Releases: centrifugal/centrifuge
v0.10.1
v0.10.0
This release is a massive rewrite of Centrifuge library (actually of some part of it) which should make library a more generic solution. Several opinionated and restrictive parts removed to make Centrifuge feel as a reasonably thin wrapper on top of strict client-server protocol.
Most work done inside #129 pr and relates to #128 issue.
Release highlights:
- Layer with namespace configuration and channel rules removed. Now developer is responsible for all permission checks and channel rules.
- Hard dependency on JWT and predefined claims removed. Users are now free to use any token implementation – like Paceto tokens for example, use any custom claims etc.
- Event handlers that not set now always lead to
Not available
error returned to client. - All event handlers now should be set to
Node
before calling itsRun
method. - Centrifuge still needs to know some core options for channels to understand whether to use presence inside channels, keep Publication history stream or not. It's now done over user-defined callback function in Node Config called
ChannelOptionsFunc
. See its detailed description in library docs. - More idiomatic error handling in event handlers, see #134.
- Aliases to
Raw
,Publication
andClientInfo
Protobuf types removed from library public API, see #136 - Support Redis Sentinel password option
Look at updated example in README and examples folder to find out more.
I hope to provide more guidance about library concepts in the future. I feel sorry for breaking things here but since we don't have v1 release yet, I believe this is acceptable. An important note is that while this release has lots of removed parts it's still possible (and not too hard) to implement the same functionality as before on top of this library. Feel free to ask any questions in our community chats.
v0.9.1
- fix
Close
method – do not use error channel since this leads to deadlock anyway, just close in goroutine. - fix presence timer scheduling
gorelease -base=v0.9.0 -version=v0.9.1
github.com/centrifugal/centrifuge
---------------------------------
Incompatible changes:
- (*Client).Close: changed from func(*Disconnect) chan error to func(*Disconnect) error
v0.9.1 is a valid semantic version for this release.
v0.9.0
This release has some API changes. Here is a list of all changes in release:
Incompatible changes:
- (*Client).Close: changed from func(*Disconnect) error to func(*Disconnect) chan error
- (*Client).Handle: removed
- Config.ClientPresencePingInterval: removed
- NewClient: changed from func(context.Context, *Node, Transport) (*Client, error) to func(context.Context, *Node, Transport) (*TransportClient, CloseFunc, error)
- NodeEventHub.ClientRefresh: removed
- RefreshHandler: changed from func(context.Context, *Client, RefreshEvent) RefreshReply to func(RefreshEvent) RefreshReply
Compatible changes:
- (*ClientEventHub).Presence: added
- (*ClientEventHub).Refresh: added
- CloseFunc: added
- Config.ClientPresenceUpdateInterval: added
- ConnectReply.ClientSideRefresh: added
- PresenceEvent: added
- PresenceHandler: added
- PresenceReply: added
- RefreshEvent.Token: added
- RefreshReply.Disconnect: added
- TransportClient: added
Now let's try to highlight most notable changes and reasoning behind:
NewClient
returnsTransportClient
andCloseFunc
to limit possible API on transport implementation levelClientPresencePingInterval
config option renamed toClientPresenceUpdateInterval
- Centrifuge now has
client.On().Presence
handler which will be called periodically while connection alive everyClientPresenceUpdateInterval
Client.Close
method now creates a goroutine internally - this was required to prevent deadlock when closing client from Presence and SubRefresh callback handlers.- Refresh handler moved to
Client
scope instead of beingNode
event handler ConnectReply
now has newClientSideRefresh
field which allows setting what kind of refresh mechanism should be used for a client: server-side refresh or client-side refresh.- It's now possible to do client-side refresh with custom token implementation (example)
- Library now uses one concurrent timer per each connection instead of 3 - should perform a bit better
All examples updated to reflect all changes here.
v0.8.2
v0.8.1
v0.8.0
This release is huge. In general, it does not change any previously defined semantics but changes API. The list of changes is big enough but fixes you need to do for this version adoption are mostly minimal. Except one thing emphasized below.
So here is that thing. Centrifuge now uses new offset
uint64
protocol field for Publication position inside history stream instead of previously used seq
and gen
(both uint32
) fields. It replaces both seq
and gen
. This change required to simplify working with history API in perspective. This is a breaking change for library users in case of using history recovery feature – read migration steps below.
Our client libraries centrifuge-js
and centrifuge-go
were updated to use offset
field. So if you are using these two libraries and utilizing recovery feature then you need to update centrifuge-js
to at least 2.6.0
, and centrifuge-go
to at least 0.5.0
to match server protocol (see the possibility to be backwards compatible on server below). All other client libraries do not support recovery at this moment so should not be affected by field changes described here.
It's important to mention that to provide backwards compatibility on client side both centrifuge-js
and centrifuge-go
will continue to properly work with a server which is using old seq
and gen
fields for recovery in its current form until v1 version of this library. It's possible to explicitly enable using old seq
and gen
fields on server side by calling:
centrifuge.CompatibilityFlags |= centrifuge.UseSeqGen
This allows doing migration to v0.8.0 and keeping everything compatible. Those CompatibilityFlags
will be supported until v1 library release. Then we will only have one way to do things.
Other release highlights:
- support Redis Streams - radically reduces amount of allocations during recovery in large history streams, also provides a possibility to paginate over history stream (an API for pagination over stream added to
Node
- seeHistory
method) - support Redis Cluster, client-side sharding between different Redis Clusters also works
- use alternative library for JWT parsing and verification - HMAC-based JWT parsing is about 2 times faster now. Related issues: #109 and #107.
- new data structure for in-memory streams (linked list + hash table) for faster insertion and recovery in large streams, also it's now possible to expire a stream meta information in case of Memory engine
- fix server side subscriptions to private channels (were ignored before)
- fix
channels
counter update frequency (commit) - drop Dep support - library now uses Go mod only for dependency management
- slightly improved test coverage
- lots of internal refactoring and style fixes
Thanks to @Skarm, @cristaloleg and @GSokol for contributions.