On 11/16/26 9:51 AM, Xavier Simonart wrote:
> Hi Rosemarie
> 
> Thanks for reviewing this patch.
> 
> On Sat, Nov 14, 2026 at 1:53 AM Rosemarie O'Riorden
> <[email protected] <mailto:[email protected]>> wrote:
> 
>     Hi Xavier, thanks for the fix. I just have a question below.
> 
>     On 11/6/26 4:08 AM, Xavier Simonart via dev wrote:
>     > Since [0], en_route_exchange has en_sb_cond_seqno as an input
>     > without change handler.
>     > en_sb_cond_seqno changes are common with conditional monitoring
>     > (e.g., a port being claimed), and with no handler such change forces
>     > a node recompute.
>     > en_route_exchange is SB_WRITE, so if this recompute occurs while
>     > OpenFlow has backlog, it gets canceled, which forces a full
>     > recompute later, including a lflow_run recompute.
>     > This was caught by sporadic failures of test [1], which was
>     > validating that no lflow_run occurs when claiming a port.
>     >
>     > Replace route_exchange_sb_ro_handler with route_exchange_sb_handler,
>     > shared by both the 'sb_ro' and 'sb_cond_seqno' inputs.
>     > It performs the sync when possible, returning
>     > EN_HANDLED_UNCHANGED/EN_HANDLED_UPDATED instead of EN_UNHANDLED.
>     >
>     > It never forces a recompute:
>     > - when the gate did not move, there is nothing to do.
>     > - when it did move but SB is not writable, it is safe to wait
>     >   quietly, since 'sb_ro' will fire again when SB becomes writable,
>     >   and we could not have synced this iteration anyway.
>     > This is unrelated to oflow backlog, as it does not program openflows.
>     >
>     > [0] 'controller: Defer route updates until routes load.'
>     > [1] "ovn-controller incremental processing".
>     >
>     > Reported-at: https://redhat.atlassian.net/browse/FDP-4699
>     <https://redhat.atlassian.net/browse/FDP-4699>
>     > Assisted-by: Claude Sonnet 5, Claude Code
>     > Fixes: a3e697fb9798 ("controller: Defer route updates until routes
>     load.")
>     > Signed-off-by: Xavier Simonart <[email protected]
>     <mailto:[email protected]>>
>     > ---
>     >  controller/ovn-controller.c      | 27 +++++++++++++++++++++------
>     >  tests/ovn-inc-proc-graph-dump.at <http://ovn-inc-proc-graph-
>     dump.at> |  4 ++--
>     >  tests/ovn-performance.at <http://ovn-performance.at>         | 21
>     +++++++++++++++++++--
>     >  3 files changed, 42 insertions(+), 10 deletions(-)
>     >
>     > diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c
>     > index 01b0119b7..c39d9b8d0 100644
>     > --- a/controller/ovn-controller.c
>     > +++ b/controller/ovn-controller.c
>     > @@ -6168,15 +6168,29 @@ route_exchange_route_table_handler(struct
>     engine_node *node, void *data)
>     >      return EN_HANDLED_UPDATED;
>     >  }
>     > 
>     > +/* Change handler shared by the 'sb_ro' and 'sb_cond_seqno'
>     inputs. */
>     >  static enum engine_input_handler_result
>     > -route_exchange_sb_ro_handler(struct engine_node *node OVS_UNUSED,
>     void *data)
>     > +route_exchange_sb_handler(struct engine_node *node, void *data)
>     >  {
>     >      struct ed_type_route_exchange *re = data;
>     > -    if (re->sb_changes_pending) {
>     > -        return EN_UNHANDLED;
>     > +
>     > +    if (!sbrec_server_has_learned_route_table(re->sb_idl)) {
>     > +        return EN_HANDLED_UNCHANGED;
>     >      }
>     > 
>     > -    return EN_HANDLED_UNCHANGED;
>     > +    bool ready = route_exchange_ready(node, re);
>     > +    if (re->initial_routes_loaded == ready && !re-
>     >sb_changes_pending) {
>     > +        /* The readiness gate did not flip and there is no SB
