With a k+m which is 3+2 each RADOS object is broken into 5 shards. By default the pool will have a min_size of k+1 (4 in this case). Which means you can lose 1 shard and still be >= min_size. If one host goes down and you use a host-based failure domain (default) you will lose 1 shard out of all PGs on that host. You will now be at min_size and so still readable/writeable. If you lose another host you will now be below min_size with 3 healthy shards for some subset of PG (those common to the 2 hosts) will be inactive and therefore not read/writeable. As you can see, the higher your M the more disks/hosts you can lose before dropping below min_size. Respectfully, *Wes Dillingham* wes@xxxxxxxxxxxxxxxxx LinkedIn <http://www.linkedin.com/in/wesleydillingham> On Mon, Nov 27, 2023 at 1:36 PM <tranphong079@xxxxxxxxx> wrote: > Hi Groups, > > Recently I was setting up a ceph cluster with 10 nodes 144 osd, and I use > S3 for it with pool erasure code EC3+2 on it. > > I have a question, how many osd nodes can fail with erasure code 3+2 with > cluster working normal (read, write)? and can i choose better erasure code > ec7+3, 8+2 etc..? > > With the erasure code algorithm, it only ensures no data loss, but does > not guarantee that the cluster operates normally and does not block IO when > osd nodes down. Is that right? > > Thanks to the community. > _______________________________________________ > ceph-users mailing list -- ceph-users@xxxxxxx > To unsubscribe send an email to ceph-users-leave@xxxxxxx > _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |