[lkml]   [2005]   [Jul]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [Linux-cluster] [RFC] nodemanager, ocfs2, dlm
On Tue, Jul 19, 2005 at 05:48:26PM -0700, Mark Fasheh wrote:
> For OCFS2 that would mean that an ocfs2_nodemanager would still exist,
> but as a much smaller module sitting on top of 'nodemanager'.

Yep, factoring out the common bits.

> So no port attribute. The OCFS2 network code normally takes port from the
> node manager in order to determine how to talk to a given node. We'll have
> to figure out how to resolve that. The easiest would be to add 'port' back,
> but I think that might be problematic if we have multiple cluster network
> infrastructures as we do today.

The port is specific to the component using it (ocfs2, dlm, etc), so
defining port as a node property doesn't make sense if nodemanager is
providing node info to multiple components.

> Another way to handle this would be to have userspace symlink to the node
> items as an attribute on an ocfs2_tcp item. We could store 'port' as a
> second attribute. This would have the added benefit of pinning node
> information while OCFS2 uses it.

I expect each component will probably use another per-node configfs object
for component-specific attributes, using the common bits from the
nodemanager object.

> > + char nd_name[NODEMANAGER_MAX_NAME_LEN+1];
> An accessor function for this would be nice for pretty prints - maybe strcpy
> into a passed string.


> > + int nd_nodeid;
> This definitely won't work with OCFS2... Nodeid (what used to be called
> node_num) needs to be unsigned. Otherwise this will break all our nodemap
> stuff which uses a bitmap to represent cluster state.


> > + struct list_head nd_status_list;
> What are these two for? They don't seem to be referenced elsewhere...

Missed ripping them out with the other ocfs-specific stuff.

> > + if (!tmp && cluster->cl_has_local &&
> > + cluster->cl_local_node == node->nd_nodeid) {
> > + cluster->cl_local_node = 0;
> I think we might want to be setting cl_local_node to NODEMANAGER_MAX_NODES
> here. It seems that ocfs2_nodemanager also does this so we might have just
> caught a bug you inherited :)


> You removed o2nm_configured_node_map but we need some sort of method for
> enumerating over the set of configured nodes.
> Also we need a method for querying the existence of a node.
> The OCFS2 code usually uses o2nm_get_node_by_num(..) != NULL for this but a
> simple boolean api call would be cleaner and would avoid exposing the node
> structure.

Right, those should be on the TODO.


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-07-20 06:15    [W:0.057 / U:0.172 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site