- [GH 2323] Improve raft stability by making peer communication timeouts configurable
- [GH 2650] Improve error messages when the etcdctl cluster-health command cannot gather cluster health
- [GH 2877] etcd can now recover after bootstrap related errors
- [GH 2884] Improve performance under high load by reusing network connections when running in proxy mode
- [GH 2902] etcd no longer sends invalid requests to the active etcd cluster members when running in proxy mode.
- [GH 2916] The etcd build script now works outside of a git repository, but the version info in the resulting binary will not contain a valid Git SHA.
- [GH 2947] The -advertise-client-urls flag is no longer required when running in proxy mode

- [GH2689] fix the bug that newly joined member may panic
- [GH2705] fix the bug that stats may be updated before it is initialized
- [GH2749] fix the typo in the string representation of raftLog
- [GH2743] fix connection leak in proxy mode
- [GH2751 GH2811] support filelock for solaris and plan9
- [GH2756] improve the reliability of `etcdctl member remove`
- [GH2758] improve the disaster recovery guide
- [GH2761] demand users to set -advertise-client-urls explicitly for better usage
- [GH2560] print out more version informantion

v2.1.0-alpha.0

- proxy shuffles endpoints to avoid being stuck to one member
- stop using resolved address to avoid TLS failure
- fix the bug that machine power-off might lead to corrupted wal
- fix the bug that etcdserver.Send does not print correct error log for messages to removed members
- etcdctl backup tool uses new file layout
- fix the bug that PUT request with prevExist=true and prevIndex/prevValue set is not parsed to CompareAndSwap
- fix the bug that etcd failed to recovery from recoverable WALs
- improve etcdctl help doc by adding ‘extended’ to output format

- etcd watchers no longer hang under when many watchers watch the same key
- etcdctl now support import hidden keys from the given snapshot

- add 'etcdctl import' command to import the migration snap generated in v0.4.8 etcd to 2.0 etcd cluster
- etcdctl takes an additional 2379, which is the IANA port of etcd client URL, as the default port
- etcd no longer builds debugging tools into bin/ directory

- revert allow_legacy related stuffs
- introduce /v2/migration/snapshot endpoint

- fix the bug that wal file is not removed in time
- support bootstrap etcd using non-empty data directory
- check given flags when join existing cluster
- check that etcd parses all arguments

- support IPv6 address in url flags
- allow user provide a name without initial-cluster update
- restrict that election timeout should be much bigger than heartbeat internval
- improve the documentation about unsafe flags
- fix the documentation about starting raft node
- fix the bug that proxy fails to restart
- fix the bug when cloning store stats
- remove starter code

- add --internal-dir flag to run legacy mode
- fix the config documentation about --bind-addr and --peer-bind-addr

- Fix a minor voting issue in raft. Under unique scenarios it was possible that two nodes in a raft cluster could be elected leader for a specific term.
- Fix unexpected HTTPS to HTTP downgrade when TLS certs and keys are not provided.
- Fix an early file unlock issue in WAL. This fix was done to improve safety of the etcd WAL implementation in preparation for the etcd 2.1.0 release.
- Fix frequent i/o timeout issues during peer communication.
- Improved documentation around setting up an etcd cluster.

v2.0.4 - jed version

- fix build issue on windows
- fix variable shadowing
- add document about running etcd in docker
- fix the bug that etcd fails to be killed when its pid equals to 1
- fix the bug that etcd cluster recovers to old status

- Fix static building of etcd binary
- Fix the bug of removing user data under /_etcd in migration
- Fix the bug of saving snapshot panic
- Fix the bug of recovering transport wrongly
- Fix the bug of skipping cluster check in discovery bootstrap
- Add support for graceful shutdown when receives SIGINT and SIGTERM

Reinstating tag v2.0.1

Fixes in 2.0.2:
- Upgrade from 0.4.6 to 2.0.2 directly works

2.0.2 replaces 2.0.1, which contained
- Add support for cluster-level upgrading from etcd 0.4.7 to etcd 2.0.1
- Add cluster-health command to etcdctl
- Add cluster tracking to proxy
- Add a rafttest pkg for raft level failure injection tests
- Change the data directory structure to /{data-dir}/member and /{data-dir}/proxy
- Fix the bug of writing zero length entries message in rafthttp streamer
- Fix a snapshot index bug in store that might lead to data inconsistency

- support allow_legacy mode for smooth upgrade

v2.0.0

- fully support and document DNS discovery using flag `--discovery-srv`
- additional tests for all packages
- fix the message flood from leader to follower when follower is unhealthy
- support scaling from two member clusters to single member clusters (not a recommended configuration)
- remove out of date documentation about the 0.4 releases
- use tls information internally when finding cluster membership
- WAL migration starts the term at 1 instead of 0
- add debug tool etcd-dump-logs
- update all documents to fit the v2.0 changes
- add `-heartbeat-interval` and `-election-timeout` flags

Thank you to: 周春寅, efrecon, Yicheng Qin, Xiang Li, Shota Fukumori (sora_h),
Rob Szumski, Peter Suschlik, Peter Lemenkov, Matthew Williams, Kelsey
Hightower, Jörg Thalheim, Jurriaan Mous, Jonathan Boulle, Brian Waldon, Brandon
Philips, Ben Darnell, Barak Michener, Andrew Bonventre, Alban Crequy

v2.0.0-rc.1

- Documentation fixes preparing for rc
- etcdctl now has env variables for x509 certs
- Testing fixes for speed and reliability

Thanks to everyone who helped with this release: Barak Michener, Brandon
Philips, Jonathan Boulle, José Luis Salvador Rufo, Kelsey Hightower, Vipin
Jain, Xiang Li, Yicheng Qin, lisael

v0.5.0-alpha.5

- extract storage interface in raft package
- print out more logs about raft and etcdserver behaviors
- support purging old WAL and snapshot files automatically
- optimize state machine processing in good path
- fix restore snapshot bug in raft
- fix the bug that may read empty state from WAL
- document how to change peer urls
- document the way to migrate member
- add external etcd-lock(implemented by ravigadde) link to the tool doc

Thank you to: Barak Michener, Ben Darnell, Brandon Philips, Brian Waldon, Johan
Euphrosine, ravigadde, Rob Szumski, Veres Lajos, Xiang Li, Yicheng Qin

v0.5.0-alpha.4