I'm (finally) getting around to putting a backup LDAP authentication server on my network. The backup uses syncrepl to grab the database, and to my eyes both LDAP servers answer read queries identically.
I'm testing the client side of this configuration on virtual CentOS 5 i386 machine. /etc/ldap.conf reads
----- %< ----- base dc=DOMAIN,dc=com timelimit 30 bind_timelimit 30 idle_timelimit 300 nss_initgroups_ignoreusers root,ldap,named,[... trimmed ...] uri ldap://ldap1.DOMAIN.com ldap://ldap2.DOMAIN.com ssl start_tls tls_cacertdir /etc/openldap/cacerts pam_password md5 ----- %< -----
The client will bind to whichever server is listed first after the 'uri' directive. In the config snippet, it's 'ldap1' -- but it works the other way too.
If the first-listed server goes away, the client never seems to try to find or bind to the second-listed server (where "never" == my patience limit of about an hour). Once the first-listed server goes away, all password authentication fails, though getent passwd and getent group still work (presumably because of nscd).
Has anyone else experienced this or, more importantly, figured out a way to get failover to work in a reasonable timeframe?