--On Friday, April 10, 2015 5:12 PM +0000 tpretz(a)gmail.com wrote:
Full_Name: Tom Pressnell
OS: debian wheezy
Submission from: (NULL) (126.96.36.199)
In testing i have encountered the following when restoring replication
from multiple providers.
slapd: slapd starting
slapd: null_callback : error code 0x10
slapd: syncrepl_entry: rid=106 be_modify failed (16)AsAslapd: do_syncrepl:
rid=106 rc 16 retrying (9 retries left)
Running 2.4.40 on debian wheezy amd64.
This occurs with both the mdb and hdb backends.
While the consumer was offline i removed all values of an indexed
attribute from hundreds of entries, all providers are synced.
Using standard syncrepl (not delta), issue occurs at process startup
where the consumer has been offline for a short while, i.e. replaying the
sessionlog not falling back to present.
Having a read of syncrepl.c it would seem that syncrepl_diff_entry is
called for a entry received from a provider in a race with the same entry
from another provider in a different thread.
Both diff calls result in an attribute delete operation, when be_modify
is hit for the second time it fails with no such attribute as the first
thread has already removed it (uses DELETE, not SOFTDEL).
Please test against current RE24, as there have been a number of fixes to
Zimbra :: the leader in open source messaging and collaboration