Hi, On Thu, Apr 16, 2020 at 09:40:17AM +0900, flamese...@yahoo.co.jp wrote: > Hi all, > I'm using latest v2.1.4, with a multiple processes config(nbproc 4), > everything works fine at the beginning.Meaning I can see 4 worker processes > are using CPU at 70~80% by top command. > The next day I noticed a drop of CPU usage on our monitoring system > dashboard, then I logged in and found that there are only 3 worker > processes are using CPU at 70~80% by top command. But all haproxy processes > are there(1 master, 4 workers). > The worker which does not show up on top's output only uses ~1% CPU. So I > assume that worker does not receive traffic. > This does not happen always, but when it happens, it happens after several > hours after start haproxy. > If you need more information, tell me how to do it, I'm willing to provide.
First, could you please share your config ? (just drop any confidential info like IP addresses or passwords). Did you have a working behaviour previously with another version ? If so, which one ? Do you observe anything abnormal in your logs ? On the stats page ? Since you're in multi-process mode, one thing that *could* possibly happen is that you'd be dealing with some heavy requests over very long connections, and that once some clients are connected to a given process and they send it very heavy requests, these ones are affected. But this should normally not happen unless you have lots of CPU-intensive rules like regex (or even worse, map_reg() with a large file), all of which should be possible to get rid of on the long term if needed. Willy