[Ocfs2-users] OCFS2 strage freezes

[Sunil Mushran]

stat-ing a file makes the fs acquire a readlock (PR) forcing the other node
to downgrade from writelock (EX) which then forces it to flush the dirty 
pages
to disk. Say you have 500M to flush and your iscsi setup's io thruput is 
30MB/s.
That'll be 16 secs.

Note: The nodes are not frozen. The ls and cp processes are waiting on the
pages to flush to disk.

You can confirm this by repeating the test and doing a "find ." instead 
of ls.
The former will not stat the file (will only stat the directory) and 
thus avoid
the force flush seen above.

Yuri Kirsanov wrote:
> Good day, everyone.
>
> I have SAN server build with Openfiler OS, with iSCSI mode turned on. I have 
> two nodes, which connect to that server via iSCSI, using one of two active 
> iSCSI partitions. I've installed ocfs2 1.3.3 with kernel 2.6.23.1, configured 
> it, made ocfs2 partition and was successful in mounting it on both nodes. 
> Everything works just fine, I can upload file from one node and delete it 
> using another one, except for one strange thing. When I'm using ocfs2 mounted 
> partition from one node, for example, uploading 1Gb test file into it, and I 
> try to access that partition from another node - just to read directory, for 
> example, - both nodes "freeze" for about 15-20 seconds, both operations just 
> stops for a while, and then copying continues on first node, while second one 
> gets directory listing.
>
> I have no errors on both nodes, neither iSCSI ones, no ocfs2. All three 
> servers are joined together using 1Gbit/sec network switch and there's no 
> other machines on that switch.
>
> If you need any information, please feel free to ask me about it. Thanks in 
> advance!
>
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>