Chris Wedgwood wrote:
> Grrr... didn't really want to have to do that if possible. It _used_
> to work 2.2.PRE2 + some of your fixes, now it doesn't....
Then you have to do some debugging to find out what is going on.
First try enabling more detailed debug logging.
If this fails to indicate where the problem is, you have to use gdb or
add more debug print statements.
It is actually not that hard to use gdb as Squid only has one thread
(not counting the "low level" I/O threads). All you need to do is to
place relevant brakepoints before you start Squid (at least this was how
I did in my glibc based system before gdb understood about Linux
threads), or if you'd like to debug a already running Squid figure out
which process that is the main thread and attach gdb to that process.
I'll try to play around some more with async-io in the weekend to see if
there is any obvious errors, but I suspect that the problem won't
reproduce in my test environment..
I must however admit that after the Squid-FS discussion have I lost
interest in the current async-io implementation of Squid. It is done on
a to low level with way to much overhead to be really beneficial. If I
ever come to writing my Squid-FS it will include threads as part of the
basic design (actually it will be a requirement of the design to make it
efficient and scalable, which is the whole point of a custom Squid-FS).
/Henrik
Received on Tue Jul 29 2003 - 13:15:57 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:04 MST