Hi,
We have a system running Ceph Pacific with a large number of delete
requests (several hundred thousands files per day) and I'm investigating
how can I increase the gc speed to keep up with our deletes (right now
there are 44 millions of objects in the gc list).
I changed max_concurrent_io to 40 , rgw_gc_max_trim_chunk to 1024,
rgw_gc_processor_max_time to 300, rgw_gc_processor_period to 300 ,
rgw_gc_obj_min_wait to 300, rgw_gc_max_objs to 1000 but I managed only
to stall the increase of the gc queue.
I'm curious about these parameters :
- if I understand the docs correctly, gc_max_objs = maximum number of
objects that may be handled by garbage collection in one garbage
collection processing cycle. Soif the gc cycle is set to 5 minutes =>
for every 24 hours we have 288 cycles => the max number of objects that
can be deleted by gc for ever 24 hours is 288 * 1000 = 288000 ? If so,
this is the total number per cluster or per rgw instance (we have 30 rgw
containers on different machines) ?
- 44 millions of objects in gc list => this means that we have ~160TB in
the garbage queue considering one object = 4MB ?
- how can I increase the gc speed beyond the current speed ? my current
settings are quite aggresive already
- is the process of increasing the pg_num/pgp_num impacting the gc speed
? I'm asking because we are doing that for several weeks, manually
increasing the pg_num in small steps
Thanks.
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
