[lustre-discuss] writes not going to new osts on new oss (lustre 2.5.3)
strosahl at jlab.org
Wed Jun 17 05:44:37 PDT 2015
A follow up,
I found a work around: https://lists.01.org/pipermail/hpdd-discuss/2014-November/001373.html
It instructed me to do the following:
[mds] echo 100 > /proc/fs/lustre/lod/<fsname>-MDT0000-mdtlov/qos_threshold_rr
Before I did that I checked, qos_threshold_rr was set to 17%, after echoing 100 into it the value went to (predictably) 100%.
----- Original Message -----
From: "Kurt Strosahl" <strosahl at jlab.org>
To: lustre-discuss at lists.lustre.org
Sent: Wednesday, June 17, 2015 7:22:04 AM
Subject: writes not going to new osts on new oss (lustre 2.5.3)
An odd issue has come up with my new lustre 2.5.3 system. I added a new oss yesterday with three osts, but the mds doesn't seem to want to write to them. I've forced it to do so by setting stripes on test directories and writting data there. I've also (since the lustre 2.5.3 system is rather small) set all other osts to read only mode. Once that is done the writes do start to go to the new oss, but only the first ost on it. So I continued set that ost offline and the writes moved to the next one, the I set that ost to read only, so now only one ost was available, which received writes.
Note that the three osts above the new one were set offline earlier in the week due to extremely low performance on their host oss...
So it was:
3 osts set read only
3 new osts that the system won't write to unless it has no other options.
I'm running lustre 2.5.3, with a zfs backend for the osts.
More information about the lustre-discuss