Hi everybody,

I am trying to understand why am I keep on getting remapped+wait_backfill
pg statuses, when doing some cluster pg shuffling. Sometimes it happens
just by doing small reweight-by-utilization operation, and sometimes when
I modify the crushmap (bigger movement of data).

Taking look at ceph health detail and investigating some of the pgs by
using pg #ipg_id query, I can see that all of the "acting" pgs are healthy
and of the same size.  "Up" pgs do have a pg folder created, but dont have
any data inside (empty head + TEMP).

I dont have any (near)full pgs, and ceph pg debug  unfound_objects_exist
yields FALSE.

Cluster is also 100% functional(but in WARN state), and I can see that if I
write some data, acting pgs are all happily syncing between each other.

Acting pgs

Up pgs

​
I can simply recover form this following this steps:

   - set  noout, norecover, norebalance to avoid unnecessary data movement
   - stopping all actingbackfill pgs (active + up) at the same time
   - remove empty "up" pgs
   - start all the pgs again
   - unset noout, norecover, norebalance

After that new "up" pgs are recreated in remapped+backfilling state, and
marked as active+clean after some time.

I have also tried to "kick the cluster in the head" with ceph pg
send_pg_creates (as stated here
https://www.mail-archive.com/[email protected]/msg12287.html), but
I get:

$ ceph pg send_pg_creates
Error EINVAL: (22) Invalid argument

BTW What is send_pg_creates really supposed to do?

Does anyone have some hints is this occurring?
Thank you,
Ivan

Jewel 12.2.2

size 3
min_size 2

#I havent been playing with tunables
# begin crush map
tunable choose_local_tries 0
tunable choose_local_fallback_tries 0
tunable choose_total_tries 50
tunable chooseleaf_descend_once 1
tunable chooseleaf_vary_r 1
tunable straw_calc_version 1

#standard ruleset
rule replicated_ruleset {
        ruleset 0
        type replicated
        min_size 1
        max_size 10
        step take dc
        step chooseleaf firstn 0 type host
        step emit
}
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to