Possible Cause of Occasional FATAL: Segment Violation

I have had the same problem that many of you have had with respect to the occasional appearance of "FATAL: Segement Violation, dying..." in the cache.log file.

I have done all manner of things to try to fix this, including changing settings for poll/select, # of fd's, async/non-async I/O. I am running on a Debian 2.1 system on Intel hardware with Linux kernel 2.2.13, SCSI hard disks, lots of RAM.

I have used PGCC for all of my compiles. Many different versions. I did a recompile using gcc-2.7.2.3 (old faithful) and the errors ceased. Also, I notice that I get far fewer messages in the cache.log file.

I would like to explore this further, but I would like the input of those of you who are more familiar with the Squid source code than I am. The only change I made so far from a broken system to a working one is the change from pgcc (patched egcs) to gcc-2.7.2.3. I did try the pgcc-2.95.2, but not the base gcc-2.95.2.

If you have had this problem, try to rebuild with 2.7.2.3, and check to see if it works for you. If any of you who know the code for Squid would like my logfiles or configuration files, etc., please let me know. I can save some off for you.

Thanks,
Matt Roberts, GRDA
mattro@grda.com
Received on Thu Dec 02 1999 - 20:28:49 MST