[.This message is going to both cgpsa-discuss and spamassassin-users, so
please watch your replies unless you intend to cross-post. Thx.]
We've been having some major problems with our SpamAssassin/Communigate
install since upgrading SA and CGPSA last week. I and another programmer
inherited the server responsibilities after our sysadmin left. (Small
ISP/hosting company, don't ask, we're pretty much stuck with admining this
thing.)
After trying to install Razor/Pyzor/DCC and failing miserably, we've
completely renamed the razor binary so SA can't find it (because it seemed
to be trying to use it in whatever horked-up crippled install state it
ended up in, and was bringing down the server in the process -- at least so
we thought, since doing that stabilized the machine).
That was Wednesday night, and all seemed good and fine...
..until this morning when I updated a bunch of custom .cf files (bigevil,
backhair, etc), trained Bayes, and restarted** CGP. (**Side question on
this, please see below...)
Wednesday's nightmare started all over again with the load climbing to
scary values like 60 and 90. (Wednesday it hit 99 and completely locked up
the machine. TWICE.) So now we've turned off the network checks in the
cgpsa.conf file, plus set debug to false and the debug_level to 0 in
it. Now the machine is "stable" again and bouncing around between 1.5 and
9.5 for load average. (Prior to this nightmare I've never seen a load on a
linux box be any higher than about 2.5.)
We're running RH 7.2 (stuck with it for now), SpamAssassin 2.63,
CommuniGate 4.1.6 and CGPSA 1.2.7
I'm watching top right now, and every time a cgpsa process appears in it,
it is using anywhere from 30% to 90% CPU. Is this normal??????
We have NO clue as to why this poor machine has been crying for mommy over
the last few days -- but since the answer to "so what have you changed?" is
that we upgraded SA and CGPSA (and tried installing Razor, Pyzor, and DCC)
that pretty much narrows the field as to what is causing it. And our tech
at the co-lo who co-manages the servers with us is just as perplexed as we are.
Anybody have any insight as to what is going on here? After the
experiences of the last couple of days I'm to the point of not wanting to
touch SA anymore, and pretty much say screw-it to training Bayes and
updating the .cf files. :-( :-(
The sad thing is, our customers weren't so much pissed that the mail server
was down for 10 hours on Wednesday than they were pissed that we had to
keep turning CGPSA/SpamAssassin on and off while diagnosing the problem
(thus letting email come through and not get flagged as spam). :-/
--JR
**Side question: how exactly does CGPSA integrate CGP to SA? If I do an
/etc/init.d/spamassassin restart will that re-read my custom .cf files? Or
is CGPSA calling SA directly some other way? Or do I have to go into CGP's
Helpers page and uncheck/Update/re-check/Update CGPSA? Or do I have to
completely restart CGServer itself (which is what I've been doing)? Thx.
please watch your replies unless you intend to cross-post. Thx.]
We've been having some major problems with our SpamAssassin/Communigate
install since upgrading SA and CGPSA last week. I and another programmer
inherited the server responsibilities after our sysadmin left. (Small
ISP/hosting company, don't ask, we're pretty much stuck with admining this
thing.)
After trying to install Razor/Pyzor/DCC and failing miserably, we've
completely renamed the razor binary so SA can't find it (because it seemed
to be trying to use it in whatever horked-up crippled install state it
ended up in, and was bringing down the server in the process -- at least so
we thought, since doing that stabilized the machine).
That was Wednesday night, and all seemed good and fine...
..until this morning when I updated a bunch of custom .cf files (bigevil,
backhair, etc), trained Bayes, and restarted** CGP. (**Side question on
this, please see below...)
Wednesday's nightmare started all over again with the load climbing to
scary values like 60 and 90. (Wednesday it hit 99 and completely locked up
the machine. TWICE.) So now we've turned off the network checks in the
cgpsa.conf file, plus set debug to false and the debug_level to 0 in
it. Now the machine is "stable" again and bouncing around between 1.5 and
9.5 for load average. (Prior to this nightmare I've never seen a load on a
linux box be any higher than about 2.5.)
We're running RH 7.2 (stuck with it for now), SpamAssassin 2.63,
CommuniGate 4.1.6 and CGPSA 1.2.7
I'm watching top right now, and every time a cgpsa process appears in it,
it is using anywhere from 30% to 90% CPU. Is this normal??????
We have NO clue as to why this poor machine has been crying for mommy over
the last few days -- but since the answer to "so what have you changed?" is
that we upgraded SA and CGPSA (and tried installing Razor, Pyzor, and DCC)
that pretty much narrows the field as to what is causing it. And our tech
at the co-lo who co-manages the servers with us is just as perplexed as we are.
Anybody have any insight as to what is going on here? After the
experiences of the last couple of days I'm to the point of not wanting to
touch SA anymore, and pretty much say screw-it to training Bayes and
updating the .cf files. :-( :-(
The sad thing is, our customers weren't so much pissed that the mail server
was down for 10 hours on Wednesday than they were pissed that we had to
keep turning CGPSA/SpamAssassin on and off while diagnosing the problem
(thus letting email come through and not get flagged as spam). :-/
--JR
**Side question: how exactly does CGPSA integrate CGP to SA? If I do an
/etc/init.d/spamassassin restart will that re-read my custom .cf files? Or
is CGPSA calling SA directly some other way? Or do I have to go into CGP's
Helpers page and uncheck/Update/re-check/Update CGPSA? Or do I have to
completely restart CGServer itself (which is what I've been doing)? Thx.