On Mon, Jul 29, 2024 at 8:16 AM Naveen Yerramneni <
[email protected]> wrote:

>
>
> > On 29 Jul 2024, at 11:22 AM, Ales Musil <[email protected]> wrote:
> >
> > CAUTION: External Email
> >
> >
> > On Fri, Jul 26, 2024 at 9:08 PM Naveen Yerramneni <
> [email protected]> wrote:
> > This change avoids northd full recompute for FDB updates.
> > Also, fixed incremental processing for port deletion to
> > delete all fdb entries associated to a port.
> >
> > Signed-off-by: Naveen Yerramneni <[email protected]>
> > ---
> > v2:
>
> >   - Addressed review comments from Ales.
>
> >   - Fixed incremental processing for port deletion to
>
> >     delete all fdb entries associated to a port.
>
> > ---
> >
> > Hi Naveen,
> >
> > thank you for the v2, I have a couple of comments down below.
> >   northd/en-northd.c       | 37 +++++++++++++++++++++++++++++++++++++
> >  northd/en-northd.h       |  1 +
> >  northd/inc-proc-northd.c |  2 +-
> >  northd/northd.c          | 11 +++++------
> >  northd/northd.h          |  3 +++
> >  5 files changed, 47 insertions(+), 7 deletions(-)
> >
> > diff --git a/northd/en-northd.c b/northd/en-northd.c
> > index 4479b4aff..badc49d43 100644
> > --- a/northd/en-northd.c
> > +++ b/northd/en-northd.c
> > @@ -259,3 +259,40 @@ en_northd_clear_tracked_data(void *data_)
> >      struct northd_data *data = data_;
> >      destroy_northd_data_tracked_changes(data);
> >  }
> > +
> > +bool
> > +sb_fdb_change_handler(struct engine_node *node, void *data)
> > +{
> > +    struct northd_data *nd = data;
> > +    const struct sbrec_fdb_table *sbrec_fdb_table =
> > +        EN_OVSDB_GET(engine_get_input("SB_fdb", node));
> > +
> > +    /* check if changed rows are stale and delete them */
> > +    const struct sbrec_fdb *fdb_e, *fdb_prev_del = NULL;
> >
> > I'm kinda confused as to why the fdb_prev_del is needed at all.
>
> I couldn’t find SAFE variant for SBREC_FDB_TABLE_FOR_EACH_TRACKED
> similar to SBREC_FDB_TABLE_FOR_EACH_SAFE hence I used fdb_prev_del
> to store and delete the previous stale entry during the iteration.
>


Ah right, nevertheless the entries to delete should be probably stored in
some structure (array maybe) to iterate over them at the end WDYT?


>
> >  +    SBREC_FDB_TABLE_FOR_EACH_TRACKED(fdb_e, sbrec_fdb_table) {
> > +        if (sbrec_fdb_is_deleted(fdb_e)) {
> > +            continue;
> > +        }
> > +
> > +        if (fdb_prev_del) {
> > +            sbrec_fdb_delete(fdb_prev_del);
> > +        }
> > +
> > +        fdb_prev_del = fdb_e;
> > +        struct ovn_datapath *od
> > +            = ovn_datapath_find_by_key(&nd->ls_datapaths.datapaths,
> > +                                          fdb_e->dp_key);
> > +        if (od) {
> > +            if (ovn_tnlid_present(&od->port_tnlids, fdb_e->port_key)) {
> > +                fdb_prev_del = NULL;
> > +            }
> > +        }
> >
> > Why not the following?
> > if (!(od && ovn_tnlid_present(...)) {
> >     sbrec_fdb_delete(fdb_e);
> > }
> >
> >
> >  +    }
> > +
> > +    if (fdb_prev_del) {
> > +        sbrec_fdb_delete(fdb_prev_del);
> > +    }
> > +
> > +    engine_set_node_state(node, EN_UPDATED);
> > +    return true;
> > +}
> > diff --git a/northd/en-northd.h b/northd/en-northd.h
> > index 9b7bda32a..9c722e401 100644
> > --- a/northd/en-northd.h
> > +++ b/northd/en-northd.h
> > @@ -19,5 +19,6 @@ bool northd_nb_logical_switch_handler(struct
> engine_node *, void *data);
> >  bool northd_nb_logical_router_handler(struct engine_node *, void *data);
> >  bool northd_sb_port_binding_handler(struct engine_node *, void *data);
> >  bool northd_lb_data_handler(struct engine_node *, void *data);
> > +bool sb_fdb_change_handler(struct engine_node *node, void *data);
> >
> >  #endif /* EN_NORTHD_H */
> > diff --git a/northd/inc-proc-northd.c b/northd/inc-proc-northd.c
> > index d56e9783a..213e6d88a 100644
> > --- a/northd/inc-proc-northd.c
> > +++ b/northd/inc-proc-northd.c
> > @@ -189,7 +189,7 @@ void inc_proc_northd_init(struct ovsdb_idl_loop *nb,
> >      engine_add_input(&en_northd, &en_sb_ha_chassis_group, NULL);
> >      engine_add_input(&en_northd, &en_sb_ip_multicast, NULL);
> >      engine_add_input(&en_northd, &en_sb_service_monitor, NULL);
> > -    engine_add_input(&en_northd, &en_sb_fdb, NULL);
> > +    engine_add_input(&en_northd, &en_sb_fdb, sb_fdb_change_handler);
> >      engine_add_input(&en_northd, &en_sb_static_mac_binding, NULL);
> >      engine_add_input(&en_northd, &en_sb_chassis_template_var, NULL);
> >      engine_add_input(&en_northd, &en_global_config,
> > diff --git a/northd/northd.c b/northd/northd.c
> > index 6898daa00..8ad3ad285 100644
> > --- a/northd/northd.c
> > +++ b/northd/northd.c
> > @@ -496,7 +496,7 @@ ovn_datapath_find(const struct hmap *datapaths,
> >      return ovn_datapath_find_(datapaths, uuid);
> >  }
> >
> > -static struct ovn_datapath *
> > +struct ovn_datapath *
> >  ovn_datapath_find_by_key(struct hmap *datapaths, uint32_t dp_key)
> >  {
> >      struct ovn_datapath *od;
> > @@ -3300,13 +3300,12 @@ delete_fdb_entry(struct ovsdb_idl_index
> *sbrec_fdb_by_dp_and_port,

