On Tue, Sep 1, 2026 at 11:04 PM Jacob Tanenbaum via dev <
[email protected]> wrote:

> Split ovn-northd's en_lflow engine node into a compute-only
> en_lflow node and a new en_dp_group_resolved node that
> handles datapath-group resolution and all SB database
> writes.  This separation makes en_lflow independent of the
> SB database, preparing for future incremental processing
> improvements.
>
> en_lflow's handlers now track dirty lflow_refs in a hmapx
> instead of calling lflow_ref_sync_lflows directly.
> en_dp_group_resolved drains the dirty set on its lflow
> handler, falling back to a full lflow_table_sync_to_sb when
> en_lflow did a full recompute or when the IGMP/MLD handler
> set the needs_full_sync flag.
>
> A new lflow_ref_unlink_and_prune() function replaces
> lflow_ref_resync_flows() for the IGMP and IC-learned
> service monitor handlers.  These handlers use a single
> shared lflow_ref whose flows are also built by the
> per-datapath pipeline (with lflow_ref = NULL).  The
> per-datapath build's dp bits are not tracked by any
> lflow_ref, so lflow_ref_unlink_lflows alone cannot clear
> them via dp_refcnt.  lflow_ref_unlink_and_prune destroys
> all lrns and orphaned lflows in-memory without SB writes,
> equivalent to lflow_ref_resync_flows' cleanup.
>
> The lflow_table_add_lflow__ upgrade-to-bitmap logic (for
> single lflow_refs that contribute multiple datapaths to the
> same lflow) is extended with per-datapath dp_refcnt
> accounting during mid-cycle bitmap upgrades, preventing
> premature dp bit release when multiple lflow_refs share a
> flow across overlapping datapaths.
>
> Reported-at: https://redhat.atlassian.net/browse/FDP-2747
> Assisted-by: Claude Opus 4.8, Claude Code
> Signed-off-by: Jacob Tanenbaum <[email protected]>
> ---
>

Hi Jacob,

thank you for the patch. I have a few comments down below.
Also did you run ovn-heater tests to see if there are any potential
issues at scale?


>  northd/automake.mk               |   2 +
>  northd/en-dp-group-resolved.c    | 129 ++++++++++++++++
>  northd/en-dp-group-resolved.h    |  30 ++++
>  northd/en-lflow.c                | 140 +++++------------
>  northd/en-lflow.h                |   3 +
>  northd/en-northd-output.c        |   4 +-
>  northd/en-northd-output.h        |   3 +-
>  northd/inc-proc-northd.c         |  29 +++-
>  northd/lflow-mgr.c               | 168 +++++++++++++++++----
>  northd/lflow-mgr.h               |   8 +-
>  northd/northd.c                  | 251 +++++++------------------------
>  northd/northd.h                  |  39 +++--
>  tests/ovn-inc-proc-graph-dump.at |  12 +-
>  tests/ovn-northd.at              |  81 +++++++---
>  14 files changed, 506 insertions(+), 393 deletions(-)
>  create mode 100644 northd/en-dp-group-resolved.c
>  create mode 100644 northd/en-dp-group-resolved.h
>
> diff --git a/northd/automake.mk b/northd/automake.mk
> index 45ca0337f..d1439294d 100644
> --- a/northd/automake.mk
> +++ b/northd/automake.mk
> @@ -16,6 +16,8 @@ northd_ovn_northd_SOURCES = \
>         northd/en-datapath-logical-router.h \
>         northd/en-datapath-sync.c \
>         northd/en-datapath-sync.h \
> +       northd/en-dp-group-resolved.c \
> +       northd/en-dp-group-resolved.h \
>         northd/en-ecmp-nexthop.c \
>         northd/en-ecmp-nexthop.h \
>         northd/en-global-config.c \
> diff --git a/northd/en-dp-group-resolved.c b/northd/en-dp-group-resolved.c
> new file mode 100644
> index 000000000..d5f9e7738
> --- /dev/null
> +++ b/northd/en-dp-group-resolved.c
>

nit: I would probably leave the resolved out of the name.
Just en-dp-group is better IMO.


> @@ -0,0 +1,129 @@
> +/*
> + * Copyright (c) 2026, Red Hat, Inc.
> + *
> + * Licensed under the Apache License, Version 2.0 (the "License");
> + * you may not use this file except in compliance with the License.
> + * You may obtain a copy of the License at:
> + *
> + *     http://www.apache.org/licenses/LICENSE-2.0
> + *
> + * Unless required by applicable law or agreed to in writing, software
> + * distributed under the License is distributed on an "AS IS" BASIS,
> + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
> implied.
> + * See the License for the specific language governing permissions and
> + * limitations under the License.
