Hi Eugen. I've tried. The system says it's not recommended but I may force it. Forcing something with the risk of losing data is not something I'm going to do.
Best regards Daniel On Sat, Mar 26, 2022 at 8:55 PM Eugen Block <[email protected]> wrote: > Hi, > > just because the autoscaler doesn’t increase the pg_num doesn’t mean > you can’t increase it manually. Have you tried that? > > Zitat von Daniel Persson <[email protected]>: > > > Hi Team. > > > > We are currently in the process of changing the size of our cache pool. > > Currently it's set to 32 PGs and distributed weirdly on our OSDs. The > > system has tried automatically to scale it up to 256 PGs without > succeeding > > and I read that cache pools are not automatically scaled so we are in the > > process of scaling. Our plan is to remove the old one and create a new > one > > with more PGs. > > > > I've run the pool in readproxy now for a week so most of the objects > should > > be available in cold storage but I want to be totally sure so we don't > lose > > any data. > > > > I read in the documentation that you could remove the overlay and that > > would redirect clients to cold storage. > > > > Is a preferred strategy to remove the overlay and then run > > cache-flush-evict-all to clear it and then replace or should I be fine > just > > to remove overlay and tiering and replace it with a new pool? > > > > Currently we have configured it to have a write caching of 0.5 hours and > > read cache of 2 days. > > > > ------ > > ceph osd pool set cephfs_data_cache cache_min_flush_age 1800 > > ceph osd pool set cephfs_data_cache cache_min_evict_age 172800 > > ---- > > > > The cache is still 25Tb in size and would be sad to lose if we have > > unwritten data. > > > > Best regards > > Daniel > > _______________________________________________ > > ceph-users mailing list -- [email protected] > > To unsubscribe send an email to [email protected] > > > > _______________________________________________ > ceph-users mailing list -- [email protected] > To unsubscribe send an email to [email protected] > _______________________________________________ ceph-users mailing list -- [email protected] To unsubscribe send an email to [email protected]
