We have had excellent results with multi-MDS - *after* we pinned every directory. directory migrations caused so much load that it was frequently no faster than a single MDS. This was on Nautilus at the time. The hard limit on strays is also per-MDS, so we ended up splitting to more MDS' to buy some time there. From what I can tell, snapshots don't get moved if you change a pin, at least not immediately, so keep that in mind.
On Wed, Nov 17, 2021 at 8:12 AM Eugen Block <[email protected]> wrote: > > Hi, > > in this thread [1] Dan gives very helpful points to consider regarding > multi-active MDS. Are you sure you need that? > One of our customers has tested such a setup extensively with > directory pinning because the MDS balancer couldn't handle the high > client load. In order to better utilize the MDS servers (only one > thread) they ran multiple daemons per server which also worked quite > well. The only issue is the rolling upgrade where you need to reduce > the max_mds to 1 which doesn't work here. But this is all still in > Nautilus. > > [1] > https://lists.ceph.io/hyperkitty/list/[email protected]/message/4LNXN6U5DTB2BFPBDGDUKTEB4THD7HCH/ > > > Zitat von [email protected]: > > > Dear Cephers, > > > > On reading a technical blog from Croit: > > https://croit.io/blog/ceph-performance-test-and-optimization > > > > It says the following: "It should be noted that it is still debated > > whether a configuration with multiple active MDS servers is OK for > > production Ceph clusters." > > > > Just wonder whether multiple active MDS servers is OK for production > > Ceph clusters OR Not? Does any one has practical lessons ? Can some > > one share more stories with us, whether success or failure > > > > Thanks a lot in advance, > > > > samuel > > > > > > > > > > > > [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] _______________________________________________ ceph-users mailing list -- [email protected] To unsubscribe send an email to [email protected]
