Skip to content

Latest commit

 

History

History
201 lines (168 loc) · 9.57 KB

RELEASES.md

File metadata and controls

201 lines (168 loc) · 9.57 KB
Raw

Versioning

LogCabin uses SemVer for its version numbers. Its "public API" consists of many components, including its network and disk formats, its client library, and the command line arguments for various executables. These components are all released together under a single version number, and the release notes below describe which components have actually changed.

Release Process

See RELEASE-PROCESS.md.

Version 1.2.0-alpha.0 (In Development)

Internal improvements:

  • #200: reset leader election timeout in follower after disk io completes

New backwards-compatible changes:

  • Added new API getConfiguration2, which behaves as getConfiguration but allows a timeout. The API returns a the configuration plus a status code to allow for a TIMEOUT response.
  • Added companion getConfiguration2Ex that behaves as getConfiguration2 but throws exceptions.
  • Added setConfiguration2 which behaves as setConfiguration but allows a timeout.
  • Added companion setConfiguration2Ex that behaves as setConfiguration2 throws exceptions.
  • See #184 for details

Version 1.1.0 (2015-07-26)

This release brings many bug fixes and improvements since v1.0.0. All users are strongly encouraged to upgrade.

Bug fixes (high severity):

  • Fixes packaging up very large AppendEntries requests. Before, it was possible for a leader to send a non-contiguous list of entries to the follower, and the follower would end up with a corrupt log (issue #160). Before, it was also possible for packing up the requests to take so long as to cause availability and performance problems (issue #161).
  • Fixes occasional hang when exiting due to unsafe access of a boolean flag (issue #144).
  • Fixes hang when exiting while removing servers from the cluster configuration (issue #183).
  • Fixes client waiting past its timeout on another client's connection attempt (issue #173).

Bug fixes (low severity):

  • Fixes Core::Debug::DebugMessage move constructor, where processName and threadName were not moved over as they should have been (git 77d7f6b).
  • Fixes signed integer overflow bug under aggressive optimizing compilers affecting SteadyTimeConverter, which is only used in producing ServerStats (git 6473400). Turns on -fno-strict-overflow compiler setting.
  • Fixes repeated PANIC in InstallSnapshot RPC after a server restarts while receiving a snapshot (issue #174). This could result in a temporary availability issue that would resolve itself on the next term change.
  • Fixes failing conditional tree operations after setting a condition with a relative path (issue #177).
  • Fixes event loop thread dumping server stats to the debug log, which had the potential for delays and deadlock (we never saw deadlock occur in practice, however). Now the stats are dumped from a separate thread (issue #159).
  • Fixes PANIC due to "No route to host" error after many minutes of the network interface going down (issue #154). This is considered low severity as it was unlikely to affect overall cluster availability.

Internal improvements:

  • Adds gcc 5.1 (which required no changes; issue #141) and clang 3.4, 3.5, 3.6, and 3.7 (issue #9) as supported compilers.
  • liblogcabin.a is now compiled with -fPIC, so it can be linked into shared object (.so) files (git 1ca169c).
  • Optimizes setting nextIndex on leaders by capping it to just past the follower's last log index. This helps with followers that are new or have fallen far behind (git fcbacbb).
  • Clients now make a best effort attempt to close their sessions when they shut down gracefully (issue #116). Before, client sessions were only ever expired after a timeout. This state could accumulate quickly when running short-lived clients in a tight loop. Enabling this change requires all servers to be updated (so that the state machine is updated); new clients talking to old clusters will issue a warning that they are unable to close their sessions.
  • SegmentedLog now coalesces back-to-back fdatasync calls, making batch log appends much more efficient (issue #165).
  • Leaders will now limit the amount of data they send to a follower when their connection to that follower is lost, which reduces wasted bandwidth (git e238fa6).

