Kaamil Jasani created DIRSERVER-2319:
----------------------------------------
Summary: Initial refresh in a refreshAndPersist replication
doesn't provide full subtree in master-master configuration
Key: DIRSERVER-2319
URL: https://issues.apache.org/jira/browse/DIRSERVER-2319
Project: Directory ApacheDS
Issue Type: Bug
Components: replication
Affects Versions: 2.0.0.AM26
Reporter: Kaamil Jasani
Attachments: apacheds-syncrepl-issue.zip
In our current setup, we have two instances of ApacheDS running (ads0 and
apacheds-main). They are configured in a master-master replication setup, both
disabled by default. Each is initialized with some users and groups.
Once both instances are started, we enable the replication consumer on ads0 and
restart it. This causes a refresh and pulls the data from apacheds-main into
ads0. We then disable the consumer and restart it again. At this point, ads0
contains a full image of the data, whereas apacheds-main only contains the data
it initially had. If we now enable the replication consumer on apacheds-main,
we would expect all the data from ads0 to be replicated into apacheds-main
during the initial refresh. However, this is not the case and what we see is
that apacheds-main still only has the data it started out with. We know that
the data exists in ads0 since we can search for it and see it in Apache
Directory Studio.
What is puzzling is that if the same procedure is carried out in the opposite
direction, it works as intended. That is, if rather than replicating from
apacheds-main to ads0 first we do it the other way, in the end both ApacheDS
instances end up with a full image. There is no functional difference in the
configuration of the two instances.
The real problem arises with this when we need to restart one of these
instances. What seems like the same problem seems to occur again fairly
erratically. The symptoms also seem kind of random, where sometimes the data
will simply not be synced over, sometimes the data is deleted from both
instances, or sometimes the entire partition gets corrupted and is no longer
available for writing or reading. Running ApacheDS with {{repair}} instead of
{{start}} brings the partition back, but not all the data seems to be retrieved.
I have attached an archive with the logs for both instances in both situations.
It contains 5 files:
* {{ADS0FIRST-ads0.log}}: logs for ads0 when replicating to ads0 first
* {{ADS0FIRST-apacheds-main.log}}: logs for apacheds-main when replicating to
ads0 first
* {{ADS0FIRST-ads0-direct-search.ldif}}: search result showing all the data
existing in ads0
* {{ADSMAINFIRST-ads0.log}}: logs for ads0 when replicating to apacheds-main
first
* {{ADSMAINFIRST-apacheds-main.log}}: logs for apacheds-main when replicating
to apacheds-main first
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
