zfault.sh fails hard read error raid0

[ryantig]

/usr/src/zfs-0.6.0-rc1$ sudo /usr/libexec/zfs/zfault.sh -cv
Destroying /dev/md0
/dev/md1
/dev/md2
/dev/md3
Creating /dev/loop0 using /tmp/zpool-lo.NgPs9GyN
Creating /dev/md0 using /dev/loop0
Destroying /dev/md0
Destroying /dev/loop0
raid0 raid10 raidz raidz2 raidz3
1 soft write error Pass Pass Pass Pass Pass
2 hard write error Pass Pass Pass Pass Pass
3 all write errors Pass Pass Pass Pass Pass
4 soft read error Pass Pass Pass Pass Pass
5 hard read error Fail (33)

[behlendorf]

So most of the zfault.sh bugs I've seen have been in the test suite. Injecting exactly the right fault at the right place and time is surprisingly difficult. In this case error code 33 indicates we injected an error which should have caused a hard read error, but zpool claims there was no error. That could mean a lot of thing you might try running zfault.sh again and see if it's transient or reproducable.

[ryantig]

I tried 3 runs and it had error 33 each time before I submitted. Then just now I ran again.... And naturally, it didn't reproduce this time. ack.
/usr/src/zfs-0.6.0-rc1$ sudo /usr/libexec/zfs/zfault.sh -cv
[sudo] password for ryan:
Destroying /dev/md0
/dev/md1
/dev/md2
/dev/md3
Creating /dev/loop0 using /tmp/zpool-lo.62BI9Sul
Creating /dev/md0 using /dev/loop0
Destroying /dev/md0
Destroying /dev/loop0
raid0 raid10 raidz raidz2 raidz3
1 soft write error Pass Pass Pass Pass Pass
2 hard write error Pass Pass Pass Pass Pass
3 all write errors Pass Pass Pass Pass Pass
4 soft read error Pass Pass Pass Pass Pass
5 hard read error Pass Pass Pass Pass Pass
6 fixable read error Pass Pass Pass Pass Pass
7 silent data corruption Pass Pass Pass Pass Pass
8 soft write timeout Pass Pass Pass Pass Pass
9 hard write timeout Skip Pass Pass Pass Pass

[behlendorf]

Still, it sounds like a flaw in the test case I'll need to run down.