search

ceph / ceph-medic

find common issues in ceph clusters
MIT License
22 stars 18 forks source link

checks: rgw_num_rados_handles should be 1 #79

Closed haklein closed 5 years ago

haklein commented 6 years ago

It'd be nice to have a check for rgw_num_rados_handles. It shouldn't be set to values higher than 1. It will cause a lot of memory growth in RadosGW at little benefit. Also possible races in multisite.

mattbenjamin commented 6 years ago

+1

alfredodeza commented 6 years ago

@haklein do you know what the command would be to check for this?

theanalyst commented 6 years ago

Isn't 1 too less a value, for clusters with decent sizes shouldn't 2-4 be ok?

haklein commented 6 years ago

depends a bit on sync/async, num of OSDs, etc. But with virtualised or containerised RGWs scalability would preferably achieved via multiple RGWs anyway.. @oritwas mentioned possible multisite issues with >1 and huge number of objects. @alfredodeza: something like that should do per RGW instance:

# ceph daemon client.rgw.rdh-cu-8 config show | grep rgw_num_rados_handles
    "rgw_num_rados_handles": "1",

But preferably this would be checked via json parsing. Depending on the discussion with @theanalyst we might consider just printing a warning.

alfredodeza commented 5 years ago

I ended up using ceph --admin-deamon /path/to/socket config show, which works great