

I have the following log entries that seem to show the OSD process attempting to descend into the PG directory on disk and create an object list of some sort: I have a cluster running Luminous 12.2.12 with Filestore and it takes my OSDs somewhere around an hour to start (They do start successfully - eventually). Subject: Luminous 12.2.12 - filestore OSDs take an hour to boot Subject: Re: Luminous 12.2.12 - filestore OSDs take an hour to bootįor what it's worth - all of our objects are generating LONG named object files like so.

I plan on testing this same scenario with BlueStore to see if it's also susceptible to these boot / read issues.

During boot the OSD goes out to lunch for increasing amounts of time based on the number of objects on disk you have that meet this criteria (With 2.4 million ish objects that meet this criteria, the OSD takes over an hour to boot).
Mirrorsync drupal full#
We have a cluster where 99% of the objects have names longer than N (220+?) characters such that it truncates the file name (as seen below with "_0_long") and stores the full object name in xattrs for the object. It does appear that long file names and filestore seem to be a real problem.
