Unexpected false negatives

[Firm]

I use http-type check to trace health of upstreams. Sometimes module wrongly marks upstream as failed although it gets 200 OK. By checking dump I noticed the following things:

• if check is marked as failed (recorded to error.log as check time out with peer) (even it gets 200 OK from remote host) then nginx sends RST immediately after getting reply
• if check is marked as passed then normal session close is happened (with FIN/ACK)
• if I lower number of nginx workers from auto (40) to 5-10 then false negatives become very rare
• if I raise timeout from 2-3 seconds to 20-30 seconds then false negatives become very rare too

Does each nginx worker run its own checks for upstream(s) or there's one 'process' which manages these checks?

[yaoweibin]

OK, I will check this problem.