> + */
> +
> +#include <config.h>
> +
> +#include "en-dp-group-resolved.h"
> +#include "en-datapath-sync.h"
> +#include "en-global-config.h"
> +#include "en-lflow.h"
> +#include "lflow-mgr.h"
> +
> +#include "lib/inc-proc-eng.h"
> +#include "northd.h"
> +#include "lib/stopwatch-names.h"
> +#include "stopwatch.h"
> +#include "timeval.h"
> +#include "openvswitch/vlog.h"
>

nit: Not ordered according to coding style.


> +
> +VLOG_DEFINE_THIS_MODULE(en_dp_group_resolved);
> +
> +void *
> +en_dp_group_resolved_init(struct engine_node *node OVS_UNUSED,
> +                          struct engine_arg *arg OVS_UNUSED)
> +{
> +    return NULL;
> +}
> +
> +static void
> +dp_group_resolved_sync_to_sb(struct engine_node *node,
> +                             struct lflow_data *lflow_data)
> +{
> +    const struct engine_context *eng_ctx = engine_get_context();
> +    struct all_synced_datapaths *all_dps =
> +        engine_get_input_data("datapath_sync", node);
> +
> +    const struct sbrec_logical_flow_table *sb_flow_table =
> +        EN_OVSDB_GET(engine_get_input("SB_logical_flow", node));
> +    const struct sbrec_logical_dp_group_table *sb_dpgrp_table =
> +        EN_OVSDB_GET(engine_get_input("SB_logical_dp_group", node));
> +
> +    struct ed_type_global_config *global_config =
> +        engine_get_input_data("global_config", node);
> +
> +    stopwatch_start(LFLOWS_TO_SB_STOPWATCH_NAME, time_msec());
> +    lflow_table_sync_to_sb(lflow_data->lflow_table,
> +                           eng_ctx->ovnsb_idl_txn,
> +                           all_dps->synced_dps,
> +                           global_config->ovn_internal_version_changed,
> +                           sb_flow_table, sb_dpgrp_table);
> +    stopwatch_stop(LFLOWS_TO_SB_STOPWATCH_NAME, time_msec());
> +}
> +
> +enum engine_node_state
> +en_dp_group_resolved_run(struct engine_node *node,
> +                         void *data OVS_UNUSED)
> +{
> +    struct lflow_data *lflow_data = engine_get_input_data("lflow", node);
> +
> +    /* The full sync below covers every lflow in the table, so any
> +     * per-ref dirty tracking left over from en_lflow's handlers is
> +     * redundant.  Clear it so the next incremental cycle starts
> +     * clean. */
> +    hmapx_clear(&lflow_data->dirty_lflow_refs);
> +    lflow_data->needs_full_sync = false;
> +
> +    dp_group_resolved_sync_to_sb(node, lflow_data);
> +    return EN_UPDATED;
> +}
> +
> +enum engine_input_handler_result
> +dp_group_resolved_lflow_handler(struct engine_node *node,
> +                                void *data OVS_UNUSED)
> +{
> +    struct lflow_data *lflow_data = engine_get_input_data("lflow", node);
> +
> +    if (hmapx_is_empty(&lflow_data->dirty_lflow_refs)
> +        || lflow_data->needs_full_sync) {
> +        hmapx_clear(&lflow_data->dirty_lflow_refs);
> +        lflow_data->needs_full_sync = false;
> +        dp_group_resolved_sync_to_sb(node, lflow_data);
> +        return EN_HANDLED_UPDATED;
> +    }
> +
> +    const struct engine_context *eng_ctx = engine_get_context();
> +    struct all_synced_datapaths *all_dps =
> +        engine_get_input_data("datapath_sync", node);
> +
> +    const struct sbrec_logical_flow_table *sb_flow_table =
> +        EN_OVSDB_GET(engine_get_input("SB_logical_flow", node));
> +    const struct sbrec_logical_dp_group_table *sb_dpgrp_table =
> +        EN_OVSDB_GET(engine_get_input("SB_logical_dp_group", node));
> +
> +    struct ed_type_global_config *global_config =
> +        engine_get_input_data("global_config", node);
>

nit: We usually get all the input data before doing any kind of operation.


> +
> +    struct hmapx_node *hmapx_node;
> +    HMAPX_FOR_EACH (hmapx_node, &lflow_data->dirty_lflow_refs) {
> +        struct lflow_ref *ref = hmapx_node->data;
> +        if (!lflow_ref_sync_lflows(
> +                ref, lflow_data->lflow_table,
> +                eng_ctx->ovnsb_idl_txn,
> +                all_dps->synced_dps,
> +                global_config->ovn_internal_version_changed,
> +                sb_flow_table, sb_dpgrp_table)) {
> +            return EN_UNHANDLED;
> +        }
> +    }
> +    hmapx_clear(&lflow_data->dirty_lflow_refs);
> +
> +    return EN_HANDLED_UPDATED;
> +}
> +
> +void
> +en_dp_group_resolved_cleanup(void *data OVS_UNUSED)
> +{
> +}
> diff --git a/northd/en-dp-group-resolved.h b/northd/en-dp-group-resolved.h
> new file mode 100644
> index 000000000..9f25072c1
> --- /dev/null
> +++ b/northd/en-dp-group-resolved.h
> @@ -0,0 +1,30 @@
> +/*
> + * Copyright (c) 2026, Red Hat, Inc.
