On Wed, 2017-10-25 at 16:39 -0400, Benjamin Coddington wrote:
> Hi Trond, thanks!
>
> This looks good, but I'm not sure it's fixed things up for me. It
> has
> definitely made my test slower to complete. My test is a bit of
> bash:
>
> set +o errexit
> i=0
> while true
> do
> i=$(($i+1))
> echo run $i $(date)
> xfstests/src/t_mtab 50 &
> xfstests/src/t_mtab 50 &
> xfstests/src/t_mtab 50 &
> wait
> done
>
> If I stick a printk before and after the wait_on_bit, I can see that
> most of
> the waits are using the full 5 second timeout rather than being
> awakened
> by
> a stateid update, which is probably the source of the slowdown.
Why would the server be taking more than 5 seconds to send a reply (or
why would the client take more than 5 seconds to process it)?
> At some times, all three processes are sleeping in wait_on_bit.. I
> need
> to
> dig into why that's happening.
Yes, that makes no sense. Are you sure that your server is following
the RFCs to the letter on that 'stateid seqids must be initialised to
1' rule?
--
Trond Myklebust
Linux NFS client maintainer, PrimaryData
trond.myklebust@xxxxxxxxxxxxxxx
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
