resubscribe logic looks ignoring sharded Pub/Sub (shardChannels Set of PubSubEndpoint)

[saiya]

Bug Report

Thank you for adding support for sharded PubSub in the 6.4.0.
Related to that implementation, I would like to confirm about resubscribe (subscription recovery after reconnection) behavior.
It looks like resubscribe handling only non-sharded PubSub subscriptions.

Current Behavior

Sharded PubSub subscriptions looks not recovered after disconnection + re-connection, despite Lettuce automatically recovers non-sharded PubSub subscriptions (the recovery behavior is very helpful for applications, and also difficult / error-prone to implement by ourselves).

In the 1457486#diff-73f0c5ac0f7d6d94f8d47e1a75564aa344cbbbd4c91ce786711b5c3673044956, PubSubEndpoint.shardChannels (private) have been added but it looks that variable is private inside PubSubEndpoint and no any code of PubSubEndpoint reading that variable's data. Also implementations of StatefulRedisPubSubConnectionImpl.resubscribe and StatefulRedisClusterPubSubConnectionImpl.resubscribe does not accessing shardChannels.

Input Code

Just make sharded Pub/Sub connection, and wait until some disconnection + re-connection occurs on that environment...

Input Code

StatefulRedisPubSubConnection<String, String> connection = client.connectPubSub()
connection.addListener(new RedisPubSubAdapter<String, String>() { ... })

RedisPubSubCommands<String, String> sync = connection.sync();
sync.ssubscribe("channel"); // Sharded subscription

Expected behavior/code

As same as non-sharded Pub/Sub, automatically recover sharded Pub/Sub subscriptions when possible to connect to that shard/node.
( May be a bit different logic from non-sharded Pub/Sub needed. Because the SSUBSCRIBE requires living connection to the particular shard's node(s) )

Environment

• Lettuce version(s): 6.4.0
• Redis version: 7.1 (I belive this issue happens even with Redis 7.0)

Possible Solution

(Not tested) Watch the connection's state somehow and manually call ssubscribe again by application's logic.

Additional context

Because currently I don't have full access to a target environment / Redis cluster, this issue is based on best guess by reading code. I hope this code reading makes some sense...

[saiya]

(Just related topic) Also Javadoc of RedisClusterClient#connectPubSub and StatefulRedisClusterPubSubConnection are bit conflicting each other. I assume the StatefulRedisClusterPubSubConnection one is correct:

RedisClusterClient#connectPubSub:

 * <li>Pub/sub commands are sent to the node with the least number of clients</li>

StatefulRedisClusterPubSubConnection:

The connection provides transparent command routing based on the first command key.

↑ By reading code, this one (routing based on first command key) is the truth and that is why ssubscribe (but also subscribe, despite it can be sent to any shard/node) implementation always sends command to the key slot's shard/node

[tishun]

Hey @saiya , thanks for the report, I think you are correct and sharded subscriptions are not resubscribed right now.
I will put this in our backlog so we can address it as soon as time permits.

[tishun]

By reading code, this one (routing based on first command key) is the truth and that is why ssubscribe (but also subscribe, despite it can be sent to any shard/node) implementation always sends command to the key slot's shard/node

Both, to the best of my own understanding:

• channel names are keys and the command will be send to a node handling the slot that handles this key
• in case more than one shard is handling this slot (replica nodes) we subscribe to the one with the least number of clients

Does that make sense?

[saiya]

Thank you for the agile response! I will await for the fix.

[saiya]

Both, to the best of my own understanding:

• channel names are keys and the command will be send to a node handling the slot that handles this key
• in case more than one shard is handling this slot (replica nodes) we subscribe to the one with the least number of clients
  Does that make sense?

I see; yes indeed that make sense.

Now I understand the RedisClusterClient#connectPubSub document's intention; Another list item saying <li>Keyless commands are send to the default connection</li>, therefore that list item talking about not-keyless command implictly.