Howard Chu wrote, with irony, on 21. feb 2007 16:43:
[...]
Unix has this cool thing called I/O redirection, perhaps you've heard of it?
[...]
and
There's also this other cool thing called job control...
[...]
This has nothing to do with that thread, but has been worrying me:
Our production RHAS4 site with 4 OpenLDAP 2.3 servers (1 delta syncrepl provider, 3 consumers) has been running exemplararily (new word?) since ~= 2.3.27.
All servers were running 2.3.33 perfectly, then I built 2.3.34 and on just one of the servers (the master), after around 4 hours of it doing hard work, I see (top) the slapd process consuming 98-99% CPU. The CPU consumption is never relinquished, even if I wait for hours (ouch). If I reinstall 2.3.33 this goes away and all is back to normal.
On my (discrete=nothing to do with the others) test rig, 2.3.34 from the same batch also functions perfectly.
Where should I be looking?
System and build details for all rigs:
OS Red Hat RHAS4, completely up2date All, but all software built and installed as rpms
My own-built db4-4.2.52 rpms with all patches (Red Hat's doesn't have all patches)
For the OpenLDAP 2.3 rpms I'm using Buchan Milne's 2.4.24 srpm spec file, basically only modified each build for the version, so that all the %prep, %configure and %build stages are the same for all builds - irrespective of OL version.
Ta in advance,
--Tonni
Tony Earnshaw wrote:
This has nothing to do with that thread, but has been worrying me:
Our production RHAS4 site with 4 OpenLDAP 2.3 servers (1 delta syncrepl provider, 3 consumers) has been running exemplararily (new word?) since ~= 2.3.27.
All servers were running 2.3.33 perfectly, then I built 2.3.34 and on just one of the servers (the master), after around 4 hours of it doing hard work, I see (top) the slapd process consuming 98-99% CPU. The CPU consumption is never relinquished, even if I wait for hours (ouch). If I reinstall 2.3.33 this goes away and all is back to normal.
For this, absolutely, make sure you have debug symbols, attach with gdb and get a stack trace for all active threads.
On my (discrete=nothing to do with the others) test rig, 2.3.34 from the same batch also functions perfectly.
Where should I be looking?
System and build details for all rigs:
OS Red Hat RHAS4, completely up2date All, but all software built and installed as rpms
My own-built db4-4.2.52 rpms with all patches (Red Hat's doesn't have all patches)
For the OpenLDAP 2.3 rpms I'm using Buchan Milne's 2.4.24 srpm spec file, basically only modified each build for the version, so that all the %prep, %configure and %build stages are the same for all builds - irrespective of OL version.
Ta in advance,
Quite far in advance, we're only on 2.4.4 here.
Howard Chu wrote, on 21. feb 2007 19:11:
This has nothing to do with that thread, but has been worrying me:
Our production RHAS4 site with 4 OpenLDAP 2.3 servers (1 delta syncrepl provider, 3 consumers) has been running exemplararily (new word?) since ~= 2.3.27.
All servers were running 2.3.33 perfectly, then I built 2.3.34 and on just one of the servers (the master), after around 4 hours of it doing hard work, I see (top) the slapd process consuming 98-99% CPU. The CPU consumption is never relinquished, even if I wait for hours (ouch). If I reinstall 2.3.33 this goes away and all is back to normal.
For this, absolutely, make sure you have debug symbols, attach with gdb and get a stack trace for all active threads.
Thanks.
On my test machine:
I reinstalled the 2.3.33 rpms, then simply copied the RPM_BUILD_DIR/openldap-2.3.34/servers/slapd/.libs/slapd to /usr/sbin/slapd (made a backup first). It seems to run ok with the 2.3.33 libs, or do I need the 2.3.34 libs too? Whatever i do on the test rig I have to replicate on the production master/provider. I'm loathe to reinstall the whole 2.3.34 gamut on that machine, I'd rather be able to copy the 2.3.33 slapd back, if I get problems (the production provider does failover to a slave/consumer if slapd goes down, so short outages shouldn't be a problem).
[...]
For the OpenLDAP 2.3 rpms I'm using Buchan Milne's 2.4.24 srpm spec file, basically only modified each build for the version, so that all the %prep, %configure and %build stages are the same for all builds - irrespective of OL version.
Ta in advance,
Quite far in advance, we're only on 2.4.4 here.
Yeah :( sorry for the typo ...
--Tonni
openldap-software@openldap.org
-
Howard Chu -
Tony Earnshaw