Took a couple of weeks but it's reliably running locally on my laptop. At higher throughput than some years ago on dedicated hardware, before the refactoring 😎 .
Happy with how this refactor of Bunny fully relying on @reactphp is working out. I got a bit of a scare earlier tonight when I found a significant performance loss (33.333% to 50%) but managed to quickly fix that and now it's on par. (We're talking 2 seconds of difference on 6 - 8 minute benchmark runs.)
Managed to break 100K messages a second both ways, really happy and proud of this 😎!
@wyri Thanks for all the work you do for the project! At work, we had a love/hate relationship with Bunny for quite some time (we switched to a Postgres-based queue recently), and we’re super grateful for the effort that was put into maintaining it.
Quick question:
> fully relying on @reactphp
This means event loop, so no “sync” version?
E: That made no sense. Guess the real question is: Will it need a specific PHP extension to work?
@merms What was the hate part of that relationship?
Behind the scenes, it has used the event loop for ages now. But did its connection handling manually. Took out the custom connection handled and now using react/socket to create the connections and such.
So no, you don't need a specific extension to work. As always please use ext-ev or ext-uv, or another event loop extension with @reactphp in production because it boosts your performance a lot. Plus drops the 1024 filedescriptor limit.
@wyri Sorry for the late reply, we had a case of kids++. 👶
Also I really struggle with a good answer. 😅
Our biggest issue was disconnecting the Bunny client in tests. See e.g. here: https://github.com/jakubkulhan/bunny/issues/93 I ended up keeping a global list of references (the & stuff) to all Bunny instances so that I could trigger their destructors by setting them to =null and calling garbage collection.
We also had some TTL/disconnection issues b/c long-running consumer jobs blocked the main loop.
@wyri (There should also be a 1.0 release at some point, I guess. But that might be just me. 😅 I think Bunny is heavily used in production, and SemVer says that 0.* interfaces are allowed to change at any time.)
@merms No need to apologise for prioritising your newborn over replying. Congratulations 🎉 ! And yes there should be a 1.0 release. Doing the merging of both clients is a bit box to tick off for that.
And yes you're right, it had its own event loop in the sync client. Kinda forgot about that after yanking it out 😅 .
To bad you can't help me test 0.6 before tagging it outside my own set up for it. But thanks for the context on it.
Add comment