On 01/17/2011 04:53 AM, Rainer Jung wrote: > On 13.01.2011 00:36, Andy Wang wrote: >> Aahh, having the maintenance thread do a periodic probe would be >> awesome. >> >> I see what you mean about parallel probing delaying request handling, >> but what would you think about modifying the loop so that after going >> through all the JK_WORKER_USABLE() workers to retry the PROBING >> workers. At that point if none of the workers are up, then it seems >> like parallel probing wouldn't be a bad idea would it? > > If none of the workers are up, then normally "forced recovery" kicks > in, i.e. the workers that erred will be put into mode > JK_LB_STATE_FORCE, which is "this worker is in error, but use it > nevertheless, because there is no good worker left". > > Regards, > > Rainer I'm having a really hard time reproducing it, but the logs from one of the times I saw it seemed to imply that the worker being tried by the first thread was put into PROBE state, and then another thread came in, and completed, with an error state because none of the other workers are alive before the first thread's worker completed and it skipped the worker the first thread was using becaause it was put into PROBE state. So the first worker thread's worker was never used by the second thread's worker resulting in the second thread.
There was never a chance for the first worker to go into FORCE mode. It was set into PROBE mode, and somehow thread 2 snuck in and before the 1st thread finished. Andy Andy --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