> + *
> + * Licensed under the Apache License, Version 2.0 (the "License");
> + * you may not use this file except in compliance with the License.
> + * You may obtain a copy of the License at:
> + *
> + *     http://www.apache.org/licenses/LICENSE-2.0
> + *
> + * Unless required by applicable law or agreed to in writing, software
> + * distributed under the License is distributed on an "AS IS" BASIS,
> + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
> implied.
> + * See the License for the specific language governing permissions and
> + * limitations under the License.
> + */
> +
> +#ifndef EN_DP_GROUP_RESOLVED_H
> +#define EN_DP_GROUP_RESOLVED_H 1
> +
> +#include "lib/inc-proc-eng.h"
> +
> +void *en_dp_group_resolved_init(struct engine_node *node,
> +                                struct engine_arg *arg);
> +enum engine_node_state en_dp_group_resolved_run(struct engine_node *node,
> +                                                void *data);
> +void en_dp_group_resolved_cleanup(void *data);
> +enum engine_input_handler_result
> +dp_group_resolved_lflow_handler(struct engine_node *node, void *data);
> +
> +#endif /* EN_DP_GROUP_RESOLVED_H */
> diff --git a/northd/en-lflow.c b/northd/en-lflow.c
> index 8cb987777..38b1639d6 100644
> --- a/northd/en-lflow.c
> +++ b/northd/en-lflow.c
> @@ -68,10 +68,6 @@ lflow_get_input_data(struct engine_node *node,
>      struct all_synced_datapaths *all_dps =
>          engine_get_input_data("datapath_sync", node);
>
> -    lflow_input->sbrec_logical_flow_table =
> -        EN_OVSDB_GET(engine_get_input("SB_logical_flow", node));
> -    lflow_input->sbrec_logical_dp_group_table =
> -        EN_OVSDB_GET(engine_get_input("SB_logical_dp_group", node));
>      lflow_input->sbrec_acl_id_table =
>          EN_OVSDB_GET(engine_get_input("SB_acl_id", node));
>
> @@ -119,8 +115,6 @@ lflow_get_input_data(struct engine_node *node,
>  enum engine_node_state
>  en_lflow_run(struct engine_node *node, void *data)
>  {
> -    const struct engine_context *eng_ctx = engine_get_context();
> -
>      struct lflow_input lflow_input;
>      lflow_get_input_data(node, &lflow_input);
>
> @@ -129,9 +123,10 @@ en_lflow_run(struct engine_node *node, void *data)
>          search_mode == LFLOW_TABLE_SEARCH_FIELDS);
>      lflow_reset_northd_refs(&lflow_input);
>      lflow_ref_clear(lflow_input.igmp_lflow_ref);
> +    hmapx_clear(&lflow_data->dirty_lflow_refs);
> +    lflow_data->needs_full_sync = false;
>
> -    build_lflows(eng_ctx->ovnsb_idl_txn, &lflow_input,
> -                 lflow_data->lflow_table);
> +    build_lflows(&lflow_input, lflow_data->lflow_table);
>
>      return EN_UPDATED;
>  }
> @@ -149,31 +144,25 @@ lflow_northd_handler(struct engine_node *node,
>          return EN_UNHANDLED;
>      }
>
> -    const struct engine_context *eng_ctx = engine_get_context();
>      struct lflow_data *lflow_data = data;
>
>      struct lflow_input lflow_input;
>      lflow_get_input_data(node, &lflow_input);
>
> -    if (!lflow_handle_northd_lr_changes(eng_ctx->ovnsb_idl_txn,
> -
> &northd_data->trk_data.trk_routers,
> -                                        &lflow_input,
> -                                        lflow_data->lflow_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_handle_northd_lr_changes(&northd_data->trk_data.trk_routers,
> +                                   &lflow_input,
> +                                   lflow_data->lflow_table,
> +                                   &lflow_data->dirty_lflow_refs);
>
> -    if (!lflow_handle_northd_port_changes(eng_ctx->ovnsb_idl_txn,
> -                                          &northd_data->trk_data.trk_lsps,
> -                                          &lflow_input,
> -                                          lflow_data->lflow_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_handle_northd_port_changes(&northd_data->trk_data.trk_lsps,
> +                                     &lflow_input,
> +                                     lflow_data->lflow_table,
> +                                     &lflow_data->dirty_lflow_refs);
>
> -    if (!lflow_handle_northd_lb_changes(
> -            eng_ctx->ovnsb_idl_txn, &northd_data->trk_data.trk_lbs,
> -            &lflow_input, lflow_data->lflow_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_handle_northd_lb_changes(&northd_data->trk_data.trk_lbs,
> +                                   &lflow_input,
> +                                   lflow_data->lflow_table,
> +                                   &lflow_data->dirty_lflow_refs);
>
