SCSI read write errors - disk failed

[Jonathan Caplette_1]

Hi there!

Since a few days, I've seen in my syslog errors related to SCSI... I'm running HP-UX 11i, I've check with ioscan the disk, fc, tape, ioscan alone with no classes, everything is CLAIMED. So I tried to fcmsutil the fiber cards they are ONLINE and I pvdisplay and diskinfo the disks related in my error (attached as my syslog error... dev_t=0x1fxxxxxx) all the disks listed as POWERFAILED respond to my pvdisplay and diskinfo!! What could I expect to give me that error???

Thanks
Jonathan

[Sundar_7 · Accepted solution]

Jonathan,

I believe PV timeout is the issue here. are your disks on a storage enclosure (EMC?) ?.

I get this errno=126 all time :-).

By default PV timeout is set to 30 secs. LVM queues the IO for the disk and if the disk did nt process the IO request within 30 secs, it assumes the LUN has gone bad and logs read/write errors.

PV TIME OUT is also the reason for those POWERFAILED error messages.

you can use pvchange command to set the PVTIMEOUT to a higher value.

man pvchange and look for -t option.

-- Sundar.

[Jonathan Caplette_1]

Sundar, and everyones,

I forgot, those disks are on a SAN.

I'm gonna check for the timeout option... Any other ideas???

Jonathan

[Jeff Schussele]

Hi Jonathan,

As well as pvchange -t to change timeout on the PV level, you can use lvchange -t to change it at the LV level.
Default is 30 seconds - bump it to 60 to give you headroom under *heavy* disk I/O loads.

Rgds,
Jeff

[Sundar_7]

any other ideas ?? -- hmm, try out the timeout settings - I am positive it will solve the problem.

[Jonathan Caplette_1]

Is this error can cause performance issues??? And setting the timeout to 60 sec, will it affect the performance of the disks access???

[Sundar_7]

Nope, it is not going to cause performance problems.

Reducing the timeout value from 30 secs to 10 secs doesnt mean you will get the IO done faster :-).

It just defines the timeout for LVM after which it can conclude the IO to the LUN has failed.

[Jonathan Caplette_1]

Thanks for the advise!! I'll check the timeout on pv and lv's!!

Ciao!

[Jonathan Caplette_1]

I just got another question!! My disks that gave me that error, is on an HDS (Hitachi Data System) SAN, do I have some specific timeout option to set on my PV??? Or these options are generic??

[Sundar_7]

The timeout value is generic though the optimal timeout value is dependent on the hardware configuration of the SAN box (like the amount of cache it has, the amount of IO traffic and the number of channel adapters etc).

But here all you can do is to hit-and-trail.

Set the timeout to 60 and observe. If you are still seeing the read errors or POWERFAILED messages for the PV then you need to increase the PV timeout.

At 60, if there are no errors, you could experiment by lowering the timeout value.