Bill Montgomery
2003-12-22 21:30:47 UTC
Hi,
I'm running mod_log_spread with apache 2.0.47, using the prefork MPM,
and spread 3.17.1. There is a spread daemon running locally on each web
server, specified with the following line in httpd.conf:
SpreadDaemon 3333
Most (75%?) of the apache log messages are being sent over the spread
ring and are recorded by spreadlogd on my logging host. However, portion
of the log messages cause an error, and are not succesfully sent to the
spread daemon. The errors look like this in the apache error log:
[Mon Dec 22 14:24:42 2003] [crit] [client 66.193.5.50] Spread Daemon
returned an error. -11 This Request could not be logged
/author/upload/upload_progress.cgi
There are also occasional -8 spread errors. Looking at sp.h in the
spread source tells me that these errors are CONNECTION_CLOSED (-8) and
ILLEGAL_SESSION (-11). These errors are returned a dozen different
places in the spread daemon code, and I wasn't able to tell why in most
cases.
READMEs, Google, and mailing list archives have not turned up anything
useful. I tried linking mod_log_spread.so against libtspread, thinking
it might be a problem with thread safety (don't think it should be, with
the apache prefork MPM), but the same symptoms are present. Could this
be a message length problem? I don't have any absurdly long URL's I'm
servering up, and I'm using the standard "combined" log format. Probably
no measureable percentage of log messages longer than 300 characters,
except that I'm losing a large fraction of the log messages.
Also, I see the same symptoms on a web server that is under a heavy load
(~10 log messages/sec) and another that has no load (just me testing). I
see no pattern to the failures on the server under heavy load, but I can
hit certain pages that will always generate a spread error on the test
server.
Anyone ever seen something like this? TIA for your help.
Best Regards,
Bill Montgomery
I'm running mod_log_spread with apache 2.0.47, using the prefork MPM,
and spread 3.17.1. There is a spread daemon running locally on each web
server, specified with the following line in httpd.conf:
SpreadDaemon 3333
Most (75%?) of the apache log messages are being sent over the spread
ring and are recorded by spreadlogd on my logging host. However, portion
of the log messages cause an error, and are not succesfully sent to the
spread daemon. The errors look like this in the apache error log:
[Mon Dec 22 14:24:42 2003] [crit] [client 66.193.5.50] Spread Daemon
returned an error. -11 This Request could not be logged
/author/upload/upload_progress.cgi
There are also occasional -8 spread errors. Looking at sp.h in the
spread source tells me that these errors are CONNECTION_CLOSED (-8) and
ILLEGAL_SESSION (-11). These errors are returned a dozen different
places in the spread daemon code, and I wasn't able to tell why in most
cases.
READMEs, Google, and mailing list archives have not turned up anything
useful. I tried linking mod_log_spread.so against libtspread, thinking
it might be a problem with thread safety (don't think it should be, with
the apache prefork MPM), but the same symptoms are present. Could this
be a message length problem? I don't have any absurdly long URL's I'm
servering up, and I'm using the standard "combined" log format. Probably
no measureable percentage of log messages longer than 300 characters,
except that I'm losing a large fraction of the log messages.
Also, I see the same symptoms on a web server that is under a heavy load
(~10 log messages/sec) and another that has no load (just me testing). I
see no pattern to the failures on the server under heavy load, but I can
hit certain pages that will always generate a spread error on the test
server.
Anyone ever seen something like this? TIA for your help.
Best Regards,
Bill Montgomery