>     write left
>     > +         * over from a previous attempt: this event is not
>     relevant to us. */
>     > +        return EN_HANDLED_UNCHANGED;
>     > +    }
>     > +
>     > +    if (!engine_get_context()->ovnsb_idl_txn) {
>     > +        return EN_HANDLED_UNCHANGED;
>     > +    }
>     > +
>     > +    return route_exchange_sync(node, re) ? EN_HANDLED_UPDATED
>     > +                                         : EN_HANDLED_UNCHANGED;
>     >  }
>     > 
>     > 
>     > @@ -7725,11 +7739,12 @@ inc_proc_ovn_controller_init(
>     >                       route_exchange_route_table_handler);
>     >      engine_add_input(&en_route_exchange,
>     &en_route_exchange_status, NULL);
>     >      engine_add_input(&en_route_exchange, &en_sb_ro,
>     > -                     route_exchange_sb_ro_handler);
>     > +                     route_exchange_sb_handler);
>     >      /* Re-evaluate the initial-dump readiness gate whenever the
>     SB monitor
>     >       * conditions change (including the final ack of the scoped
>     >       * Advertised_Route dump). */
>     > -    engine_add_input(&en_route_exchange, &en_sb_cond_seqno, NULL);
>     > +    engine_add_input(&en_route_exchange, &en_sb_cond_seqno,
>     > +                     route_exchange_sb_handler);
>     > 
>     >      engine_add_input(&en_addr_sets, &en_sb_address_set,
>     >                       addr_sets_sb_address_set_handler);
>     > diff --git a/tests/ovn-inc-proc-graph-dump.at <http://ovn-inc-
>     proc-graph-dump.at> b/tests/ovn-inc-proc-graph-dump.at <http://ovn-
>     inc-proc-graph-dump.at>
>     > index b3df37a01..7853e5001 100644
>     > --- a/tests/ovn-inc-proc-graph-dump.at <http://ovn-inc-proc-graph-
>     dump.at>
>     > +++ b/tests/ovn-inc-proc-graph-dump.at <http://ovn-inc-proc-graph-
>     dump.at>
>     > @@ -480,8 +480,8 @@ digraph "Incremental-Processing-Engine" {
>     >       SB_port_binding -> route_exchange
>     [[label="engine_noop_handler"]];
>     >       route_table_notify -> route_exchange
>     [[label="route_exchange_route_table_handler"]];
>     >       route_exchange_status -> route_exchange [[label=""]];
>     > -     sb_ro -> route_exchange
>     [[label="route_exchange_sb_ro_handler"]];
>     > -     sb_cond_seqno -> route_exchange [[label=""]];
>     > +     sb_ro -> route_exchange [[label="route_exchange_sb_handler"]];
>     > +     sb_cond_seqno -> route_exchange
>     [[label="route_exchange_sb_handler"]];
>     >       garp_rarp [[style=filled, shape=box, fillcolor=white,
>     label="garp_rarp"]];
>     >       OVS_open_vswitch -> garp_rarp [[label=""]];
>     >       SB_chassis -> garp_rarp [[label=""]];
>     > diff --git a/tests/ovn-performance.at <http://ovn-performance.at>
>     b/tests/ovn-performance.at <http://ovn-performance.at>
>     > index 114917832..de56ca5f8 100644
>     > --- a/tests/ovn-performance.at <http://ovn-performance.at>
>     > +++ b/tests/ovn-performance.at <http://ovn-performance.at>
>     > @@ -359,11 +359,28 @@ for i in 1 2; do
>     >          [hv1 hv2], [lflow_run],
>     >          [ovn-nbctl --wait=hv acl-add $ls to-lport 1000 'outport
>     == \"$lp\"' drop]
>     >      )
>     > -
> 
>     Nit: random empty line deleted
> 
>     >      # Bind port $lp and wait for it to come up
>     > +    #
>     > +    # Create openflow backlog by pausing ovs/ovsdb.
>     > +    # We can't simply pause ovs to delay flow handling, as this
>     would also prevent ofport creation.
>     > +    # So, follow this pattern:
>     > +    # 1) Pause ovs to prevent oflow creation, but keep ovsdb
>     running so we can add interface.
>     > +    # 2) Create interface.
