Re: syncrepl_del_nonpresent leading to directory implosion
On Tue, 8 Jul 2014, Aaron Bennett wrote:
Jul 7 04:19:30 zoot slapd[15011]: bdb(dc=clarku,dc=edu): Lock table
is out of available lock entries
Jul 7 04:19:30 zoot slapd[15011]: => bdb_idl_insert_key: c_get failed: Cannot
allocate memory (12)
[...]
That goes on, deleting things, until my pager exploded when some
critical role accounts got nuked and our radius servers died. I pulled
an LDIF when I realized what was happening and between 18:55 and 19:35,
our LDIF went from 75MB to 45MB ? so about 30MB of stuff disappeared.
What happened? Did I get hit by an oddball BDB/Syncrepl behavior? Do I
have a bad misconfiguration in my replication setup?
Usually what the syslogs say is right ... underconfigured re: number of
locks. Check your DB_CONFIG, per
http://www.openldap.org/faq/data/cache/893.html and the associated links.
I suppose there's a potential argument to be made for "I wish it died fast
instead of continued into a weird syncrepl state."
Other possibilities include a BDB bug, or perhaps you really were out of
memory (but I'd also expect core portions of slapd and possibly other
processes to be complaining in an OOM situation, which you didn't indicate
in your report).
#include obligatory lmdb plug here (no need to play around with this
stuff). This may help you justify the transition to management, if needed!