Hi, If I activate error logging for ACL Control Summary or similar, the machine totaly goes into IO-Wait. It just writes maybe 100K/second but is totaly unusable any more... From normal 1% CPU Load (on a 2 Way Xeon) it moves to 200% CPU Utilization. Debugging ACLs therefore is nearly impossible on a productive system... We run the Fedora-DS 1.0.4 (fedora-ds-1.0.4-1.RHEL4) on an up2date CentOS System... Is there any perfomance tuning option like with the log-buffering for access log. I can''t see, why logging kills the machine! cheers.roman
David Boreham
2007-Oct-09 13:52 UTC
Re: [Fedora-directory-users] Error Logging Performance
Roman RATHLER wrote:> If I activate error logging for ACL Control Summary or similar, the > machine totaly goes into IO-Wait. It just writes maybe 100K/second but > is totaly unusable any more... From normal 1% CPU Load (on a 2 Way > Xeon) it moves to 200% CPU Utilization. > Debugging ACLs therefore is nearly impossible on a productive system... > > We run the Fedora-DS 1.0.4 (fedora-ds-1.0.4-1.RHEL4) on an up2date > CentOS System... Is there any perfomance tuning option like with the > log-buffering for access log. I can''t see, why logging kills the machine!Errors often occur before a crash, and therefore the error log is flushed to persistent store often in order to improve the chances that any message emitted before a crash will be retained. I seem to remember that buffering can be enabled on the error log but I can''t remember the details. Probably in the documentation somewhere though. I think the underlying code is the same for the error log vs. the access log (which is optimized for performance by default) , so it should be possible to configure the error log to buffer. Alternatively you could put the error log file on a ramdisk.
Richard Megginson
2007-Oct-09 15:43 UTC
Re: [Fedora-directory-users] Error Logging Performance
David Boreham wrote:> Roman RATHLER wrote: >> If I activate error logging for ACL Control Summary or similar, the >> machine totaly goes into IO-Wait. It just writes maybe 100K/second >> but is totaly unusable any more... From normal 1% CPU Load (on a 2 >> Way Xeon) it moves to 200% CPU Utilization. >> Debugging ACLs therefore is nearly impossible on a productive system... >> >> We run the Fedora-DS 1.0.4 (fedora-ds-1.0.4-1.RHEL4) on an up2date >> CentOS System... Is there any perfomance tuning option like with the >> log-buffering for access log. I can''t see, why logging kills the >> machine! > Errors often occur before a crash, and therefore the error log > is flushed to persistent store often in order to improve the chances > that any message emitted before a crash will be retained. > > I seem to remember that buffering can be enabled on the error log > but I can''t remember the details. Probably in the documentation > somewhere though. > I think the underlying code is the same for the error log vs. the > access log > (which is optimized for performance by default) , so it should be > possible > to configure the error log to buffer.I don''t think this is possible. It looks like only the access log has this switch.> > Alternatively you could put the error log file on a ramdisk.Another option is to replace the error log file with a named pipe. The other end of the pipe is connected to a python script that keeps a circular log buffer in memory. The script below allows you to configure the size of that buffer, and if you want to use a fifo for the access and audit logs.> > > -- > Fedora-directory-users mailing list > Fedora-directory-users@redhat.com > https://www.redhat.com/mailman/listinfo/fedora-directory-users