Hi,
We try to build a HA environment for a Document Management System. Before I come
to the problem I will briefly describe our setup. We have two buildings and a
fibrechannel switch in each building and attached to it a disk arrays, a DB
server for the metadata (Linux with Oracle) and a fileserver under W2K for the
bulk data. The fileserver runs under W2K, because the application server runs
under W2K. The servers are also connected to the switch in the other building.
The idea is to mirror the data over the buildings. All machines are connected
with two fibres (using QLA2202F cards) to each switch. Also the arrays are using
two connections to the switch. is true for. So we have a multiple pathes from
the machines to the arrays.
The Linux machines are using md-raidtools to mirror over the buildings. To see
the correct number of devices we have used first the failover qlogic driver 6.01
and afterwards the standard version upto 6.04beta4 with the multipath
personality of the mdtools.
The system runs till we need the failover. But if pathes are not available we
get a kernel Oops in the RAID1 personality of the md and any IO to the disk
arrays hangs forever, the machine does also not shutdown correctly. Because we
see this behaviour also if a windows machine boots and sends a LIP reset over
the fibre channel, this is even for normal operations not acceptable. Needless
to say that the W2K do not have this problem. Therefore we concluded that the
setup of the hardware is ok,(BIOS settings are the same for Linux and W2K). Of
course we are unsure whether the problem is in the mdtools or in the qlogic
driver (who should handle the LIP reset). We try to get help from Linux
companies, but we where not very successful.I could send the list Oops and more
information if it would help. Maybe it is well known problem having a raid1
personality over two multipath personalities. (The Oops says something from a
NULL pointer which he can't follow and if I understood it correctly, it happens
after all pathes are gone, due to LIP reset). The LIP reset problem was first
seen while the mirror was resyncing and a windows machine got rebooted.
Thank you for some feedback
Frank Behner
Do you see an OOPS when one of the paths of the multipath leaves, or
only when all paths are dead or in LIP.
Thanks
-steve
frank@behner.org wrote:
Hi,
We try to build a HA environment for a Document Management System. Before I come
to the problem I will briefly describe our setup. We have two buildings and a
fibrechannel switch in each building and attached to it a disk arrays, a DB
server for the metadata (Linux with Oracle) and a fileserver under W2K for the
bulk data. The fileserver runs under W2K, because the application server runs
under W2K. The servers are also connected to the switch in the other building.
The idea is to mirror the data over the buildings. All machines are connected
with two fibres (using QLA2202F cards) to each switch. Also the arrays are using
two connections to the switch. is true for. So we have a multiple pathes from
the machines to the arrays.
The Linux machines are using md-raidtools to mirror over the buildings. To see
the correct number of devices we have used first the failover qlogic driver 6.01
and afterwards the standard version upto 6.04beta4 with the multipath
personality of the mdtools.
The system runs till we need the failover. But if pathes are not available we
get a kernel Oops in the RAID1 personality of the md and any IO to the disk
arrays hangs forever, the machine does also not shutdown correctly. Because we
see this behaviour also if a windows machine boots and sends a LIP reset over
the fibre channel, this is even for normal operations not acceptable. Needless
to say that the W2K do not have this problem. Therefore we concluded that the
setup of the hardware is ok,(BIOS settings are the same for Linux and W2K). Of
course we are unsure whether the problem is in the mdtools or in the qlogic
driver (who should handle the LIP reset). We try to get help from Linux
companies, but we where not very successful.I could send the list Oops and more
information if it would help. Maybe it is well known problem having a raid1
personality over two multipath personalities. (The Oops says something from a
NULL pointer which he can't follow and if I understood it correctly, it happens
after all pathes are gone, due to LIP reset). The LIP reset problem was first
seen while the mirror was resyncing and a windows machine got rebooted.
Thank you for some feedback
Frank Behner
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi,
only if all pathes are dead. I did in the meantime change multipath.c so
that in those cases the path/disk is marked bad. Now the Oops has gone.
But I still have a SCSI error in the logs which I could not yet identify
who is sending it. The log entries are for example:
Jan 15 18:30:01 dmslsp1 kernel: SCSI disk error : host 1 channel 0 id 0
lun 0 re
turn code = 20008
Jan 15 18:30:01 dmslsp1 kernel: I/O error: dev 08:11, sector 9379848
And of course main sympthoms I/O is hanging machine does not cleanly
shutdown stay.
Can you point me whether I have to look for this in the raidtools, in
the driver or in the mid-scsi code.
BTW the kernel is 2.4.19 with a SuSE SLES8 distribution.
By Frank
Steven Dake schrieb:
Do you see an OOPS when one of the paths of the multipath leaves, or
only when all paths are dead or in LIP.
Thanks
-steve
frank@behner.org wrote:
quoted
Hi,
We try to build a HA environment for a Document Management System.
Before I come
to the problem I will briefly describe our setup. We have two
buildings and a
fibrechannel switch in each building and attached to it a disk
arrays, a DB
server for the metadata (Linux with Oracle) and a fileserver under
W2K for the
bulk data. The fileserver runs under W2K, because the application
server runs
under W2K. The servers are also connected to the switch in the other
building.
The idea is to mirror the data over the buildings. All machines are
connected
with two fibres (using QLA2202F cards) to each switch. Also the
arrays are using
two connections to the switch. is true for. So we have a multiple
pathes from
the machines to the arrays.
The Linux machines are using md-raidtools to mirror over the
buildings. To see
the correct number of devices we have used first the failover qlogic
driver 6.01
and afterwards the standard version upto 6.04beta4 with the multipath
personality of the mdtools.
The system runs till we need the failover. But if pathes are not
available we
get a kernel Oops in the RAID1 personality of the md and any IO to
the disk
arrays hangs forever, the machine does also not shutdown correctly.
Because we
see this behaviour also if a windows machine boots and sends a LIP
reset over
the fibre channel, this is even for normal operations not acceptable.
Needless
to say that the W2K do not have this problem. Therefore we concluded
that the
setup of the hardware is ok,(BIOS settings are the same for Linux and
W2K). Of
course we are unsure whether the problem is in the mdtools or in the
qlogic
driver (who should handle the LIP reset). We try to get help from Linux
companies, but we where not very successful.I could send the list
Oops and more
information if it would help. Maybe it is well known problem having a
raid1
personality over two multipath personalities. (The Oops says
something from a
NULL pointer which he can't follow and if I understood it correctly,
it happens
after all pathes are gone, due to LIP reset). The LIP reset problem
was first
seen while the mirror was resyncing and a windows machine got rebooted.
Thank you for some feedback
Frank Behner
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html