[Gluster-devel] quota.t hangs on NetBSD machines

[Emmanuel Dreyfus]

On Thu, Dec 31, 2015 at 02:51:41PM +0530, Raghavendra Talur wrote:
> To our surprise though, the hung test started proceeding.

You mean a process gets stuck into a system call for hours and then 
is able to ascape?

Some hints:

1) ps -axl shows the waiting channel (WCHAN column) for a process stuck
in kernel. What is it?

2) crash is a kernel debugger that can be used while running multiuser.
Of course since the system is running, the output is obsolete most 
of the time, but for a stuck process we can extract valuable information.

Run crash from the shell, then inside crash, run the ps command. Find the
relevant process and note the address in the STRUCT LWP * column. For an
example, let us say it is c63452a0.

bt/a c63452a0 will produce a kernel backtrace for the process. This can
be extremely valuable to understand hat is going on. If we are awaiting 
for a lock, we can track what process is holdoing it.

 
-- 
Emmanuel Dreyfus
manu at netbsd.org