14 KiB
0.8.2 (UNRELEASED)
FEATURES:
IMPROVEMENTS:
- agent: Fixed a missing case where gossip would stop flowing to dead nodes for a short while. [GH-451]
- agent: Uses the go-sockaddr library to look for private IP addresses, which prefers non-loopback private addresses over loopback ones when trying to automatically determine the advertise address. [GH-451]
- agent: Properly seeds Go's random number generator using the seed library. [GH-451]
- agent: Serf is now built with Go 1.10.x. [GH-520]
- agent: Improved address comparison during conflict resolution. [GH-433]
- agent: Updated memberlist to latest to pull several cleanups and fixes. [GH-491]
- agent: Improved handling of leave intent messages to make sure they propagate and are processed correctly. [GH-510]
- agent: Added CLI option to disable compression for debugging messages. [GH-529]
- library: Moved close of shutdown channel until after network resorces are released. [GH-453]
- library: Fixed several race conditions with QueryResponse [GH-460]
- library: Made snapshot writing asyncronous and will less aggressive compaction on large clusters to avoid blocking message handler on disk IO [GH-524][GH-525]
BUG FIXES:
- agent: Added defenses against invalid network coordinates with NaN and Inf values. [GH-468]
- agent: Fixed an issue on Windows where "wsarecv" errors were logged when clients accessed the RPC interface. [GH-479]
- agent: Fixed an issue where calling the serf Stats function could result in a deadlock. [Consul Issue 4011]
0.8.1 (February 6, 2017)
IMPROVEMENTS:
- agent: Added support for relaying query responses through N other nodes for redundancy. [GH-439]
- agent: Added the ability to tune the broadcast timeout, which might be necessary in very large clusters that experience very large, simultaneous changes to the cluster. [GH-412]
- agent: Added a checksum to UDP gossip messages to guard against packet corruption. [GH-432]
- agent: Added a short window where gossip will still flow to dead nodes so that they can more quickly refute. [GH-440]
- build: Serf now builds with Go 1.7.5. [GH-443]
0.8 (September 14, 2016)
FEATURES:
- Lifeguard Updates: Implemented a new set of feedback controls for the gossip layer that help prevent degraded nodes that can't meet the soft real-time requirements from erroneously causing flapping in other, healthy nodes. This feature tunes itself automatically and requires no configuration. [GH-394]
IMRPOVEMENTS:
- Modified management of intents to be per-node to avoid intent queue overflow errors in large clusters. [GH-402]
- Joins based on a DNS lookup will use TCP and attempt to join with the full list of returned addresses. [GH-387]
- Serf's Go dependencies are now vendored using govendor. [GH-383]
- Updated all of Serf's dependencies. [GH-387] [GH-401]
- Moved dist build into a Docker container. [GH-409]
BUG FIXES:
- Updated memberlist to pull in a fix for leaking goroutines when performing TCP fallback pings. This affected users with frequent UDP connectivity problems. [GH-381]
0.7 (December 21, 2015)
FEATURES:
- Added new network tomography subsystem that computes network coordinates for
nodes in the cluster which can be used to estimate network round trip times
between any two nodes; exposes new
GetCoordinate
API as as well as a a newserf rtt
command to query RTT interactively
IMPROVEMENTS:
- Added support for configuring query request size and query response size [GH-346]
- Syslog messages are now filtered by the configured log-level
- New
statsd_addr
for sending metrics via UDP to statsd - Added support for sending telemetry to statsite
serf info
command now displays event handlers [GH-312]- Added a
MemberLeave
message to theEventCh
for a force-leave so higher- level applications can handle the leave event - Lots of documentation updates
BUG FIXES:
- Fixed updating cached protocol version of a node when an update event fires [GH-335]
- Fixed a bug where an empty remote state message would cause a crash in
MergeRemoteState
0.6.4 (Febuary 12, 2015)
IMPROVEMENTS:
- Added merge delegate to Serf library to support application specific logic in cluster merging.
SERF_RPC_AUTH
environment variable can be used in place of CLI flags.- Display if encryption is enabled in Serf stats
- Improved
join
behavior when using DNS resolution
BUG FIXES:
- Fixed snapshot file compaction on Windows
- Fixed device binding on Windows
- Fixed bug with empty keyring
- Fixed parsing of ports in some cases
- Fixing stability issues under high churn
MISC:
- Increased user event size limit to 512 bytes (previously 256)
0.6.3 (July 10, 2014)
IMPROVEMENTS:
- Added
statsite_addr
configuration to stream to statsite
BUG FIXES:
- Fixed issue with mDNS flooding when using IPv4 and IPv6
- Fixed issue with reloading event handlers
MISC:
- Improved failure detection reliability under load
- Reduced fsync() use in snapshot file
- Improved snapshot file performance
- Additional logging to help debug flapping
0.6.2 (June 16, 2014)
IMPROVEMENTS:
- Added
syslog_facility
configuration to set facility
BUG FIXES:
- Fixed memory leak in in-memory stats system
- Fixed issue that would cause syslog to deadlock
MISC:
- Fixed missing prefixes on some log messages
- Docs fixes
0.6.1 (May 29, 2014)
BUG FIXES:
-
On Windows, a "failed to decode request header" error will no longer be shown on every RPC request.
