> On 7/31/25 10:52 PM, Lorenzo Bianconi wrote:
> > This patch handles the logical switch creation incrementally
> > in the northd engine node. The dependent engine nodes - ls_stateful,
> > lflow and few others still fall back to full recompute, which
> > will be handled in separate patches.
> >
> > Reported-at: https://issues.redhat.com/browse/FDP-754
> > Co-authored-by: Numan Siddique <num...@ovn.org>
> > Signed-off-by: Numan Siddique <num...@ovn.org>
> > Signed-off-by: Lorenzo Bianconi <lorenzo.bianc...@redhat.com>
> > ---
>
> Hi Lorenzo,
>
> Here's one more issue I found while testing this on top of current main:
Hi Dumitru,
thx for the review.
>
> >
> > +static void
> > +destroy_tracked_dps(struct tracked_dps *trk_dps)
> > +{
> > + hmapx_clear(&trk_dps->crupdated);
> > +
> > + struct hmapx_node *n;
> > + HMAPX_FOR_EACH (n, &trk_dps->deleted) {
> > + free(n->data);
>
> We leak memory here. n->data is of type 'struct ovn_datapath *'. We
> need to call ovn_datapath_destroy() instead. However, that's trickier
> because we already removed the element from 'ls_datapaths.datapaths' so
> we need to something like:
>
> Change the ovn_datapath_destroy() function to not do hmap removal
> (expect the record to not be in any map). And change
> ovn_datapaths_destroy() to:
>
> static void
> ovn_datapaths_destroy(struct ovn_datapaths *datapaths)
> {
> struct ovn_datapath *dp;
> HMAP_FOR_EACH_POP (dp, key_node, &datapaths->datapaths) {
> ovn_datapath_destroy(dp);
> }
> hmap_destroy(&datapaths->datapaths);
>
> bitmap_free(datapaths->dps_index_map.map);
> vector_destroy(&datapaths->dps);
> }
ack, I will fix it in v5.
Regards,
Lorenzo
>
> > + }
> > + hmapx_clear(&trk_dps->deleted);
> > +}
> > +
>
> Regards,
> Dumitru
>
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
