v7.0.0

[mosquito]

No description provided.

[torarvid]

Oh, wow, I just tested this branch, and it seems like it fixes a problem we're having in production, where a burst of message can cause the lock in basic_publish to sometimes not be released. PR #175 tries to partially address it, but I saw other issues that I couldn't figure out...

But using the branch from this PR, I'm not able to reproduce the problem 🥳

@mosquito Do you have any schedule for when you're thinking of releasing it? 😅

[mosquito]

@torarvid Oh, there are so many changes that I don't even know how long it will take to make it fine.

[dgoeries]

@torarvid what version are you using and how do you define burst? How many messages in which interval?

For now, we are very happy with version 3.3.1, higher versions degrade a bit the performance. However, last time we tested this MR, it seemed to remove the performance degradation we saw before

[torarvid]

@dgoeries To be honest, I'm not sure what burst numbers are needed to trigger it.

Someone on my team made a test script to reproduce on a local rabbit server. It sets up a client for pubsub (with publisher ack) using a channel-pool with max 10 channels. Then it goes into a loop where it asks the user how many messages to publish, and then publishes that amount of messages. Rinse and repeat. All messages are using the same timeout (usually we use 0.3 seconds in the test script, but usually between 2 and 5 seconds in production).

What we see is that if we send 100 msgs, it's always fine. If we do 1000, then usually most will time out, but after that; trying with 100 again works fine.

The interesting thing is if we try 10,000 or 20,000 msgs. Then 100% of them fail. And most importantly — if I try even just a single message after this point, it always fails. It will never recover for this channel (because the channel lock hasn't been released).

We use 6.7.1 at the moment.