On Thu, Apr 28, 2016 at 08:23:59PM -0400, Ramu Ramamurthy wrote:
> On graceful-restart of ovn-controller, the chassis row is
> inserted in the Chassis table. During this transaction,
> there is a window of time where an idl row-read may not
> return the newly created row - even though the row should exist,
> but the transaction is in an incomplete state.
> As a result, get_chassis() in binding_run() returns a null chassis record
> binding_run exits early, and does not
> create local_datapaths, and patch_run deletes
> localnet patch ports. In a later run, the localnet patch ports are recreated.
>
> This is reproducable consistently but not on every restart.
> The fix is to handle the case that the chassis record
> may be null in binding_run, and yet create local_datapaths.
>
> RFC because, I am not sure if the fix is correct - ie,
> can there be local_datapaths with no chassis record ? And if this
> fix is not correct, then what is the correct fix ? Would it be
> to not run binding_run() and patch_run() when ovnsb_idl_txn is null ?
I think that this is correct, or at any rate reasonable.
I applied this to master, folding in the following logic simplification:
diff --git a/ovn/controller/binding.c b/ovn/controller/binding.c
index 9f50cd5..a0d8b96 100644
--- a/ovn/controller/binding.c
+++ b/ovn/controller/binding.c
@@ -187,10 +187,8 @@ binding_run(struct controller_ctx *ctx, const struct
ovsrec_bridge *br_int,
if (iface_rec && ctx->ovs_idl_txn) {
update_qos(iface_rec, binding_rec);
}
- if (chassis_rec && binding_rec->chassis == chassis_rec) {
- continue;
- }
- if (ctx->ovnsb_idl_txn && chassis_rec) {
+ if (ctx->ovnsb_idl_txn && chassis_rec
+ && binding_rec->chassis != chassis_rec) {
if (binding_rec->chassis) {
VLOG_INFO("Changing chassis for lport %s from %s to %s.",
binding_rec->logical_port,
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