>      return EN_HANDLED_UPDATED;
>  }
> @@ -189,17 +178,14 @@ lflow_lr_stateful_handler(struct engine_node *node,
> void *data)
>          return EN_UNHANDLED;
>      }
>
> -    const struct engine_context *eng_ctx = engine_get_context();
>      struct lflow_data *lflow_data = data;
>      struct lflow_input lflow_input;
>
>      lflow_get_input_data(node, &lflow_input);
> -    if (!lflow_handle_lr_stateful_changes(eng_ctx->ovnsb_idl_txn,
> -                                          &lr_sful_data->trk_data,
> -                                          &lflow_input,
> -                                          lflow_data->lflow_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_handle_lr_stateful_changes(&lr_sful_data->trk_data,
> +                                     &lflow_input,
> +                                     lflow_data->lflow_table,
> +                                     &lflow_data->dirty_lflow_refs);
>
>      return EN_HANDLED_UPDATED;
>  }
> @@ -214,17 +200,14 @@ lflow_ls_stateful_handler(struct engine_node *node,
> void *data)
>          return EN_UNHANDLED;
>      }
>
> -    const struct engine_context *eng_ctx = engine_get_context();
>      struct lflow_data *lflow_data = data;
>      struct lflow_input lflow_input;
>
>      lflow_get_input_data(node, &lflow_input);
> -    if (!lflow_handle_ls_stateful_changes(eng_ctx->ovnsb_idl_txn,
> -                                          &ls_sful_data->trk_data,
> -                                          &lflow_input,
> -                                          lflow_data->lflow_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_handle_ls_stateful_changes(&ls_sful_data->trk_data,
> +                                     &lflow_input,
> +                                     lflow_data->lflow_table,
> +                                     &lflow_data->dirty_lflow_refs);
>
>      return EN_HANDLED_UPDATED;
>  }
> @@ -235,35 +218,20 @@ lflow_multicast_igmp_handler(struct engine_node
> *node, void *data)
>      struct multicast_igmp_data *mcast_igmp_data =
>          engine_get_input_data("multicast_igmp", node);
>
> -    const struct engine_context *eng_ctx = engine_get_context();
>      struct lflow_data *lflow_data = data;
>      struct lflow_input lflow_input;
>      lflow_get_input_data(node, &lflow_input);
>
> -    if (!lflow_ref_resync_flows(mcast_igmp_data->lflow_ref,
> -                                lflow_data->lflow_table,
> -                                eng_ctx->ovnsb_idl_txn,
> -                                lflow_input.dps,
> -                                lflow_input.ovn_internal_version_changed,
> -                                lflow_input.sbrec_logical_flow_table,
> -
> lflow_input.sbrec_logical_dp_group_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_ref_unlink_and_prune(mcast_igmp_data->lflow_ref,
> +                              lflow_data->lflow_table);
>
>      build_igmp_lflows(&mcast_igmp_data->igmp_groups,
>                        &lflow_input.ls_datapaths->datapaths,
>                        lflow_data->lflow_table,
>                        mcast_igmp_data->lflow_ref);
>
> -    if (!lflow_ref_sync_lflows(mcast_igmp_data->lflow_ref,
> -                               lflow_data->lflow_table,
> -                               eng_ctx->ovnsb_idl_txn,
> -                               lflow_input.dps,
> -                               lflow_input.ovn_internal_version_changed,
> -                               lflow_input.sbrec_logical_flow_table,
> -                               lflow_input.sbrec_logical_dp_group_table))
> {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_data->needs_full_sync = true;
> +    hmapx_add(&lflow_data->dirty_lflow_refs, mcast_igmp_data->lflow_ref);
>
>      return EN_HANDLED_UPDATED;
>  }
> @@ -280,7 +248,6 @@ lflow_group_ecmp_route_change_handler(struct
> engine_node *node,
>          return EN_UNHANDLED;
>      }
>
> -    const struct engine_context *eng_ctx = engine_get_context();
>      struct lflow_data *lflow_data = data;
>
>      struct lflow_input lflow_input;
> @@ -289,25 +256,13 @@ lflow_group_ecmp_route_change_handler(struct
> engine_node *node,
>      struct group_ecmp_datapath *route_node;
>      struct hmapx_node *hmapx_node;
>
> -    /* We need to handle deletions before additions as they could
> potentially
> -     * overlap. */
>      HMAPX_FOR_EACH (hmapx_node,
>
>  &group_ecmp_route_data->trk_data.deleted_datapath_routes) {
>          route_node = hmapx_node->data;
>          lflow_ref_unlink_lflows(route_node->lflow_ref);
