High CPU usage for web

[kholikt]

Hi,

I am running iplanet server with hpux11. I have very strange problem here. Sometimes the web service process is running at 100% Unless I restarted the process or it will remain 100%

I have use the truss to trace the system call. I have found that the web process is entering into an endless loop of poll and recv. What is that mean?

Here is the sample of the output

poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN
poll(0x7af6ea90, 1, 1000) ............................................................................. = 1
recv(33, 0x7af6d824, 4096, 0) ......................................................................... ERR#11 EAGAIN

[rick jones]

the tusc output you are showing gives every indication that either the kernle is erroneously reporting a file descriptor ready for reading, or the application is getting confused as to which FD is being reported ready by poll().

a couple things:

1) make sure you are on the latest rev of tusc and take a verbose tusc trace to see what FD is being reported ready by poll() If poll() is reporting an FD other than the one the app is passing to recv() the bug is in the app. If poll() is reporting the FD that the app is passing to recv() the bug would likely be in the kernel. At that point it is time to contact the RC. They may want a kernel dump.

2) make sure you are on the latest ARPA Transport and Streams patches

[Santosh Nair_1]

Hi,

I may be misreading this, but doesn't EAGAIN indicate
no more processes, i.e. either maxuprc or nproc was exceeded?

-Santosh