What's New in beta-20160829

Aug 29, 2016

Get future release notes emailed to you:


General Changes

  • CockroachDB now uses Go 1.7. #8579
  • CockroachDB now uses RocksDB 4.9. #8815

Command-Line Interface Changes

  • The cockroach zone set command no longer takes its input directly on the command line. Instead, it accepts a --file flag to read from a file, or --file=- to read from standard input. #7953

New Features

  • AS OF SYSTEM TIME can now be used with JOIN queries. #8198
  • The type BIT now works correctly (as a shorthand for BIT(1)). The length limit is now enforced. #8326
  • The SHOW commands now only show tables that the current user has permission to access. #8070
  • Foreign keys can now use a prefix of an index. #8059
  • The standard information_schema database is now partially supported. New tables will be added to this database in future beta releases. #7965, #8119

Bug Fixes

  • Clusters with a large number of ranges no longer experience persistently broken connections on node restarts. #8828
  • The RENAME command now requires the DROP privilege on the table or database. It is no longer possible to rename the system database. #7998
  • The repeat() and substr() functions with very large numeric arguments will no longer crash the server as easily. #8073, #8078
  • Certain errors now cause a replica to be marked as corrupt so the corruption will not be replicated to other nodes. #7684
  • A CloseComplete packet is now sent in response to a Close command, improving compatibility with come PostgreSQL client drivers (including the Elixir driver). #8309
  • SQL name parsing has been improved to fix several panics and allow foreign key constraints that span databases. #8152
  • Fixed some panics. #8283, #8282
  • Fixed a deadlock if a range grew too large. #8387
  • Fixed a race in which multiple conflicting snapshots could be accepted at the same time. #8365
  • Decimal values are now represented correctly in the binary protocol. #8319

Performance Improvements

  • Snapshots are now sent synchronously during replica changes. This controls the rate of the replication process and prevents spikes in memory usage that often caused servers to crash. #8613
  • Raft logs are now truncated less aggressively, reducing the chance that replication will need to send a snapshot instead of the log. #8343, #8629, #8656
  • Snapshots and the raft log are now written more efficiently. #8644
  • Raft log entries are now cached. #8494
  • Raft groups are now created lazily at startup. #8592
  • Raft heartbeats are now sent less often by default. #8695
  • Improved performance and reliability of clusters with large numbers of ranges. #8125, #8162, #8495
  • The heuristics used by the rebalancing system have been improved. #8124
  • Some noisy log messages have been removed or reduced. #8071, #8021, #8240, #8292, #8529, #8687, #8689
  • The gossip network reconnects more reliably after a failure. #8128
  • RPC connections to failed nodes are now detected sooner. #8163
  • The cache of the first range descriptor is now properly invalidated. #8163
  • Removed replicas are now garbage-collected sooner in many cases. #8172
  • The resolution of the block profiler has been reduced, saving CPU. #8384
  • Range lease extensions no longer block concurrent reads. #8352

UI Changes

  • The "capacity used" is now computed correctly. #8048
  • The CPU and garbage-collection graphs now display averages. #8048
  • The Databases section now includes more details. #8364

Doc Updates


This release includes 280 merged PRs by 26 authors. We would like to thank the following contributors from the CockroachDB community:

  • Christian Koep
  • Dolf Schimmel
  • songhao
YesYes NoNo