[ovs-dev] [PATCH] No need to restart DB / OVS on changing dscp value.
blp at nicira.com
Fri Jun 15 10:43:00 PDT 2012
On Thu, Jun 14, 2012 at 05:31:07PM -0700, Mehak Mahajan wrote:
> I have taken care of the json connection.
> The reason to propagate the changed information to the top was in future if
> any other configurable options for the socket get added, we can just modify
> all the options and then reconnect once (though i cannot foresee any such
> parameters as of now). Hence I have taken this change out.
> However in case of the controller it may not be possible to make these
> changes unless I can call rconn_disconnect() followed directly by
> [As of now I am calling rconn_connect (I assumed that rconn_disconnect
> modifies the 'target' in some way and hence needs to be re-written to rconn
> using rconn_set_target__() before the reconnect()).
> Also it seems to be the norm to repopulate all the data structures from
> what was just read from the db, in case anything has changed. Hence I
> called rconn_connect() instead of reconnect()]
> rconn_set_dscp() does not get passed the 'target' and 'name'.
I still don't understand how the above adds up to rconn_reconnect()
being inappropriate following a DSCP change. It seems to me that it
does exactly what we want if the DSCP changes: it drops the connection
and reconnects to the existing target. Can you explain?
> As for the ovs-schema, the current documentation reads
> "The connection must be reset for the new DSCP values to take effect." ...
> for both the Manager and Controller dscp configuration. Though we are still
> resetting the connection, I am not sure if this needs to be explicitly
> stated in man page as the user does not need to do anything now?
Your documentation update seems OK to me (although you can delete the
sentence about 8 seconds, since I think we've fixed that). I was
talking about vswitch.ovsschema. You can drop that change, since as far
as I can see it isn't useful.
More information about the dev