> -
> -        bool handled = lflow_ref_sync_lflows(
> -            route_node->lflow_ref, lflow_data->lflow_table,
> -            eng_ctx->ovnsb_idl_txn, lflow_input.dps,
> -            lflow_input.ovn_internal_version_changed,
> -            lflow_input.sbrec_logical_flow_table,
> -            lflow_input.sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            return EN_UNHANDLED;
> -        }
> +        hmapx_add(&lflow_data->dirty_lflow_refs, route_node->lflow_ref);
>      }
>
> -    /* Now we handle created or updated route nodes. */
>      struct hmapx *crupdated_datapath_routes =
>          &group_ecmp_route_data->trk_data.crupdated_datapath_routes;
>      HMAPX_FOR_EACH (hmapx_node, crupdated_datapath_routes) {
> @@ -316,16 +271,7 @@ lflow_group_ecmp_route_change_handler(struct
> engine_node *node,
>          build_route_data_flows_for_lrouter(
>              route_node->od, lflow_data->lflow_table,
>              route_node, lflow_input.bfd_ports);
> -
> -        bool handled = lflow_ref_sync_lflows(
> -            route_node->lflow_ref, lflow_data->lflow_table,
> -            eng_ctx->ovnsb_idl_txn, lflow_input.dps,
> -            lflow_input.ovn_internal_version_changed,
> -            lflow_input.sbrec_logical_flow_table,
> -            lflow_input.sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            return EN_UNHANDLED;
> -        }
> +        hmapx_add(&lflow_data->dirty_lflow_refs, route_node->lflow_ref);
>      }
>
>      return EN_HANDLED_UPDATED;
> @@ -338,7 +284,6 @@ lflow_ic_learned_svc_mons_handler(struct engine_node
> *node,
>      struct ic_learned_svc_monitors_data *ic_learned_svc_monitors_data =
>          engine_get_input_data("ic_learned_svcs", node);
>
> -    const struct engine_context *eng_ctx = engine_get_context();
>      struct lflow_data *lflow_data = data;
>      struct lflow_input lflow_input;
>      lflow_get_input_data(node, &lflow_input);
> @@ -349,16 +294,8 @@ lflow_ic_learned_svc_mons_handler(struct engine_node
> *node,
>              &ic_learned_svc_monitors_data->ic_learned_svc_monitors_map,
>              ic_learned_svc_monitors_data->lflow_ref);
>
> -    if (!lflow_ref_resync_flows(
> -            ic_learned_svc_monitors_data->lflow_ref,
> -            lflow_data->lflow_table,
> -            eng_ctx->ovnsb_idl_txn,
> -            lflow_input.dps,
> -            lflow_input.ovn_internal_version_changed,
> -            lflow_input.sbrec_logical_flow_table,
> -            lflow_input.sbrec_logical_dp_group_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_ref_unlink_and_prune(ic_learned_svc_monitors_data->lflow_ref,
> +                              lflow_data->lflow_table);
>
>      build_lswitch_arp_nd_ic_learned_svc_mon(
>          &svc_mons_data,
> @@ -366,16 +303,9 @@ lflow_ic_learned_svc_mons_handler(struct engine_node
> *node,
>          lflow_input.svc_monitor_mac,
>          lflow_data->lflow_table);
>
> -    if (!lflow_ref_sync_lflows(
> -            ic_learned_svc_monitors_data->lflow_ref,
> -            lflow_data->lflow_table,
> -            eng_ctx->ovnsb_idl_txn,
> -            lflow_input.dps,
> -            lflow_input.ovn_internal_version_changed,
> -            lflow_input.sbrec_logical_flow_table,
> -            lflow_input.sbrec_logical_dp_group_table)) {
> -        return EN_UNHANDLED;
> -    }
> +    lflow_data->needs_full_sync = true;
> +    hmapx_add(&lflow_data->dirty_lflow_refs,
> +              ic_learned_svc_monitors_data->lflow_ref);
>
>      return EN_HANDLED_UPDATED;
>  }
> @@ -386,6 +316,7 @@ void *en_lflow_init(struct engine_node *node
> OVS_UNUSED,
>      struct lflow_data *data = xmalloc(sizeof *data);
>      data->lflow_table = lflow_table_alloc();
>      lflow_table_init(data->lflow_table);
> +    hmapx_init(&data->dirty_lflow_refs);
>      return data;
>  }
>
> @@ -393,4 +324,5 @@ void en_lflow_cleanup(void *data_)
>  {
>      struct lflow_data *data = data_;
>      lflow_table_destroy(data->lflow_table);
> +    hmapx_destroy(&data->dirty_lflow_refs);
>  }
> diff --git a/northd/en-lflow.h b/northd/en-lflow.h
> index 99bcfda15..5dab6cc19 100644
> --- a/northd/en-lflow.h
> +++ b/northd/en-lflow.h
> @@ -7,12 +7,15 @@
>  #include <stdlib.h>
>  #include <stdio.h>
>
> +#include "lib/hmapx.h"
>  #include "lib/inc-proc-eng.h"
>
>  struct lflow_table;
>
>  struct lflow_data {
>      struct lflow_table *lflow_table;
> +    struct hmapx dirty_lflow_refs; /* lflow_refs changed by handlers. */
> +    bool needs_full_sync;          /* Full lflow_table_sync_to_sb needed.
