On Tue, Feb 16, 2016 at 11:57 AM, Varada Kari <Varada.Kari@xxxxxxxxxxx> wrote:
> Hi all,
>
> Apologies for the long mail.
>
> We are testing the latest jewel branch with kernel RBD in our test
> clusters. When running some benchmark tests with various tools like
> vdbench and medusa, we are facing io's getting timed out and some
> messenger issues when osd's in a particular node are not reachable(crashed).
>
> we simulated the same problem by killing the osds in a specific node.
> Here is some information about the use case what we are following.
> please note, steps listed are used for reproducing the problem.
>
> [...]
>
> Even retires seemed to be replied back to the client, but client is
> handling the request or not received by the libceph?
> Enabled logs on libceph end find any handle_reply messages, but they are
> present in the kernel logs for this tid, except some keep alives been
> sent to the osd. Have tried with latest stable kernel(4.4.1) to check if
> any recent fixes resolves the issues, but problem still persists.
I assume you mean aren't present? If you have a matching set of osd
and kernel client logs, please compress and upload them somewhere so
I can take a look.
Could you try 4.5-rc4? There is a related fix in 4.5-rc which isn't in
4.4.1 (although it's been scheduled for 4.4-stable). 4.5-rc4 would also
be easier to debug.
Thanks,
Ilya
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
