v0.12: periodic performance drop due to GC

[bsnote]

We have a pub/sub system which is running fine on v0.10. When I tried to run it on v0.12, it failed after some time - the system went unreachable. I made the assumption that this has something to do with GC. My next step was to try to specify --max_old_space_size value. I set it to 364 MB (10% of freeMem). This time it didn't crash but every 3 hours performance dropped. See New Relic stats:

System details:
It's running on Amazon EC2 c3.large instance:
CPU: 2 cores
memory: 3.68 GB
OS: Amazon Linux (3.14.44-32.39.amzn1.x86_64)

[CGavrila]

Not sure if this is such a great idea, but I'm just going to throw it here.

You could try having a gc(); call somewhere inside some piece of code which is executed frequently and run your app with --expose_gc. This way you might at least spread the load, even though I'm not sure if you'd want this as a permanent solution.

[bsnote]

Thanks @CGavrila . Still, I think, it's a common problem that needs to be addressed.

[bsnote]

What is the best way to diagnose this issue?
cc @bnoordhuis

[bsnote]

Same problem when running on iojs v2.3.4.

[ChALkeR]

@bsnote Is this still an issue with recent Node.js versions (v4.x / v5.x)?

[bsnote]

It's running OK with Node.js v4.x/v5.x.