VxFS volume - 99% full - SCSI tran_err

• Previous message: Andrew Luande ALT: "Tar extract"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Date: Sat, 7 Feb 04 17:47:45 +0100
To: <sunmanagers@sunmanagers.org>

Hi Sun specialists,

I'm faced with the following situation: I have a Fire 280 in use with
Solaris 8, Veritas VolumeManager and FileSystem 3.5 and a 4 TB hardware
RAID system connected via two Fibre Channel links and Qlogic HBAs (by
using Veritas VxDMP multipathing).
All of the RAID capacity is tied together via VxVM (Stripe over 4 LUNs
w/1 TB ea.) and this single volume than formatted with VxFS.

The system worked fine for months. Since the time the file system is
become full (around 98-99%) I'm seeing VxFS messages in the log like

vxfs: mesg 001: vx_nospace - /dev/vx/dsk/datadg/raid file system full (1
block extent)

which I understand - but also I'm seeing various times tran_err's from
the SCSI driver:

SCSI transport failed: reason 'tran_err': retrying command

All these tran_err's are commented with "retrying command" so I think no
real data error has occured until yet. Some of the tran_err messages are
syncing with the vx_nospace messages, but not all. Also not all of the
vx_nospace messages are syncing with a daily backup process which copies
hundreds of GB onto this file system.

Since the tran_err's are occuring since the file system is virtually
full, and there where no messages like this for months before, I'd bet
that there is a conjunction between.
I've tried to find information on Sunsolve as well as Veritas Support but
didn't found anything except a plain vx_nospace explanation.

My theory:
Since there is actually no entry for throttling down the queue length
"sd_max_throttle" in /etc/system and Solaris is seeing a total of 8 LUNs
(4x 1 TB ea. on 2 paths) the maximum load of SCSI commands can reach 2048
(8x 256) - but this is too much even for a fast RAID controller like the
Eon2510 we are using.
The missing link I'm like to ask you about is the logical conjunction
between a virtually full VxFS file system and the occurence of such a
high SCSI command onto the volumes' LUNs.

Is there like any like - let me say - automatic defragmentation process
with VxFS when running out of space???

Please send me your ideas - I will summarize.

Best regards,

Sven Heydecke
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers

• Previous message: Andrew Luande ALT: "Tar extract"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]