[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: back-meta hangs



Sébastien Georget wrote:

First I set threads=2 in slapd.conf.

Using 'charge' (http://sourceforge.net/projects/loadtesting/), I start 2 threads (clients) doing the following : bind, search base dc=domain, unbind.
Almost every time the server hangs. If I hit ctrl-c I returns the following message and I have to kill it :
slapd shutdown: waiting for 4 threads to terminate


Notes :
- If I replace the search with base o=A|B,dc=domain there is no problem.
- With the default configuration (threads=16) I have to launch 17 clients to freeze slapd.


I can produce a full log (67KB) of this scenario if you want.


I've fixed the problem in HEAD; a similar fix also applies to back-ldap, although might be a bit more difficult to trigger. I'll try to backport to 2.2 asap.

This is the intended usage; if all databases reside on the same server and you don't need to rewrite or do some fancy stuff, I suggest you create a third bdb instance rooted at "dc=domain", and glue the three together using the "subordinate" qualifier in the two subordinate databases.

I use two bdb + a meta because I have read-write access to A, and B is a replica of another site (using syncrepl).

backglue can be used regardless of the type of database (e.g. shadow or not). I think a backglue approach is preferable to back-meta when all data resides in the same DSA.


p.



   SysNet - via Dossi,8 27100 Pavia Tel: +390382573859 Fax: +390382476497