-
Avoiding holding a lock which can cause monitor/stream commands to fail when an event handler is blocking
-
Fixing conflict response decoding errors
IMPROVEMENTS:
-
Improved agent CLI usage documentation
-
Warn if an event handler is slow, potentially blocking other events
0.6.0 (May 8, 2014)
FEATURES:
-
Support for key rotation when using encryption. This adds a new
serf keys
command, and a-keyring-file
configuration. Thanks to @ryanuber. -
New
-tags-file
can be specified to persist changes to tags made via the RPC interface. Thanks to @ryanuber. -
New
serf info
command to provide operator debugging information, and to get info about the local node. -
Adding
-retry-join
flag to agent which enables retrying the join until success or-retry-max
attempts have been made.
IMPROVEMENTS:
-
New
-rejoin
flag can be used along with a snapshot file to automatically rejoin a cluster. -
Agent uses circular buffer to invoke handlers, guards against unbounded output lengths.
-
Adding support for logging to syslog
-
The SERF_RPC_ADDR environment variable can be used instead of the
-rpc-addr
flags. Thanks to @lalyos [GH-209]. -
serf query
can now output the results in a JSON format. -
Unknown configuration directives generate an error [GH-186]. Thanks to @vincentbernat.
BUG FIXES:
-
Fixing environmental variables with invalid characters. [GH-200]. Thanks to @arschles.
-
Fixing issue with tag changes with hard restart before failure detection.
-
Fixing issue with reconnect when using dynamic ports.
MISC:
-
Improved logging of various error messages
-
Improved debian packaging. Thanks to @vincentbernat.
0.5.0 (March 12, 2014)
FEATURES:
-
New
query
command provides a request/response mechanism to do realtime queries across the cluster. [GH-139] -
Automatic conflict resolution. Serf will detect name conflicts, and use an internal query to determine which node is in the minority and perform a shutdown. [GH-167] [GH-119]
-
New
reachability
command can be used to help diagnose network and configuration issues. -
Added
member-reap
event to get notified of when Serf removes a failed or left node from the cluster. The reap interval is controlled byreconnect_timeout
andtombstone_timeout
respectively. [GH-172]
IMPROVEMENTS:
-
New Recipes section on the site to share Serf tips. Thanks to @ryanuber. [GH-177]
-
members
command has new-name
filter flag. Thanks to @ryanuber [GH-164] -
New RPC command "members-filtered" to move filtering logic to the agent. Thanks to @ryanuber. [GH-149]
-
reconnect_interval
andreconnect_timeout
can be provided to configure agent behavior for attempting to reconnect to failed nodes. [GH-155] -
tombstone_interval
can be provided to configure the reap time for nodes that have gracefully left. [GH_172] -
Agent can be provided
rpc_auth
config to require that RPC is authenticated. All commands can take a-rpc-auth
flag now. [GH-148]
BUG FIXES:
-
Fixed config folder in Upstart script. Thanks to @llchen223. [GH-174]
-
Event handlers are correctly invoked when BusyBox is the shell. [GH-156]
-
Event handlers were not being invoked with the correct SERF_TAG_* values if tags were changed using the
tags
command. [GH-169]
MISC:
- Support for protocol version 1 (Serf 0.2) has been removed. Serf 0.5 cannot join a cluster that has members running version 0.2.
