Hey Alexey, sorry for the long time without any response of Gerhard, I am
Holger a colleague of Gerhard. So what has happend since his last response
... (1) We checked our source code again and implemented some little
modifications in the two critical servlet algorithms, so that we can be sure
that every servlet ends with an 'response.flushBuffer();' (2) After the
modification we updated the productive systems and for one week it seems to
be the solution. The server works without any problems. (See attached picture
of the CPU load graph for week 37) (3) During this time we patched our local
glassfish in the netbeans environment. It works without any problems. (4)
Unfortunatly on week 38 the productive systems raises the same problems as
mentioned before. In week 39 we had to stop the glassfish manually, because
it was unavailable for the regular http-requests. Yesterday we uploaded your
patch to our productive system and it works for more than 24 hours without
any problems. Actually we recognized at one system that one http-thread stack
into something like a loop. The difference since patching the sever is that
just one log message appears in the glassfish log with "idle thread". But the
server load is still high. One thread consumes 100% CPU power. Do you need
some more informations about our source code to understand the big picture?
Do you have any suggestions what we can do or check in our source code to
react on those bad threads? Thanks in advance, Gerhard and Holger
--
[Message sent by forum member 'HolgerD']
View Post: http://forums.java.net/node/889790