Hide Forgot
Description of problem: osd_journal_size is currently set to 1GB in OSP-d while Ceph's default is 5GB. This RFE is to align OSP-d's osd_journal_size value to Ceph's default. source: https://github.com/ceph/ceph/blob/hammer/src/common/config_opts.h#L492 Version-Release number of selected component (if applicable): 7
This bug did not make the OSP 8.0 release. It is being deferred to OSP 10.
Hi Angus, I could try to drive this change via upstream getting TripleO providing sensible defaults to the deployment. If you are ok with that feel free to assign the bug to me and I'll start working on it.
Regarding the upgrade thing, if we change the default value, we need to orchestrate the upgrade in THT upgrade scripts, so increase the journal size before puppet runs, so when puppet runs, journal size is already good. Can we have Ceph experts on this task? Can we scale-up the journal size without downtime? Thanks
After reading into ceph documentation[1] I'm not convinced that we should change the default. I do not know where that 5GB default is from that is mentioned in the bug description but current default for journal size is 0 and there is clear note that the required journal size is based on the deployment characteristics and should be calculated based on those. the current 1GB might be something that works in our CI well and every single deployment should overwrite this based on the 'osd journal size = {2 * (expected throughput * filestore max sync interval)}' calculation. Perhaps there would be rather documentation change in place than changing the default value on the deployment scripts that should be overwritten anyways at the runtime? [1] http://docs.ceph.com/docs/master/rados/configuration/osd-config-ref/#journal-settings
Ah just realized where that was coming from. Interestingly the hammer config reference section is exactly the same claiming the default being 0 http://docs.ceph.com/docs/hammer/rados/configuration/osd-config-ref/#journal-settings
That's likely a documentation error. It's been to 5GB for a while now (https://github.com/ceph/ceph/blob/hammer/src/common/config_opts.h#L495) Things may have changed a bit since the last time I looked into it but my understanding of this variable is that it will only be used to create a journal partition of osd_journal_size MB when running ceph-disk prepare and used when creating the Ceph journal (ceph-osd -i <ID> --mkjournal). Changing this value won't matter for existing OSDs.
Thanks Alexander, So if I understand correctly this should be really straight forward change to make not affecting the upgrades.
I've proposed change to the upstream to remove the overwrite of osd_journal_size. As that option affects only during the journal creation and ceph does have sensible default lets just stay on that and if ceph developers sees reason to change the default we get that benefit as well without changing the overwrite again.
The change merged into master. Bug opened for backporting into mitaka.
Quality Engineering Management has reviewed and declined this request. You may appeal this decision by reopening this request.
Adding to clarify, as comment 12 has continually confused me. This is already fixed in OSP9, therefore it is closed in the context of OSP10 and the desired functionality is present.