This topic has 2 replies, 3 voices, and was last updated 5 years, 9 months ago by 
-
Posts
-
We are trying to set up OpenDJ replication and have a few questions on how the topology should look.
Our current topology has 2 data centers each with 2 standalone Directory Servers, 2 standalone Replication Controllers.
– The Directory Servers are connected to both Replication controllers in the same data center.
– The replication Controllers are all connected to each other.What I am currently seeing is that when a Directory Server loses connection to one of the two Replication Controllers it blocks all incoming writes, but allows reads. This is because we have the isolation-policy set to reject-all-updates.
My question: Is there a way where if the connection to one Replication Server goes down, it would fail over to the other Replication Controller and not block updates until connection to both Replication Controllers goes down.
A Directory Server is only connected to a single RS at a time, but should be aware of all RS and reconnect to any of them, when it’s primary RS no longer respond.
If a DS cannot reach out any RS, yes, it is rejecting all updates (because it’s the responsibility of a RS to store and forward all changes to the other servers).
You need to make sure that the RS are addressable by all DS. i.e. their hostnames are resolvable or their IP addresses : ports are accessible from the DS.
OpenDJ Administration Guide has a section on replication with multiple datacenters : https://backstage.forgerock.com/docs/opendj/3/admin-guide#repl-dedicated-servers
As Ludo pointed out a DS must be able to reach an RS or it will reject all updates.
However, I wonder if your issue is what we were seeing recently: We had 2 DS with 2 RS and when our 2nd DS would go down, at some point thereafter reads worked against our 1st DS but not writes, as our 1st DS was not able to reach any RS’s including its very own which never went down and should have been able to keep everything working okay.
Issue was with OpenDJ 3.5.0 and 3.5.1. We opened a ticket with support and are testing a T-patch that looks very promising. We still want to test more though.
–Nikolaos
You must be logged in to reply to this topic.
