DM-Multipath setup for ORacle RAC Environment
For some reason as ORacle is running the multipath daemon seeem to be croaking on the i/o. I get path failures to 3 of the 4 paths and they bounce from failed running, active running. Once I stop RAC the failures dont happen anymore. THis is RHEL 6.8 unpatched (unable to get access to internet), I do have the path_grouping_policy set to multibus. when the policy is set to default I get no problems. All hardware is new, and dont see any failures of any sorts.
Responses
Is the multibus path_grouping_policy appropriate for your storage hardware?
Some storages will require that the disks are accessed primarily via a particular path or paths. Usually this means the storage system does not have all its storage controllers fully active simultaneously, but has some kind of active/standby scheme going on. When switching to the paths connected through the standby controller, the storage system then needs to execute a controller failover, which can block the I/O activity for a short time.
With such storage systems, using the multibus path_grouping_policy will cause frequent storage controller failovers, as the multibus policy assumes all the paths can be used simultaneously. This will ruin the storage performance - not only for you, but for all the users of the storage system. Your symptoms look like this might be happening.
The "multibus" policy should only be used with storages with capability to have all I/O paths simultaneously active. For active/standby systems, you should probably set the group_by_prio policy, and set the correct path priority mechanism (the "prio" setting) - ALUA is the standardized way to discover the appropriate path priorities, but there are several vendor-specific mechanisms that predate the ALUA standard.
When group_by_prio is used and the correct path priority mechanism is selected, you should have some paths "active running" and others "enabled running". The number of "active running" paths should usually be at least equal to the number of HBAs your server has for connecting to that particular LUN.
I understand RedHat cooperates with the storage vendors to include the recommended multipath settings for each storage system model into the built-in defaults of DM-Multipath as soon as possible. But if your storage system is newer than your current patch level, or e.g. storage firmware updates have enabled new features, then it might be necessary to specify the path_grouping_policy and other multipath settings manually. In such cases, you (or your storage administrator) should choose the appropriate settings based on the capabilities of the storage system and the vendor's latest compatibility information. If you're uncertain, relying on the DM-Multipath built-in defaults is usually much better than trying to blindly guess the correct settings. The "multibus" policy is not always the best or right choice.
Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.