> */
>

This has very strange relation, it is cleared by the sync node?
Shouldn't this be part of the tracked_data and cleared before
every engine run? That's why we have those. Same goes for the
hmapx, it should be modified only by lflow node and consumed
by the sync.

 };
>
>  enum engine_node_state en_lflow_run(struct engine_node *node, void *data);
> diff --git a/northd/en-northd-output.c b/northd/en-northd-output.c
> index b492a771c..ed8dec0ac 100644
> --- a/northd/en-northd-output.c
> +++ b/northd/en-northd-output.c
> @@ -51,8 +51,8 @@ northd_output_sync_to_sb_handler(struct engine_node
> *node OVS_UNUSED,
>  }
>
>  enum engine_input_handler_result
> -northd_output_lflow_handler(struct engine_node *node OVS_UNUSED,
> -                            void *data OVS_UNUSED)
> +northd_output_dp_group_resolved_handler(struct engine_node *node
> OVS_UNUSED,
> +                                        void *data OVS_UNUSED)
>  {
>      return EN_HANDLED_UPDATED;
>  }
> diff --git a/northd/en-northd-output.h b/northd/en-northd-output.h
> index b7053e60c..bb5e35900 100644
> --- a/northd/en-northd-output.h
> +++ b/northd/en-northd-output.h
> @@ -13,7 +13,8 @@ enum engine_input_handler_result
>  northd_output_sync_to_sb_handler(struct engine_node *node,
>                                   void *data OVS_UNUSED);
>  enum engine_input_handler_result
> -northd_output_lflow_handler(struct engine_node *node, void *data
> OVS_UNUSED);
> +northd_output_dp_group_resolved_handler(struct engine_node *node,
> +                                        void *data OVS_UNUSED);
>  enum engine_input_handler_result
>  northd_output_mac_binding_aging_handler(struct engine_node *node,
>                                          void *data OVS_UNUSED);
> diff --git a/northd/inc-proc-northd.c b/northd/inc-proc-northd.c
> index d2f28666c..4c0c8acaa 100644
> --- a/northd/inc-proc-northd.c
> +++ b/northd/inc-proc-northd.c
> @@ -37,6 +37,7 @@
>  #include "en-multicast.h"
>  #include "en-northd.h"
>  #include "en-lflow.h"
> +#include "en-dp-group-resolved.h"
>  #include "en-northd-output.h"
>  #include "en-meters.h"
>  #include "en-sampling-app.h"
> @@ -160,7 +161,8 @@ enum sb_engine_node {
>  static ENGINE_NODE(northd, CLEAR_TRACKED_DATA, SB_WRITE);
>  static ENGINE_NODE(sync_from_sb, SB_WRITE);
>  static ENGINE_NODE(sampling_app);
> -static ENGINE_NODE(lflow, SB_WRITE);
> +static ENGINE_NODE(lflow);
> +static ENGINE_NODE(dp_group_resolved, SB_WRITE);
>  static ENGINE_NODE(mac_binding_aging, SB_WRITE);
>  static ENGINE_NODE(mac_binding_aging_waker);
>  static ENGINE_NODE(northd_output);
> @@ -399,9 +401,7 @@ void inc_proc_northd_init(struct ovsdb_idl_loop *nb,
>      engine_add_input(&en_multicast_igmp, &en_sb_igmp_group, NULL);
>
>      engine_add_input(&en_lflow, &en_sync_meters, NULL);
> -    engine_add_input(&en_lflow, &en_sb_logical_flow, NULL);
>      engine_add_input(&en_lflow, &en_sb_multicast_group, NULL);
> -    engine_add_input(&en_lflow, &en_sb_logical_dp_group, NULL);
>      engine_add_input(&en_lflow, &en_bfd_sync, NULL);
>      engine_add_input(&en_lflow, &en_route_policies, NULL);
>      /* Route changes are propagated to en_lflow through the
> en_group_ecmp_route
> @@ -436,6 +436,25 @@ void inc_proc_northd_init(struct ovsdb_idl_loop *nb,
>      engine_add_input(&en_lflow, &en_ic_learned_svc_monitors,
>                       lflow_ic_learned_svc_mons_handler);
>
> +    engine_add_input(&en_dp_group_resolved, &en_lflow,
> +                     dp_group_resolved_lflow_handler);
> +    engine_add_input(&en_dp_group_resolved, &en_sb_logical_flow, NULL);
> +    engine_add_input(&en_dp_group_resolved, &en_sb_logical_dp_group,
> NULL);
> +    /* dp_group_resolved reads synced datapath arrays (dps[DP_MAX]) when
> +     * calling lflow_table_sync_to_sb / lflow_ref_sync_lflows.  A noop
> +     * handler is sufficient because en_lflow already recomputes when
> +     * datapath_sync changes; dp_group_resolved just needs access to
> +     * the data. */
> +    engine_add_input(&en_dp_group_resolved, &en_datapath_sync,
> +                     engine_noop_handler);
> +    /* dp_group_resolved reads ovn_internal_version_changed from
> +     * global_config when syncing flows to SB.  Changes to
> +     * global_config propagate through en_lflow (which has its own
> +     * global_config handler), so a noop handler here avoids a
> +     * redundant recompute. */
> +    engine_add_input(&en_dp_group_resolved, &en_global_config,
> +                     engine_noop_handler);
> +
>      engine_add_input(&en_sync_to_sb_addr_set, &en_northd, NULL);
