A
Avizz
PS :I am cross posting this in some more forums in hopes of getting
some answers Please excuse me if it irritates some one
Hi All
I have the following setup
Jdk1.4.1_02 ( i have tested it with JDK1.4.0 and 1.4.2 )
Win2k SP 4 (all updates and patches made)
Apache Tomcat/4.1.18-LE-jdk14 ( Tested on newer version too)
I have a Async Server (listening for socket connections on one port )
which is being init in a static block in a Servlet ,I did this as I
need to access the Http Functionality
of Tomcat .In other words my server shares the same process as the
Servlet .So far so good ,things were working fine
Until I started noticing that some times (initially once in a month)
tomcat started taking up huge CPU resources. Driving the processor to
100% utilization,
The server still works but the machine becomes very slow,
-- My first step was to introduce 'Time outs' in all the While(true)
and places where
there were blocking operations. As it looked more like something stuck
in a for loop which was not ending But to my dismay the problem
continued
Things became worse this abnormality became more frequent
-- Next i found that when ever the this happened , i had some
connections in CLOSE_WAIT state ( doing netstat -a ).but when i
emulated the problem on my local machine ,i never had 100%cpu
utilization ,even when i had 70 - 80 CLOSE_WAIT
-- Next i found that when the N/w Goes Down the NIO Selector's
select() operation
stops blocking and returns with 0 selected keys .It went into a
infinite loop as it was in a while (true) loop ,and consumed lots of
processing as select operation was not blocking
I learnt that it was an acknowledged bug and i came up with a work
around where in i recreated / re initialized the selector when ever i
got zero selected keys (after a threshold i.e after continues 50 zero
select operation) . This worked fine on my local machine as well as
the production machine .I pulled out the n/w cable and reconfirmed. I
thought that my nightmare was over
-- But again after working fine for 3 days it has again gone into a
loop
no errors / no exception/ nw connectivity is fine too
Can any one throw some light on the topic .I can not find any thing
wrong in the code
, and all the NIO bugs i have patched. Still it behaves abnormally.I
have to restart the server for it to behave properly ,but it
disconnecs all connected clients .This is not desirable.
Is it due to the CLOSE_WAITS or the my approach to recreate selector
is wrong (works fine for some days though) ,Or some thing I don't know
yet any ways please let me know If any one has experienced any thing
like this OR can add any thing you will save fellow developer from
going insane..
Thanks and regards
Avizz
some answers Please excuse me if it irritates some one
Hi All
I have the following setup
Jdk1.4.1_02 ( i have tested it with JDK1.4.0 and 1.4.2 )
Win2k SP 4 (all updates and patches made)
Apache Tomcat/4.1.18-LE-jdk14 ( Tested on newer version too)
I have a Async Server (listening for socket connections on one port )
which is being init in a static block in a Servlet ,I did this as I
need to access the Http Functionality
of Tomcat .In other words my server shares the same process as the
Servlet .So far so good ,things were working fine
Until I started noticing that some times (initially once in a month)
tomcat started taking up huge CPU resources. Driving the processor to
100% utilization,
The server still works but the machine becomes very slow,
-- My first step was to introduce 'Time outs' in all the While(true)
and places where
there were blocking operations. As it looked more like something stuck
in a for loop which was not ending But to my dismay the problem
continued
Things became worse this abnormality became more frequent
-- Next i found that when ever the this happened , i had some
connections in CLOSE_WAIT state ( doing netstat -a ).but when i
emulated the problem on my local machine ,i never had 100%cpu
utilization ,even when i had 70 - 80 CLOSE_WAIT
-- Next i found that when the N/w Goes Down the NIO Selector's
select() operation
stops blocking and returns with 0 selected keys .It went into a
infinite loop as it was in a while (true) loop ,and consumed lots of
processing as select operation was not blocking
I learnt that it was an acknowledged bug and i came up with a work
around where in i recreated / re initialized the selector when ever i
got zero selected keys (after a threshold i.e after continues 50 zero
select operation) . This worked fine on my local machine as well as
the production machine .I pulled out the n/w cable and reconfirmed. I
thought that my nightmare was over
-- But again after working fine for 3 days it has again gone into a
loop
no errors / no exception/ nw connectivity is fine too
Can any one throw some light on the topic .I can not find any thing
wrong in the code
, and all the NIO bugs i have patched. Still it behaves abnormally.I
have to restart the server for it to behave properly ,but it
disconnecs all connected clients .This is not desirable.
Is it due to the CLOSE_WAITS or the my approach to recreate selector
is wrong (works fine for some days though) ,Or some thing I don't know
yet any ways please let me know If any one has experienced any thing
like this OR can add any thing you will save fellow developer from
going insane..
Thanks and regards
Avizz