On 09/28/2010 02:04 PM, Sean S wrote:
This is a continuation of my older post here:
That thread had gotten a little off topic so I decided to start a new
I'm experience a random disconnect of my iSCSI drive on a system using
an iSCSI partition as the root partition. I'm currently running with
open-iscsi-872 on a CentOS 5.1 machine. (2.6.18-53 kernel).
I was able to capture a wireshark log of the disconnect. I'm pasting
it inline below (sorry about the line wrap, if theres a better way to
post it let me know). .40 is the initiator and .13 is the target. It
appears the last thing that happens is the initiator sends a write to
the target(13473) the target successfully completes the write and
sends the "good" response (13476) and then the initiator never
responds after that, not even sending the TCP ACK. I'm making the
assumption now that I'm looking at some sort of networking issue, and
that the iSCSI partition going down is by product of that. Does anyone
have any opinions on what's going wrong here?
Any tips on how to proceed? I've already tried updating my Ethernet
driver to the latest available from intel, but the same thing occurs.
Sorry for the late response. Could you send me the trace offlist. Just
attach it to a mail. Did you take the log from the target side? It makes
some sense with what we saw in the initiator logs you sent last time
where at some point the initiator is not able to communicate to the
target. Something is not allowing each other to talk to each other.
And this is still with IET, right? On that box do you have multiple
nics, if so could you try dm-multipath? Just discover the IET target and
it should return both portals (in older versions you might have to do
discovery to both portals instead of just one). Then on the iscsi side
if you set the node.startup to automatic we will log into both. Then
just run /sbin/multipath and that will create a multipath device.
I think for Centos though, you need the current version to do root over
dm-multipath over iscsi, and also with that it will do the multipath for
you, so it is a lot easier. This way if one path goes bad the OS should
be able to use the other one and we can see the logs and do some tests
like just a ping through the interface that was not working.
You received this message because you are subscribed to the Google Groups
To post to this group, send email to open-is...@googlegroups.com.
To unsubscribe from this group, send email to
For more options, visit this group at