Re: more problems with NFS. sort of repeatable problem with vmplayer

[Charles Hedrick <hedrick@rutgers.edu>]

Thanks. We’re now into classes, so our priorities are supporting students. Unfortunately this isn’t the only issue we have to resolve.

 I’ll try to get this info, just not today. We asked the course using vmplayer to put their VM in local storage, or to use kvm. So far no problems. The default kvm setup doesn’t use direct.

Does direct mode get turned into direct mode on the server? If so I should look into the state of direct support in the version of ZFS we have.

> On Oct 14, 2021, at 6:28 PM, NeilBrown <neilb@suse.de> wrote:
> 
> On Wed, 06 Oct 2021, Charles Hedrick wrote:
> 
>> 
>> It tried running vmplayer.  Shortly after starting to create a new VM,
>> vmplayer hung.  I had another window with a shell.  I went into the
>> directory with the vm files and did “ls -ltrc”.  It didn’t quite hang,
>> but look about a minute to finish I also saw log entries from VMware
>> complaining that disk operations took several seconds. 
> 
> Useful information to provide when a process appears to hang on NFS
> include:
> 
> - cat /proc/$PID/stack
> 
> - rpcdebug -m nfs -s all; rpcdebug -m rpc -s all ; sleep 2 ; 
>   rpcdebug -m rpc -c all; rpcdebug -m nfs -c all
> then collect kernel logs
> 
> - tcpdump -w filename.pcap -s 0 -c 1000 port 2049
>  and compress filename.pcap and put it somewhere we can find it.
> 
> - trace-cmd record -e 'nfs:*' sleep 2
>  trace-cmd report > filename
> 
>> 
>> We’re probably an unusual installation.  We’re a CS department, with
>> researchers and also a large time-sharing environment for students
>> (spread across many machines, with a graphical interface using Xrdb,
>> etc).  Our people use every piece of software under the sun. 
> 
> Probably not all that unusual.  There certainly are lots of large and
> varied NFS sites out there.
> 
>> 
>> Client and server are both Ubuntu 20.04. Server is on ZFS with NVMe storage.
> 
> If it is possible to reproduce without ZFS, that would provide useful
> information.
> I don't think it is *likely* that ZFS causes the problem, but neither
> would I be surprised if it did.
> 
> NeilBrown