https://issues.apache.org/bugzilla/show_bug.cgi?id=56119
--- Comment #21 from Sebb <[email protected]> --- Testing with the privately provided test plan shows that the problem occurs on one of the hosts only. This host only occurs once in the plan, which means that its connection is relatively idle compared with the others. When you add in all the other samples and delays, the elapsed time between samples of this particular host is about 60 seconds. I just tried a simple GET on the same host, and that also fails when the elapsed time between samples is around 15 secs or more (10 secs seems to be OK) I think this is the immediate cause of the issue. The staleCheck option adds an extra sample just before real sample. This causes the old connection to be dropped by the server, but the stale check code does not report this: it just creates a new connection, which then works because it has not been idle. That is why disabling Keep-Alive works - there is no attempt to reuse the connection which has been idle for a while. The work round - if one does not want to use staleCheck for everything - is to disable keep-alive for samples that are likely to trigger the idle timeout. We need to have a dev@ discussion on what we might want to do to fix this in JMeter. I'll start that off shortly. -- You are receiving this mail because: You are the assignee for the bug.