>      sbrec_fdb_index_set_dp_key(target, dp_key);
> >      sbrec_fdb_index_set_port_key(target, port_key);
> >
> > -    struct sbrec_fdb *fdb_e =
> sbrec_fdb_index_find(sbrec_fdb_by_dp_and_port,
> > -                                                   target);
> > -    sbrec_fdb_index_destroy_row(target);
> > -
> > -    if (fdb_e) {
> > +    struct sbrec_fdb *fdb_e;
> > +    while ((fdb_e = sbrec_fdb_index_find(sbrec_fdb_by_dp_and_port,
> > +                                           target))) {
> >
> > Why is this change needed and wasn't before? What has changed?
> >
>
> In previous code, assume there are 2 entries associated to a port. When
> port got deleted,
> one of the 2 entries gets removed from FDB table, this triggers FDB update
> and since there
> was no change handler full recompute used to happen. As part of full
> recompute, the remaining
> stale entries used to get removed.
>
> Now, we have change handler in which we are iterating only tracked flows
> hence unmodified stale
> entries still remain in the FDB table without this change.
>
> > Also the proper way to iterate over index is to use the FOR_EACH_EQUAL
> macro:
> >
> > struct sbrec_fdb *fdb_e;
> > SBREC_FDB_FOR_EACH_EQUAL (fdb_e, target, sbrec_fdb_by_dp_and_port) {
> >     sbrec_fdb_delete(fdb_e);
> > }
>
> Ack, will address in v3.
>
>
Also it should be made explicit by the function name that it is removing
more than one
row now e.g. delete_fdb_entries().


> >           sbrec_fdb_delete(fdb_e);
> >      }
> > +    sbrec_fdb_index_destroy_row(target);
> >  }
> >
> >  struct service_monitor_info {
> > diff --git a/northd/northd.h b/northd/northd.h
> > index d4a8d75ab..26fe24df3 100644
> > --- a/northd/northd.h
> > +++ b/northd/northd.h
> > @@ -87,6 +87,9 @@ ods_size(const struct ovn_datapaths *datapaths)
> >      return hmap_count(&datapaths->datapaths);
> >  }
> >
> > +struct ovn_datapath *
> > +ovn_datapath_find_by_key(struct hmap *datapaths, uint32_t dp_key);
> > +
> >  bool od_has_lb_vip(const struct ovn_datapath *od);
> >
> >  struct tracked_ovn_ports {
> > --
> > 2.36.6
> >
> > _______________________________________________
> > dev mailing list
> > [email protected]
> > https://mail.openvswitch.org/mailman/listinfo/ovs-dev [
> mail.openvswitch.org]
> >
> >
> > Thanks,
> > Ales
> >
> > --
> > Ales Musil
> > Senior Software Engineer - OVN Core
> > Red Hat EMEA [redhat.com][email protected] [red.ht]
>
>
>

-- 

Ales Musil

Senior Software Engineer - OVN Core

Red Hat EMEA <https://www.redhat.com>

[email protected]
<https://red.ht/sig>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to