On ven, 2005-04-01 at 12:27 -0800, Nicholas A. Bellinger wrote: > On Fri, 2005-04-01 at 10:35 +0200, Christophe Varoqui wrote: > > On Thu, Mar 31, 2005 at 03:52:11PM -0800, Nicholas A. Bellinger wrote: > > > > > > ... > > > > > > IO never goes back across iSCSI Channel 20. I also noticed that this > > > does not occur every time iSCSI Channel 20 is paused. Sometimes IO will > > > stop over iSCSI Channel 21 as well, but iSCSI Channel 20 can be started > > > and IO resumes correctly to both iSCSI Channels. > > > > > > So my main question is this, once the iSCSI Channel is restarted and IO > > > is allowed to flow again, how does dm-multipath know that the device is > > > unlocked and online? > > > > > It's userspace responsability to maintain the device map. > > In the current available implementation, multipathd does the path checking and reinstate offline paths when they come back up. > > > > The problem that I have been encountering is multipathd never testing > for the path (iSCSI Channel 20 in the above example) once the struct > scsi_device goes from a blocked to unblocked state. Is there anything > that I am missing in the setup in the above example? > You may be hit by a bug in multipath/main.c:pgcmp() I discovered today. You can try 0.4.4-pre7 or simply : --- multipath-tools-0.4.4-pre6/multipath/main.c 2005-03-31 11:05:32.000000000 +0200 +++ multipath-tools-0.4.4-pre7/multipath/main.c 2005-04-01 21:54:44.000000000 +0200 @@ -503,10 +512,9 @@ if (found) break; } - if (found) { + if (found) found = 0; - break; - } else + else return 1; -- christophe varoqui <christophe.varoqui@xxxxxxx>