Re: questions about pg_log mechanism

[chen jan <janchen2015@xxxxxxxxx>]

Hi all,

Sorry, I need to correct something about what I observed.  I found pg_log file was on running OSD disk, but the file size was 0, so I think maybe OSD kept all pg log data in memory.

BTW, the CEPH version is 0.80.6

Thanks,

Jan

2014-11-08 10:04 GMT+08:00 chen jan <janchen2015@xxxxxxxxx>:

Hi all,

 I'm trying to test pg_log mechanism under stress test by using a simple 3 nodes with 15 osds ceph cluster(replica size is 2). The following were test steps:

1. Set mon_osd_down_out_interval to 2 days 

2. Using at least 10 threads FIO and librbd to send random 4KB r/w IOs continuously.

3. Shutdown one node, that means 5 osds were down, and pg_log mechanism started to work.

After running 1.5 days, I observed two results:

1. There were no pg_log file on all running OSD disk

2. Almost every running OSD process occupied up to 1 GB memory.

So my question was: It seemed like OSD kept all pg_log data in memory, was that true? If not, could someone help to explain what happen to this? 

BTW, XFS was used in this environment.

Thanks,

Jan

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com