>      engine_add_input(&en_sync_to_sb_addr_set, &en_lr_stateful, NULL);
>      engine_add_input(&en_sync_to_sb_addr_set, &en_sb_address_set, NULL);
> @@ -490,8 +509,8 @@ void inc_proc_northd_init(struct ovsdb_idl_loop *nb,
>      engine_add_input(&en_northd_output, &en_sync_from_sb, NULL);
>      engine_add_input(&en_northd_output, &en_sync_to_sb,
>                       northd_output_sync_to_sb_handler);
> -    engine_add_input(&en_northd_output, &en_lflow,
> -                     northd_output_lflow_handler);
> +    engine_add_input(&en_northd_output, &en_dp_group_resolved,
> +                     northd_output_dp_group_resolved_handler);
>      engine_add_input(&en_northd_output, &en_mac_binding_aging,
>                       northd_output_mac_binding_aging_handler);
>      engine_add_input(&en_northd_output, &en_fdb_aging,
> diff --git a/northd/lflow-mgr.c b/northd/lflow-mgr.c
> index ce9c4f854..7a61212bc 100644
> --- a/northd/lflow-mgr.c
> +++ b/northd/lflow-mgr.c
> @@ -261,6 +261,28 @@ lflow_table_set_size(struct lflow_table *lflow_table,
> size_t size)
>      lflow_table->entries.n = size;
>  }
>
> +/* An lflow that applies to no datapaths (its dp group bitmap is empty)
> must
> + * not exist in the SB DB.  Such lflows can be left in the table by
> en_lflow's
> + * incremental handlers, which unlink lflows (clearing their dp bits) and
> defer
> + * the teardown to the dp-group-resolved sync stage.  Release the lflow's
> dp
> + * group and destroy it; its SB row (if any) is removed by the caller.
> Returns
> + * true if the lflow was destroyed. */
> +static bool
> +lflow_prune_if_no_datapaths(struct lflow_table *lflow_table,
> +                            struct ovn_lflow *lflow)
> +{
> +    if (dynamic_bitmap_count1(&lflow->dpg_bitmap)) {
> +        return false;
> +    }
> +
> +    enum ovn_datapath_type dp_type =
> ovn_stage_to_datapath_type(lflow->stage);
> +    ovs_assert(dp_type < DP_MAX);
> +    ovn_dp_group_release(&lflow_table->dp_groups[dp_type], lflow->dpg);
> +    lflow->dpg = NULL;
> +    ovn_lflow_destroy(lflow_table, lflow);
> +    return true;
> +}
> +
>  void
>  lflow_table_sync_to_sb(struct lflow_table *lflow_table,
>                         struct ovsdb_idl_txn *ovnsb_txn,
> @@ -287,6 +309,12 @@ lflow_table_sync_to_sb(struct lflow_table
> *lflow_table,
>              ovn_lflow_destroy(lflow_table, lflow);
>              continue;
>          }
> +        /* An lflow with no datapaths must be removed from the SB DB.  It
> is
> +         * skipped here (not added to 'sb_uuid_set'), so its SB row, if
> any, is
> +         * deleted by the reconciliation loop below. */
> +        if (lflow_prune_if_no_datapaths(lflow_table, lflow)) {
> +            continue;
> +        }
>          sbflow = NULL;
>          if (!uuid_is_zero(&lflow->sb_uuid)) {
>              sbflow = sbrec_logical_flow_table_get_for_uuid(sb_flow_table,
> @@ -372,7 +400,7 @@ lflow_table_sync_to_sb(struct lflow_table *lflow_table,
>              lflows, &stage,
>              sbflow->priority, sbflow->match, sbflow->actions,
>              sbflow->controller_meter, acl_ct_translation, sbflow->hash);
> -        if (lflow) {
> +        if (lflow && !lflow_prune_if_no_datapaths(lflow_table, lflow)) {
>              const struct ovn_synced_datapaths *datapaths;
>              struct hmap *dp_groups;
>              dp_groups = &lflow_table->dp_groups[dp_type];
> @@ -393,6 +421,9 @@ lflow_table_sync_to_sb(struct lflow_table *lflow_table,
>          if (search_mode != LFLOW_TABLE_SEARCH_FIELDS) {
>              break;
>          }
> +        if (lflow_prune_if_no_datapaths(lflow_table, lflow)) {
> +            continue;
> +        }
>          const struct ovn_synced_datapaths *datapaths;
>          struct hmap *dp_groups;
>          enum ovn_datapath_type dp_type =
> @@ -681,20 +712,32 @@ lflow_ref_unlink_lflows(struct lflow_ref *lflow_ref)
>      }
>  }
>
> -bool
> -lflow_ref_resync_flows(struct lflow_ref *lflow_ref,
> -                       struct lflow_table *lflow_table,
> -                       struct ovsdb_idl_txn *ovnsb_txn,
> -                       const struct ovn_synced_datapaths dps[DP_MAX],
> -                       bool ovn_internal_version_changed,
> -                       const struct sbrec_logical_flow_table
> *sbflow_table,
> -                       const struct sbrec_logical_dp_group_table
> *dpgrp_table)
> +/* Unlinks and destroys all lrns in 'lflow_ref', then destroys any lflow
> + * whose referenced_by list is empty (no other lflow_ref references it).
