[lustre-discuss] "Not on preferred path" error
ball at umich.edu
Tue Sep 20 10:28:33 PDT 2016
Stabbing in the dark, but this sounds like a multipath problem. Perhaps
you have 2 or more paths to the storage, and one or more of them is down
for some reason, perhaps the hardware itself, perhaps a cable is
pulled.... You could look for LEDs in a bad state.
I always find it instructive to reboot such a system and watch what
comes up on the console during the startup.
On 9/20/2016 12:29 PM, Joe Landman wrote:
> On 09/20/2016 12:21 PM, Lewis Hyatt wrote:
>> We do not know if it's related, but this same OSS is in a very bad
>> state, with very high load average (200), very high I/O wait time, and
>> taking many seconds to respond to each read request, making the array
>> more or less unusable. That's the problem we are trying to fix.
> This sounds like a storage system failure. Queuing up of IOs to drive
> the load to 200 usually means something is broken elsewhere in the
> stack at a lower level. Not always ... sometimes you have users who
> like to write several million/billion small ( < 100 byte ) files.
> What does dmesg report? Try to do a pastebin/gist of it, and point it
> to the list.
> Things that come to mind are
> a) offlined RAID (most likely): This would explain the user load, and
> all sorts of strange messages about block devices and file systems in
> the logs
> b) A user DoS against the storage: usually someone writing many tiny
> There are other possibilities, but these seem more likely.
More information about the lustre-discuss