--On Tuesday, November 17, 2009 09:11:23 PM -0800 Quanah Gibson-Mount quanah@zimbra.com wrote:

On Nov 16, 2009, at 9:03 PM, Bill MacAllister whm@stanford.edu wrote:

...

--On Monday, November 16, 2009 07:37:49 PM -0800 Quanah Gibson-Mount quanah@zimbra.com wrote:

...

--On November 16, 2009 8:27:21 PM +0000 whm@stanford.edu wrote:

...

Full_Name: Bill MacAllister Version: 2.4.19+ OS: Debian 5 URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (171.64.19.165)

On a system where slapd is the top CPU process and is consuming 4% of the CPU simple queries are taking ten of seconds to complete. Here is an example query:

Last time we saw this it appeared that it might be related to DNS. How long does a netstat -a take on the affected systems when the LDAP queries are slow?

I don't have that number. I have only see it once in the test environment I am not sure how long it will take me to get it. Since this plastered the production service today, I will have to reproduce it though before we attempt another production upgrade.

Not clear to me why a DNS issue would appear only under load and not at other times. And it seems strange that it appears in lenny/openldap-2.4 and not in etch/openldap-2.3.

Bill

-- Bill MacAllister, System Software Programmer Unix Systems Group, Stanford University

For historical note this was caused by Cyrus-sasl being built incorrectly by the debian packagers when heimdal is used.

--Quanah

I don't understand why you refer to this finding as historical. If I am reading this correctly you and Howard have found the underlying cause. Now that the problem is understood can you suggest a way for us to cause the problem in our test environments? At this point we will really need to convince ourselves that the problem is indeed fixed before we try to deploy 2.4 in our production environment again.

Bill