Filesystem goes read-only when LVM mirror leg is lost and mirrorlog type is "mirrored" on Red Hat Enteprise Linux 6
Issue
- When the cluster was started, and the service was at site-1, I simulated a split in the cluster by:
- Starting the iptables on both nodes (blocking UDPU heartbeats and the ability to fence the other node).
- Unpresenting/unallocating storage from the alternate site.
- Since the shared volume group is mirrored with a LUN at each site, both nodes should be able to bring up cluster service on their site-local LUN.
- However, when running fence_ack_manual, node1 sharedvg goes readonly and I have to re-enable the service to bring it online. Node2 fails to bring the service online altogether.
Environment
- Red Hat Enterprise Linux Server 6
- May occur on Red Hat Clusters but could also occur on standalone systems.
- LVM2 logical volume management
- At least 1 LVM2 mirrored device with at least 2 legs
- mirrorlog is type "mirrored" (as opposed to "corelog" or "disk" mirrorlogging)
- Legs of the mirror may be disabled/disconnected independently (ie. leg1 could become unavailable but leg2 may still be available).
- This can most commonly occur when the legs come from different SAN's or storage back-ends.
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.