc.3] - 2018-03-23

New Features

  • Support Region Merge
  • Add the Raw DeleteRange API
  • Add the GetMetric API
  • Support streaming in Coprocessor
  • Support modifying RocksDB parameters online

Improvements

  • Inform PD immediately once the Raft snapshot process is completed, to speed up balancing
  • Reduce the I/O fluctuation caused by RocksDB sync files
  • Optimize the space reclaiming mechanism after deleting data
  • Improve the data recovery tool tikv-ctl
  • Fix the issue that it is slow to make nodes down caused by snapshot
  • Increase the raw_get/get/batch_get by 30% with ReadPool
  • Support configuring the request timeout of Coprocessor
  • Carry time information in Region heartbeats
  • Limit the space usage of snapshot files to avoid consuming too much disk space
  • Record and report the Regions that cannot elect a leader for a long time
  • Speed up garbage cleaning when starting the server
  • Update the size information about the corresponding Region according to compaction events
  • Limit the size of scan lock to avoid request timeout
  • Use DeleteRange to speed up Region deletion

[2.0.0-rc.2] - 2018-03-15

New Features

  • Implement IngestSST API
  • tikv-ctl now can send consistency-check requests to TiKV
  • Support dumping stats of RocksDB and malloc in tikv-ctl

Improvements

  • Reclaim disk space after data have been deleted

[2.0.0-rc.1] - 2018-03-09

New Features

  • Protect important configuration which cannot be changed after initial configuration
  • Check whether SSD is used when you start the cluster

Improvements

  • Fix the issue that gRPC call is not cancelled when PD leaders switch
  • Optimize the read performance using ReadPool, and improve the performance by 30% for raw get
  • Improve metrics and optimize the usage of metrics

[1.1.0-beta] - 2018-02-24

Improvements

  • Traverse locks using offset + limit to avoid potential GC problems
  • Support resolving locks in batches to improve GC speed
  • Support GC concurrency to improve GC speed
  • Update the Region size using the RocksDB compaction listener for more accurate PD scheduling
  • Delete the outdated data in batches using DeleteFilesInRanges, to make TiKV start faster
  • Configure the Raft snapshot max size to avoid the retained files taking up too much space
  • Support more recovery operations in tikv-ctl
  • Optimize the ordered flow aggregation operation

[1.0.8] - 2018-02-11

Improvements

  • Use DeleteFilesInRanges to clear stale data and improve the TiKV starting speed
  • Sync the metadata of the received Snapshot compulsorily to ensure its safety

Bug Fixes

  • Use Decimal in Coprocessor sum

[1.0.7] - 2018-01-22

Improvements

  • Support key-only option in Table Scan executor
  • Support the remote mode in tikv-ctl
  • Fix the loss of scheduling command from PD

Bug Fixes

  • Fix the format compatibility issue of tikv-ctl proto
  • Add timeout in Push metric

[1.1.0-alpha] - 2018-01-19

New Features

  • Support Raft learner
  • Support TLS

Improvements

  • Optimize Raft Snapshot and reduce the I/O overhead
  • Optimize the RocksDB configuration to improve performance
  • Optimize count (*) and query performance of unique index in Coprocessor
  • Solve the reconnection issue between PD and TiKV
  • Enhance the features of the data recovery tool tikv-ctl
  • Support the Delete Range feature
  • Support splitting according to table in Regions
  • Support setting the I/O limit caused by snapshot
  • Improve the flow control mechanism
You can’t perform that action at this time.