Cent

======

Minor backwards incompatible changes here when using `client_user_connection_limit` option – see below.

Centrifugo v2.8.0 has many internal changes that could affect overall performance and latency. In general, we expect better latency between a client and a server, but servers under heavy load can notice a small regression in CPU usage.

Improvements:

* Centrifugo can now maintain a single connection from a user when personal server-side channel used. See [396](https://github.com/centrifugal/centrifugo/issues/396) and [docs](https://centrifugal.github.io/centrifugo/server/server_subs/#maintain-single-user-connection)
* New option `client_concurrency`. This option allows processing client commands concurrently. Depending on your use case this option has potential to radically reduce latency between a client and Centrifugo. See [detailed description in docs](https://centrifugal.github.io/centrifugo/server/configuration/#client_concurrency)
* When using `client_user_connection_limit` and user reaches max amount of connections Centrifugo will now disconnect client with `connection limit` reason instead of returning `limit exceeded` error. Centrifugo will give a client advice to not reconnect.

======

No backwards incompatible changes here.

Fixes:

* Fix client reconnects due to `InsufficientState` errors. There were two scenarios when this could happen. The first one is using Redis engine with `seq`/`gen` legacy fields (i.e. not using **v3_use_offset** option). The second when publishing a lot of messages in parallel with Memory engine. Both scenarios should be fixed now.
* Fix non-working SockJS transport close with custom disconnect code: this is a regression introduced by v2.6.2

======

No backwards incompatible changes here.

Fixes:

* Fix non-working websocket close with custom disconnect code: this is a regression introduced by v2.6.2

======

This release has minor backwards incompatible changes in some Prometheus/Graphite metric names. This means that you may need to adapt your monitoring dashboards a bit. See details below.

Improvements:

* Previously metrics exposed by Centrifuge library (which Centrifugo is built on top of) belonged to `centrifuge` Prometheus namespace. This lead to a situation where part of Centrifugo metrics belonged to `centrifugo` and part to `centrifuge` Prometheus namespaces. Starting from v2.7.0 Centrifuge library specific metrics also belong to `centrifugo` namespace. So the rule to migrate is simple: if see `centrifuge` word in a metric name – change it to `centrifugo`.
* Refreshed login screen of admin web interface with moving Centrifugo logo on canvas – just check it out!
* New gauge that shows amount of running Centrifugo nodes
* Centrifugal organization just got [the first baker on Opencollective](https://opencollective.com/centrifugal) ❤️. This is a nice first step in making Centrifugo development sustainable.

Fixes:

* Fix `messages_sent_count` counter which did not show control, join and leave messages

**Coming soon 🔥:**

* Official Grafana Dashboard for Prometheus storage is on its way to Centrifugo users. [Track this issue](https://github.com/centrifugal/centrifugo/issues/383) for a status, the work almost finished.
* Official Centrifugo Helm Chart for Kubernetes. [Track this issue](https://github.com/centrifugal/centrifugo/issues/385) for a status, the work almost finished.

======

No backwards incompatible changes here.

Improvements:

* Internal refactoring of WebSocket graceful close, should make things a bit more performant (though only in apps which read lots of messages from WebSocket connections)
* Disconnect code is now `uint32` internally
* A bit more performant permission checks for publish, history and presence ops
* Connect proxy request payload can optionally contain `name` and `version` of client if set on client side, see updated [connect proxy docs](https://centrifugal.github.io/centrifugo/server/proxy/#connect-proxy)
* New blog post [Experimenting with QUIC and WebTransport in Go](https://centrifugal.github.io/centrifugo/blog/quic_web_transport/) in Centrifugo blog

Fixes:

* fix panic on connect in 32-bit ARM builds, see [387](https://github.com/centrifugal/centrifugo/issues/387)

======

No backwards incompatible changes here.

Improvements:

* Add `grpc_api_key` option, see [in docs](https://centrifugal.github.io/centrifugo/server/grpc_api/#api-key-authorization)

Fixes:

* Fix Redis Engine errors related to missing epoch in Redis HASH. If you see errors in servers logs, like `wrong Redis reply epoch` or `redigo: nil returned`, then those should be fixed here. Also take a look at v2.5.2 release which contains backport of this fix if you are on v2.5.x release branch.