On 7/2/26 3:46 PM, Eli Britstein via dev wrote: > When ovs-vswitchd exits without --cleanup, kernel datapath state can > outlive the daemon. If the next configuration uses no bridge of that > datapath type, nothing reopens the backer and the orphan persists. > > After each bridge_reconfigure(), delete enumerated datapaths for types > that no configured bridge uses and that ovs-vswitchd is not already > managing. Kernel datapaths are preserved across restarts that use > flow-restore-wait when the database still lists system bridges. > > Signed-off-by: Eli Britstein <[email protected]> > ---
Hi, Eli. Thanks for the patch. Though I don't think that the right direction to go. It is technically possible with the careful management to run multiple OVS instances within the same network namespace (with different datapath types). And it is also possible to run multiple OVS datapaths alongside each other in the kernel without an ovs-vswitchd (some service meshes do that). So, the ovs-vswitchd should not make a unilateral decision to destroy detapaths that do not belong to it, as that will break other instances or other users of the kernel datapath. Best regards, Ilya Maximets. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