New backwards-compatible changes:

  • LogLevel, getLogFilename, setLogFilename, reopenLogFromFilename, getLogPolicy, logPolicyFromString, logPolicyToString were introduced in include/LogCabin/Debug.h.
  • The LogCabin daemon will now reopen its log file on SIGUSR2 (useful for log rotation; issue #150). Signal handling was not listed as part of LogCabin's public API until now; signals listed in --help messages are now subject to semantic versioning.
  • The build/Examples/ServerControl or /usr/bin/logcabinctl program can be used to inspect and manipulate an individual server's state (issue #151). Its command line is now part of LogCabin's public API.
  • All clients now have --verbose and --verbosity to control the debug log level and policy (issue #153). The server's config file now has a new option logPolicy to control the same.
  • Exceptions due to bad user input are now caught by broad exception handlers. They were uncaught before, causing the process to abort() and create unnecessary core files (issue #166).
  • Adds several new ServerStats metrics.

Changes to unstable APIs:

  • build/Examples/ServerStats or /usr/bin/logcabin-serverstats along with scripts/serverstats.py have been removed. The logcabinctl program can now be used to fetch the stats instead, and the Python wrapper wasn't kept up-to-date anyhow. External clients linked to old versions of LogCabin::Client::Cluster::getServerStats() will not work with new servers, and logcabinctl stats get and other clients linked to the same function will not work with old servers.
  • Examples/HelloWorld is no longer installed to /usr/bin/logcabin-helloworld.
  • Fixes RPM build on Scons 2.3.0 (git e77d217).
  • Removes the 'reload' command from the Red Hat init script, which would previously kill the server (git 28044cb).
  • Changes the log path in the Red Hat init script to /var/log/logcabin/logcabin.log (git e9e466f).
  • Changes the RPM package and Red Hat init script to launch the LogCabin server as user logcabin instead of root (issue #178). Changes the storage path.
  • The client library in testing mode (MockClientImpl) will now return obvious timeout errors, making it easier to unit test client code (git a8b22be).

Version 1.0.0 (2015-04-29)

This is the first stable release of LogCabin. We encourage others to try this release out, and we believe it to be ready for production use. As it is the very first release, users are advised to check back frequently in case serious bugs are found.

The public API with respect to versioning consists of the following:

  • include/LogCabin/Client.h: API
  • include/LogCabin/Debug.h: API
  • build/LogCabin or /usr/bin/logcabind: command line
  • build/Examples/Reconfigure or /usr/bin/logcabin-reconfigure: command line
  • build/Examples/TreeOps or /usr/bin/logcabin: command line
  • config file format and options: defined by sample.conf
  • client-to-server network protocol: compatibility
  • server-to-server network protocol: compatibility
  • replicated state machine behavior: compatibility
  • storage layout on disk: compatibility
  • snapshot format on disk: compatibility
  • log format on disk of Segmented storage module: compatibility

Command line APIs consist of argv, zero vs nonzero exit statuses, and side effects, but not necessarily stdout and stderr. These are documented with -h and --help flags.

The interfaces/protocols indicating "compatibility" are not documented publicly, but interoperability with different versions of the code is maintained. Interoperability with third-party implementations is not guaranteed, as there is no explicit protocol specification.

  • Network protocols indicating "compatibility" provide forwards and backwards compatibility: older code and newer code must be able to interoperate within a MAJOR release. This is desirable in the network protocols to allow non-disruptive rolling upgrades.

  • Disk formats indicating "compatibility" provide backwards compatibility: newer code must be able to accept formats produced by older code within a MAJOR release. However, older code may not be able to accept disk formats produced by newer code. This reflects the expectation that servers will be upgraded monotonically from older to newer versions but never back.

  • The replicated state machine (which contains the core Tree data structure that clients interact with, among other things) provides backwards compatibility for a limited window of time. LogCabin will only update the externally visible behavior of its replicated state machine when all currently known servers support the new version. At that point, servers running the old version of the code may not be able to participate in the cluster (they will most likely PANIC repeatedly until their code is upgraded).

The following are specifically excluded from the public API and are not subject to semantic versioning (they may be added to the public API in future releases):

  • client library ABI
  • scripts/logcabin-init-redhat command line
  • various other scripts
  • build/Examples/ServerStats command line
  • various other build/Examples executables
  • build/Storage/Tool command line
  • ServerStats ProtoBuf fields
  • log format on disk of SimpleFile storage module