8.0-RC1: Some ZFS issues

[Borja Marcos]

Hello,

I'm still running into some issues with FreeBSD 8-RC1, using ZFS.  
Maybe we are abusing ZFS a bit, and I will be glad to help diagnosing  
the problem.

Right now I'm trying to reproduce it and get some diagnostic  
information. A brief description follows.

We have a couple of machines running 8.0RC1,
FreeBSD ppll-publi1.XXXXX.com 8.0-RC1 FreeBSD 8.0-RC1 #4: Tue Sep 22  
10:43:16 CEST 2009     root at ppll-publi1.XXXXXX.com:/usr/obj/usr/src/ 
sys/PPLL  amd64


and we are using ZFS to replicate content to them.

They are an active/passive arrangement using Heartbeat to set up ftp  
and NFS services. The contents to be exported by NFS are updated by  
ftp, everything works in the active system. We replicate some datasets  
from the active to the passive system using zfs send/zfs receive. We  
schedule the replications to be made (right now) each 2 minutes.

Apart from the stability problems we have seen in FreeBSD 7 (we know  
ZFS is experimental) at least none of the replications (and we have  
been doing this for months) has corrupted anything.

Now we have updated to 8.0RC1, updated pools and datasets to the v13  
layout, and still running into ONE problem. If I understood well, with  
the latest ZFS version it's possible to apply a zfs receive to a  
dataset that is actually being accessed.

I've tried, and if I have heavy read access to the dataset *and* at  
the same time I'm receiving an incremental snapshot on it, sometimes  
there seems to be a deadlock and I loose access to the whole ZFS pool.

Something like:

ZFS dataset, say pool/zfsdataset

While I'm using rsync (only purpose: heavy I/O access) to copy the  
contents to another dataset, I usue a zfs receive to synchronize it  
with the latest version.


I was using a no witness, no debug kernel because I needed to know how  
well it works, but I will try to get debug information. Please let me  
know if something in particular is needed.







Borja.