[Ocfs2-users] DLMFS on OracleVM 2.1 (OEL5.0 based)

[Heinzmann, Robert]

Hi List, 

I want to use DLMFS of OCFS2 to avoid multiple start of virtual machines
on OracleVM. I want to use a wrapper around xm that spawns a deamon that
keeps a file open in /dlm/DOMAIN. Now I played around a bit and followed
the procedure in the document
http://oss.oracle.com/projects/ocfs2/src/branches/ocfs2-1.2/dlmfs.txt
for DLMFS. 

Theres one problem. The "O_NONBLOCK" option is not working. I have 2
nodes in my setup that have both mounted the /dlm as dlmfs and both have
mounted a 5 GB OCFS2 test volume on FS SAN.

I run the getlocks script below without O_NONBLOCK on node1 with a sleep
of - eg - 30. Then I run - while the script is running on node1 - the
same script on node2 the input appears on node2 as soon as the script
has finished on node1. Thats what I expeced. 

If I run the getlocks script with O_NONBLOCK enabled, node1 successfully
sleeps e.g. 30 sec. While the script is running I execute the getlocks
script on node2. The following ouput appears - as expected: 


--- OUTPUT ----
Traceback (most recent call last):
  File "/INSTALL/getlocks.py", line 12, in ?
    fd=os.open(dir + "/" + mach, os.O_RDWR|os.O_CREAT|os.O_NONBLOCK)
OSError: [Errno 26] Text file busy: '/dlm/xen/TermServ'
--- OUTPUT ----

Thats what I expected. But when the getlocks script is finished on node1
and I run getlocks on node2 again the ERROR above appers also. 

Shouldn't it be possible that the lock is aquired on node2 when node1
releases the FH ??? Is there something wrong in the OCFS2 DLM code or
did I get something wrong here ? 

Regards,
Robert 

--- getlocks test script ---
#!/usr/bin/python

import os
import time

dir="/dlm/xen"
mach="TermServ"

if ( not os.path.isdir(dir)):
        os.mkdir(dir)
try:
        # fd=os.open(dir + "/" + mach, os.O_RDWR|os.O_CREAT)
        fd=os.open(dir + "/" + mach, os.O_RDWR|os.O_CREAT|os.O_NONBLOCK)
        sleeptime=raw_input("How long to sleep ? ")
        print "Sleeping " + sleeptime + " seconds"
        time.sleep(float(sleeptime))
        os.close(fd)
except IOError, e:
        print "ERROR: could not open file\n"
        raise Exception, "%s [%d]" % (e.strerror, e.errno)
--- end getlocks test script ---


Does anyone have a hint here ? 

I already wrote an e-mail to mark fasheh (listed in
http://oss.oracle.com/projects/ocfs2/src/branches/ocfs2-1.2/dlmfs.txt),
but he's did not answer yet. 

Robert