>     > +    # 3) Pause ovs-db and restart ovs. ovs now creates ofport,
>     but ovsdb is sleeping.
>     > +    # 4) After one second (i.e. enough for ovs to have time to
>     write ofport), pause ovs and restart ovsdb.
>     > +    # 5) Now ovn-controller should get ofport, claim and program
>     flows (while ovs is paused).
>     > +    #    Sleep 2 more seconds to let ovn-controller handle the
>     diff, start programming openflows, hitting backlog and cancelling.
>     > +    #    If previous pauses are not long enough (which should be
>     rare), we might miss some failures on buggy code, but this should
>     not cause flaky tests.
>     > +    # 6) Wake up ovs so flows get programmed.
>     > +    # 7) Wait for port up + sync.
>     > +    # 8) This used to cause recomputes.
>     > +    sleep_ovs hv$i
>     >      OVN_CONTROLLER_EXPECT_NO_HIT(
>     >          [hv$i hv$j], [lflow_run],
>     > -        [as hv$i ovs-vsctl add-port br-int $vif -- set Interface
>     $vif external-ids:iface-id=$lp &&
>     > +        [as hv$i ovs-vsctl --no-wait add-port br-int $vif -- set
>     Interface $vif external-ids:iface-id=$lp &&
>     > +         sleep_ovsdb hv$i && wake_up_ovs hv$i &&
>     > +         sleep 1 && sleep_ovs hv$i && wake_up_ovsdb hv$i &&
>     > +         sleep 2 && wake_up_ovs hv$i &&
> 
>     Looking at this test made me recall somewhat recently seeing on the
>     mailing list that OVS is trying to avoid all sleeps in tests going
>     forwards. So I was just wondering, could this be done without sleeps?
>     I'm not sure exactly how it would look... but it's just a thought. Did
>     you consider this? Do you think it's even feasible?
> 
> Thanks for your suggestion.
> I initially skipped this because implementing it cleanly wasn't obvious.
> So, I took another look this morning, as I agree avoiding sleeps is
> usually a good idea.
> It is technically feasible by enabling the relevant debug logs and
> waiting for specific log messages instead of using fixed sleeps:
> Replace 'sleep 1' with 'waiting for proper ofport' in hv$i/ovs-vswitchd.log
> Replace 'sleep 2' with 'waiting for 'ofctrl_put can.t be performed'' in
> hv$i/ovn-controller.log
> However, after fighting a bit due to the initial sleeps being embedded
> in some m4 macros, the cleanest shell equivalent I could come up with
> ends up looking like this for resp. 'sleep 1' and 'sleep 2':
> seconds=0 && while ! grep -q 'ofport.:1001' hv$i/ovs-vswitchd.log; do
> [[ \$seconds -ge 30 ]] && break; sleep 1; seconds=\$((seconds + 1)); done &&
> seconds=0 && while ! grep -q 'ofctrl_put can.t be performed' hv$i/ovn-
> controller.log; do [[ "\$seconds" -ge 30 ]] && break; sleep 1; seconds=\
> $((seconds + 1)); done &&
> Given how much extra complexity this adds to an already dense macro, and
> since the worst-case scenario with the current sleep 1/sleep 2 on a slow
> system is simply missing a regression check rather than causing a false-
> positive test failure, I think keeping the simple sleep calls is the
> better trade-off here. Please let me know if you feel otherwise.

Hi Xavier, thanks for looking into this! That was my concern too, that
the replacement for a simple sleep would be some monstrosity like that.
Because since we're already inside of the OVN_CONTROLLER_EXPECT_NO_HIT
macro, in the middle of a sequence of bash commands, we can't simply use
OVS_WAIT_UNTIL... Well, thanks anyways for taking a look! I'm fine with
the sleeps then.

Acked-by: Rosemarie O'Riorden <[email protected]>

> 
> 
>     Thanks,
>     Rosemarie
> 
> Thanks
> Xavier 
> 
> 
>     >           check ovn-nbctl wait-until Logical_Switch_Port $lp
>     'up=true' &&
>     >           check ovn-nbctl --wait=hv sync]
>     >      )
> 

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to