Re: (ITS#3672) Odd behavior of back meta when tortured by concurrency test

[ando@sys-net.it]

OK, I took this chance to flush my todo list a bit, and I continued 
reworking connection handling to adhere a bit more to what's being done 
in back-ldap.  I've added locks to connection data and set up the 
infrastructure for retry.  The problem persisted on a more reliable 
system, although much less frequent.  Now I'm unable to reproduce it 
with back-bdb or back-hdb, while it sometimes appears with back-ldbm.  
As a consequence, my guess is that it's not fully gone, but its cause 
might be somewhere else.  Note that the test is painfully slow with 
back-ldbm, while with comparable number of loops I don't see an 
appreciable overhead with respect to test008 (if I turn logging off; 
otherwise, libldap's logging may become very verbose...).

I'd mark this as suspended right now.

p.

-- 
Pierangelo Masarati
mailto:pierangelo.masarati@sys-net.it



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