[ ssic-linux-Bugs-1704047 ] CFS non-init nodes missing other non-init node\'s /dev

Bugs item #1704047, was opened at 2007-04-19 23:12
Message generated for change (Settings changed) made by rogertsang
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=405834&aid=1704047&group_id=32541

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Filesystem
Group: None
>Status: Closed
Resolution: Fixed
Priority: 5
Private: No
Submitted By: Roger Tsang (rogertsang)
Assigned to: Roger Tsang (rogertsang)
>Summary: CFS non-init nodes missing other non-init node\'s /dev

Initial Comment:
Seeing this bug in 2.6.11-ssi, but not sure if this goes back to previous SSI-1.9 releases.

When a node joins the cluster only the init node mounts the node's udev at /cluster/node{nodenum}/dev.  The other nodes in the cluster that are UP before this node silently fail to mount the node's udev.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-05-22 20:13

Message:
Logged In: YES 
user_id=1246761
Originator: YES

After testing with more nodes, the initial bugfix needs to be revised. 
Only the secondary CLMS master and CLMS master get notified.

Though cfs_notify_all notifies all nodes, the ICS client code that handles
the RPC refuses to send to non-ICS UP nodes.  ICS client thinks these
non-init nodes have not joined the cluster because the client is in CLMS
pre-root and does not pull the list of nodes from the CLMS coordinator
until CLMS post-root.

The revised fix is to simply have the node tell other nodes to discover
non-NFS mounts after CLMS post-root and before transitioning to UP state. 
They will discover this node's UDEV mount.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-04-21 21:48

Message:
Logged In: YES 
user_id=1246761
Originator: YES

Still need to checkin bug fix.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-04-21 19:45

Message:
Logged In: YES 
user_id=1246761
Originator: YES

Testing cfs_notify_all all possible nodes in the cluster for udev.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-04-21 18:01

Message:
Logged In: YES 
user_id=1246761
Originator: YES

Minor correction: all OpenSSI nodes mount /dev before CLMS post-root.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-04-21 17:58

Message:
Logged In: YES 
user_id=1246761
Originator: YES

This is possible because non-CLMS master OpenSSI nodes mount /dev before
CLMS post-root.  CLMS post-root synchronizes the CLMS nodelist.  So when we
mount /dev before post-root in OpenSSI cfs_notify_all looks up the CLMS
nodelist and only knows to notify the CLMS master.  Some instrumentation
shows the other OpenSSI nodes (non-init nodes) did not get notified.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=405834&aid=1704047&group_id=32541

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >>  http://get.splunk.com/