Skip to content

Latest commit

 

History

History
257 lines (149 loc) · 7.41 KB

CHANGELOG.md

File metadata and controls

257 lines (149 loc) · 7.41 KB
Raw

Change log

master

0.6.5 (2020-03-29)

  • Change alpine version from edge to latest (@ericmatte)

0.6.4 (2019-09-16)

Now it's possible to access ENV in custom log handlers written in mruby.

0.6.3 (2019-08-05)

It can be used on a RPC server side to access remote connection IP address.

  • Use request id for session uids if available. (@sponomarev)

If a load balancer standing in front of WebSocket server assigns X-Request-ID header, this request ID will be used for session identification.

0.6.2 (2019-03-25)

  • Configure maximum message size via --max_message_size. Defaults to 65536 (64kb).

0.6.1 (2018-12-21) "X-mas time is here again!" 🎅

Go to /health (you can configure the path via --health-path) to see the health message. You can use this endpoint as readiness/liveness check (e.g. for load balancers).

0.6.0 (2018-11-12)

  • Add force termination support (by sending the second signal). (@palkan)

  • Add session ID to outgoing RPC call's metadata. (@palkan)

All RPC calls contains the related session ID to metadata (sid key).

  • Send transmissions to client even if authentication failed. (@palkan)

  • Fix websocket close status to reflect the reason for the closure. (@sponomarev)

  • Reduce binary size. (@sponomarev)

0.6.0-preview6

  • Add experimental mruby integration for custom metrics logging. (@palkan)

NOTE: This feature experimental, i.e. the final API may (and likely will) change.

You can write custom Ruby script to implement statistics logging.

For example, to provide Librato-comatible output you can write a custom formatter like this:

# my-metrics-formatter.rb
# This MetricsFormatter name is required!
module MetricsFormatter
  KEYS = %w(clients_num clients_unique_num goroutines_num)

  # `data` is a Hash containing all the metrics data
  def self.call(data)
    parts = []

    data.each do |key, value|
      parts << "sample##{key}=#{value}" if KEYS.include?(key)
    end

    parts.join(' ')
  end
end

And then use it like this:

anycable-go --metrics_log_formatter="my-metrics-formatter.rb"

>INFO 2018-04-27T14:11:59.701Z sample#clients_num=0 sample#clients_uniq_num=0 sample#goroutines_num=0

0.6.0-preview5

  • [Fixes #34] Fix panic when trying to send to a closed channel. (@palkan)

  • Add --metrics_host option. (@shedimon)

0.6.0-preview4

  • Fixed memory leak (caused by leaking mutex goroutines).

  • Other minor fixes.

0.6.0-preview2

  • Add --metrics_log_interval option. (@palkan)

  • Add HTTP /metrics endpoint. (@palkan)

Serves Prometheus-formatted metrics.

Enabled with --metrics_http=/metrics option.

Specify custom port with --metrics_http_port=1234.

  • Add RPC metrics. (@palkan)

  • Refactor metrics: add descriptions, change names. (@palkan)

0.6.0-preview1

Added metrics package with some basic metrics (counters and gauges).

Metrics collected (cached) every 15s (will be configurable in the future).

Currently, only logging mode is available (disabled by default): run with --metrics_log command to flush stats every 15s to log (with info level).

NOTE: Counter values contain delta counts (for the last period).

Example metrics log entry:

INFO 2018-03-05T16:20:07.367Z auth_failures_count=0 broadcast_msg_count=228 client_msg_count=2228 clients_num=10000 context=metrics disconnect_queue_size=0 goroutines_num=21273 streams_num=1 uniq_clients_num=10000 unknown_broadcast_msg_count=0 unknown_client_msg_count=0
  • Add signal handling and graceful shutdown. (@palkan)

When receiving SIGINT or SIGTERM we:

  • Stop the server (to not accept new connections)

  • Close all registered sessions (authenticated clients)

  • Wait for pending Disconnect requests to complete

  • Wait for active RPC calls to finish.

  • [Breaking Change] New configuration and CLI options. (@palkan)

Environment variables now should be prefixed with ANYCABLE_.

REDIS_URL and PORT env variables are recognized by default.

Setting server address is split into two params: --port and --host (instead of --addr).

Run anycable-go -h to learn more.

Now we use structured logging with the help if apex/log. For example:

INFO 2018-03-05T08:44:57.684Z context=main Starting AnyCable unknown
INFO 2018-03-05T08:44:57.684Z context=main Handle WebSocket connections at /cable
INFO 2018-03-05T08:44:57.684Z context=http Starting HTTP server at 0.0.0.0:8080
INFO 2018-03-05T08:44:57.685Z context=rpc RPC pool initialized: 0.0.0.0:50051
INFO 2018-03-05T08:44:57.695Z context=pubsub Subscribed to Redis channel: __anycable__

Also, json format is supported out-of-the-box (--log_format=json or ANYCABLE_LOG_FORMAT=json).

  • Add unique identifiers to connections. (@palkan)

Helps to identify connections in logs. Will be included into RPC calls in the future releases.

  • [Closes#24] No more ping logs even in debug mode.

0.5.4 (2018-02-08)

  • Automatically reconnect to Redis when connection is lost. (@palkan)

Fixes #25.

0.5.3 (2017-12-22)

  • Fix bug with non-JSON messages. (@palkan)

Fixes #23.

0.5.1 (2017-11-08)

To secure your anycable-go server provide the paths to SSL certificate and private key:

anycable-go -addr=0.0.0.0:443 -ssl_cert=path/to/ssl.cert -ssl_key=path/to/ssl.key

=> Running AnyCable websocket server (secured) v0.5.1 on 0.0.0.0:443 at /cable
  • Handle RPC errors gracefully. (@palkan)

Avoid panic when RPC server is unavailable. All RPC call now return (response, error).

0.5.0 (2017-10-20)

  • Support passing arbitrary headers to RPC. (@palkan)

Added new CLI option to pass a list of headers:

anycable-go -headers=cookie,x-api-token,origin

By default equals "cookie".

  • Send control frame before closing connections. (@palkan)

0.4.2 (2017-09-27)

  • Fixed bug with race conditions in hub. (@palkan)

Fixed #10.

0.4.0 (2017-03-18)

  • Follow AnyCable versioning conventions. (@palkan)

Add -version flag to show current version. Print current version on startup.

0.3.0 (2017-01-22)

Use one Command call instead of separate calls (Subscribe, Unsubscribe, Perform).

Do not add identifier field.

0.2.0 (2016-12-28)

  • Add DisconnectNotifier. (@palkan)

DisconnectNotifier invokes RPC Disconnect gracefully, i.e. with the rate limit (100 requests per second by default).

Pinger now is always running and track the number of active connections by itself (no need to call hub.Size()). No more race conditions.