On Thu, Sep 08, 2011 at 06:59:46PM -0700, Ethan Jackson wrote:
> This patch no longer rate limits database updates due to CFM
> changes. Due to recent changes, the fault status of CFM only
> changes once per 3.5 tx_interval seconds. There doesn't seem to be
> a good reason to add an additional rate limit on top of this.
This patch changes bridge_run() to commit a database transaction on
every trip through the main loop. That's going be troublesome because
of how transactions work. A transaction is always constructed against
the last version of the database contents received from ovsdb-server.
Calling ovsdb_idl_txn_commit() sends the transaction to ovsdb-server
and *does a rollback* of the changes from ovs-vswitchd's in-memory
copy of the database. Only when ovsdb-server commit succeeds or fails
and ovsdb-server sends the updated version of the database will the
in-memory copy of the database be permanently updated.
This means that, if it takes more than one trip through the main loop
for ovsdb-server to update the database and send back the updated
version, then, with this patch, ovs-vswitchd will send an extra copy
of the updates on every trip.
I think that the solution is to keep the transaction around until the
commit finishes, something like this:
if (!txn) {
txn = ovsdb_idl_txn_create(idl);
HMAP_FOR_EACH (br, node, &all_bridges) {
struct iface *iface;
HMAP_FOR_EACH (iface, name_node, &br->iface_by_name) {
iface_refresh_cfm_stats(iface);
}
}
}
if (ovsdb_idl_txn_commit(txn) != TXN_INCOMPLETE) {
ovsdb_idl_txn_destroy(txn);
txn = NULL;
}
'txn' needs to be static now so probably a longer name is better.
We'd also need an ovsdb_idl_txn_wait() call in bridge_wait().
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