0.4.5 (February 25, 2014)
FEATURES:
- New
tags
command is available to dynamically update tags without reloading the agent. Thanks to @ryanuber. [GH-126]
IMPROVEMENTS:
-
Upstart receipe logs output thanks to @breerly [GH-128]
-
members
can filter on any tag thanks to @hmrm [GH-124] -
Added vagrant demo to make a simple cluster
-
members
now columnizes the output thanks to @ryanuber [GH-138] -
Agent passes its own environment variables through thanks to @mcroydon [GH-142]
-
-iface
flag can be used to bind to interfaces [GH-145]
BUG FIXES:
-
-config-dir would cause protocol to be set to 0 if there are no configuration files in the directory [GH-129]
-
Event handlers can filter on 'member-update'
-
User event handler appends new line, this was being omitted
0.4.1 (February 3, 2014)
IMPROVEMENTS:
- mDNS service uses the advertise address instead of bind address
0.4.0 (January 31, 2014)
FEATURES:
-
Static
role
has been replaced with dynamic tags. Each agent can have multiple key/value tags associated using-tag
. Tags can be updated using a SIGHUP and are advertised to the cluster, causing themember-update
event to be triggered. [GH-111] [GH-98] -
Serf can automatically discover peers uing mDNS when provided the
-discover
flag. In network environments supporting multicast, no explicit join is needed to find peers. [GH-53] -
Serf collects telemetry information and simple runtime profiling. Stats can be dumped to stderr by sending a
USR1
signal to Serf. Windows users must use theBREAK
signal instead. [GH-103] -
advertise
flag can be used to set an advertise address different from the bind address. Used for NAT traversal. Thanks to @benagricola [GH-93] -
members
command now takes-format
flag to specify either text or JSON output. Fixed by @ryanuber [GH-97]
IMPROVEMENTS:
-
User payload always appends a newline when invoking a shell script
-
Severity of "Potential blocking operation" reduced to debug to prevent spurious messages on slow or busy machines.
BUG FIXES:
-
If an agent is restarted with the same bind address but new name, it will not respond to the old name, causing the old name to enter the
failed
state, instead of having duplicate entries in thealive
state. -
leave_on_interrupt
set to false when not specified, if any config file is provided. This flag is deprecated forskip_leave_on_interrupt
instead. [GH-94]
MISC:
-
-role
configuration has been deprecated in favor of-tag role=foo
. The flag is still supported but will generate warnings. -
Support for protocol version 0 (Serf 0.1) has been removed. Serf 0.4 cannot join a cluster that has members running version 0.1.
0.3.0 (December 5, 2013)
FEATURES:
- Dynamic port support, cluster wide consistent config not necessary
- Snapshots to automaticaly rejoin cluster after failure and prevent replays [GH-84] [GH-71]
- Adding
profile
config to agent, to support WAN, LAN, and Local modes - MsgPack over TCP RPC protocol which can be used to control Serf, send events, and receive events with low latency.
- New
leave
CLI command and RPC endpoint to control graceful leaves - Signal handling is controlable, graceful leave behavior on SIGINT/SIGTERM can be specified
- SIGHUP can be used to reload configuration
IMPROVEMENTS:
- Event handler provides lamport time of user events via SERF_USER_LTIME [GH-68]
- Memberlist encryption overhead has been reduced
- Filter output of
members
using regular expressions on role and status replay_on_join
parameter to control replay withstart_join
monitor
works even if the client is behind a NAT- Serf generates warning if binding to public IP without encryption
BUG FIXES:
- Prevent unbounded transmit queues [GH-78]
- IPv6 addresses can be bound to [GH-72]
- Serf join won't hang on a slow/dead node [GH-70]
- Serf Leave won't block Shutdown [GH-1]
0.2.1 (November 6, 2013)
BUG FIXES:
- Member role and address not updated on re-join [GH-58]
0.2.0 (November 1, 2013)
FEATURES:
- Protocol versioning features so that upgrades can be done safely. See the website on upgrading Serf for more info.
- Can now configure Serf with files or directories of files by specifying
the
-config-file
and/or-config-dir
flags to the agent. - New command
serf force-leave
can be used to force a "failed" node to the "left" state. - Serf now supports message encryption and verification so that it can be used on untrusted networks [GH-25]
- The
-join
flag onserf agent
can be used to join a cluster when starting an agent. [GH-42]
IMPROVEMENTS:
- Random staggering of periodic routines to avoid cluster-wide synchronization
- Push/Pull timer automatically slows down as cluster grows to avoid congestion
- Messages are compressed to reduce bandwidth utilization
serf members
now provides node roles in output- Joining a cluster will no longer replay all the old events by default,
but it can using the
-replay
flag. - User events are coalesced by default, meaning duplicate events (by name) within a short period of time are merged. [GH-8]
BUG FIXES:
- Event handlers work on Windows now by executing commands through
cmd /C
[GH-37] - Nodes that previously left and rejoin won't get stuck in 'leaving' state. [GH-18]
- Fixing alignment issues on i386 for atomic operations [GH-20]
- "trace" log level works [GH-31]
0.1.1 (October 23, 2013)
BUG FIXES:
- Default node name is outputted when "serf agent" is called with no args.
- Remove node from reap list after join so a fast re-join doesn't lose the member.
0.1.0 (October 23, 2013)
- Initial release