Chunked Pub/Sub message receive with interleaved command responses leaves commands uncompleted

lettuce version: 5.1.2.RELEASE

When i subscribe multi channels at one time, is there a problem? see blew.

// io.lettuce.core.pubsub.PubSubCommandHandler#decode
// when i call: pubSubCommands.subscribe("test1","test2","test3"...);
protected void decode(ChannelHandlerContext ctx, ByteBuf buffer) throws InterruptedException {
        if (!getStack().isEmpty()) {
            //first channel("test1") decoded here, and cmd polled from stack
            super.decode(ctx, buffer);
        }

        ReplayOutput<K, V> replay;
        while ((replay = queue.poll()) != null) {
            replay.replay(output);
            endpoint.notifyMessage(output);
            output = new PubSubOutput<>(codec);
        }

        while (super.getStack().isEmpty() && buffer.isReadable()) {
            // other channels decode here, like "test2" "test3"...
           // is this a problem here, in my opinion, all channels should decode on the first place where "test1" decoded.
            if (!super.decode(buffer, output)) {
                return;
            }
            endpoint.notifyMessage(output);
            output = new PubSubOutput<>(codec);
        }
        buffer.discardReadBytes();
    }

When i attempt to subscribe multi channels cyclic with 100 threads in one redis connection, I get an RedisCommandTimeoutException . Maybe you thought i shouldn’t call it like this. But i truly have this requirement in my Application. Let’s see the blew code.

// code in io.lettuce.core.pubsub.PubSubCommandHandler#decode
// when i call pubSubCommands.subscribe("test1","test2","test3"...) cyclic in 100 threads in one redis connection.
if (!super.decode(buffer, output)) {
       //channel "test2" "test3"... may be decoded here. image that, if super.decode(buffer, output)  
       //return false, it means the byte pack is not enough for "test2" decoding. Then function returned. 
       //at the same time, other thread put a command to stack. When netty channel get another bytes. 
       //we step this method at second time, super.decode(ctx, buffer) while perform nothing
       //because output.type() is not null. super.getStack().isEmpty() will be false, 
       //then the bytes will never be decoded .
                return;
}

If i just subscribe one channel. but there also maybe a problem.it same to previous problem. Image that, if i get a message from a redis topic(channel),then decode message here

if (!super.decode(buffer, output)) {return}

if the message byte pack is not complete, super.decode(buffer, output) will return false and method return. at the same time i call a subscribe use the same connection, then output.type() will not be null, and super.getStack().isEmpty() will be false, we get a death loop, bytes will never be decoded ,like previous problem.

My Application use one connection to dynamic subscribe lots of channel，so I will subscribe and receive message at same time in multi threads.

[mp911de]

I pushed a change to our feature branch for this ticket and the branch build that has produced artifact snapshots with version 5.2.0.gh936-SNAPSHOT. Care to upgrade to see whether the change fixes your issue? The above mentioned reproducer no longer produces timeouts.

Make sure to include our snapshot repository:

<dependencies>
  <dependency>
    <groupId>io.lettuce</groupId>
    <artifactId>lettuce-core</artifactId>
    <version>5.2.0.gh936-SNAPSHOT</version>
  </dependency>
</dependencies>

<repositories>
  <repository>
    <id>sonatype-snapshots</id>
    <url>https://oss.sonatype.org/content/repositories/snapshots</url>
    <snapshots>
      <enabled>true</enabled>
    </snapshots>
  </repository>
</repositories>

[mp911de]

The artifact was cleared out from the repository. 5.1.4 is scheduled for Feb 12.