Some requests to my site are failed with "Server unexpectedly dropped the connection" and there is no errors in error logs.
RPM is 1-1.2k.
Do you have any ideas?
Here is configs of haproxy, nginx and unicorn:
Asked
Active
Viewed 657 times
2 Answers
3
Before you share logs which is critical you have misconfiguration:
Nginx worker_connections 3000 Reason: keep it 2-3x more than actual active connections number
Nginx worker_rlimit_nofile 20000. Reason: each single connection requires 2 file descriptors for serving request via proxying
Nginx absence of location that serves static files straight from file system bypass backend. Reason: if you pass any static file via backend it reach the limit very quick
Unicorn worker_processes 4. Reason: 4 workers are not enough to serve 1K RPM
System: I guess there is lack of kernel optimisation. Take a look Unicorn System Kernel tips
Anatoly
- 15,298
- 5
- 53
- 77
-
1Looks like the main problem was in worker_connections.Thanks a lot! – Ian Bernatcki Dec 09 '14 at 11:31
0
This looks like it could be an issue with your Unicorn config - it's set to have a backlog of 1024, which could cause some issues despite being the default setting.
From the Unicorn documentation: "If you are running unicorn on multiple machines, lowering this number can help your load balancer detect when a machine is overloaded and give requests to a different machine."
It may also be worth monitoring your system resources when this happens. Please let us know how this goes for you!
Philip Kallberg
- 177
- 3
-
May be, but there is no errors in unicorn's log files and ha-proxy had not captured any bad responses. – Ian Bernatcki Dec 09 '14 at 14:05