should we run sshd at higher priority on DUTs and devservers? |
||
Issue descriptionWe see occasional test failures in the lab from ssh errors, some of which may be due to timeouts. I am pretty sure that most of these timeouts are NOT due to poor scheduling of sshd. But on the other hand, sshd is always working on behalf of some other process, and rarely doing long operations on its own (I am not sure about some of the key verifications, but I suspect that they are dominated by network latency). Thus I don't see any contraindications to giving sshd some negative nice (i.e. some mean), and it may have benefits.
,
Nov 28 2016
Right. It doesn't hurt, but likely doesn't help either. The special case that made me think of this is the errors we seem to get when we run ssh <DUT> "touch <some-file>; reboot" It *seems* that in a number of cases the ssh command fails. In some experiment, I can make ssh fail (i.e. hang) with ssh <DUT> "reboot; sleep <T>" with T = 0.1s. It works with T = 0.05s. So it's tight, and I am wondering if there are other situations in which we don't sufficiently prioritize sshd. However: 1. It's not clear nice would help because there are other operations, like flushing dirty buffers to disk, which also compete 2. There are other ways of dealing this particular case. So maybe it's not a good idea. I enjoy using the bug tracker for my streams of consciousness.
,
Dec 5 2016
Wouldn't help. |
||
►
Sign in to add a comment |
||
Comment 1 by vapier@chromium.org
, Nov 22 2016