> + * Unlike lflow_ref_unlink_lflows (which only clears dp bits and sets
> + * linked=false), this function removes the lrns and orphaned lflows
> + * from the in-memory table entirely, without writing to SB. */
> +void
> +lflow_ref_unlink_and_prune(struct lflow_ref *lflow_ref,
> +                           struct lflow_table *lflow_table)
>  {
>      lflow_ref_unlink_lflows(lflow_ref);
> -    return lflow_ref_sync_lflows__(lflow_ref, lflow_table, ovnsb_txn,
> -                                   dps,
> -                                   ovn_internal_version_changed,
> sbflow_table,
> -                                   dpgrp_table);
> +
> +    struct lflow_ref_node *lrn;
> +    HMAP_FOR_EACH_SAFE (lrn, ref_node, &lflow_ref->lflow_ref_nodes) {
> +        struct ovn_lflow *lflow = lrn->lflow;
> +        lflow_ref_node_destroy(lrn);
> +
> +        if (ovs_list_is_empty(&lflow->referenced_by)) {
> +            enum ovn_datapath_type dp_type =
> +                ovn_stage_to_datapath_type(lflow->stage);
> +            ovs_assert(dp_type < DP_MAX);
> +            ovn_dp_group_release(&lflow_table->dp_groups[dp_type],
> +                                 lflow->dpg);
> +            lflow->dpg = NULL;
> +            ovn_lflow_destroy(lflow_table, lflow);
> +        }
> +    }
>  }
>
>  bool
> @@ -781,10 +824,78 @@ lflow_table_add_lflow__(struct lflow_table
> *lflow_table,
>              }
>              ovs_list_insert(&lflow->referenced_by, &lrn->ref_list_node);
>              hmap_insert(&lflow_ref->lflow_ref_nodes, &lrn->ref_node,
> hash);
> +        } else if (sdp) {
> +            /* Single-datapath (re-)add of an existing node. */
> +            if (!lrn->dpgrp_lflow) {
> +                if (!lrn->linked) {
> +                    /* First add of this (re-)link cycle establishes the
> base
> +                     * datapath this reference tracks for the lflow. */
> +                    lrn->dp_index = sdp->index;
> +                } else if (lrn->dp_index != sdp->index) {
> +                    /* The same 'lflow_ref' references this lflow L(M, A)
> for a
> +                     * second datapath in this cycle (e.g. the single
> lflow_ref
> +                     * shared by all IGMP flows).  A single-datapath
> +                     * lflow_ref_node tracks only one datapath index, so
> +                     * lflow_ref_unlink_lflows() would leave the other
> +                     * datapaths set in the lflow's dp group bitmap.
> Upgrade
> +                     * the node to track a datapath bitmap instead, so
> that
> +                     * unlinking clears every datapath this reference
> +                     * contributed. */
> +                    size_t len = sparse_array_len(&sdp->dps->dps_array);
> +                    lrn->dpgrp_bitmap = bitmap_allocate(len);
> +                    lrn->dpgrp_bitmap_len = len;
> +                    bitmap_set1(lrn->dpgrp_bitmap, lrn->dp_index);
> +                    bitmap_set1(lrn->dpgrp_bitmap, sdp->index);
> +                    lrn->dpgrp_lflow = true;
> +
> +                    /* This reference already accounted for
> 'lrn->dp_index'
> +                     * (the first datapath it added this cycle) in the
> block
> +                     * below.  This add contributes a second datapath,
> which
> +                     * the block below will not see because it only runs
> on
> +                     * the first link of the cycle.  Account for it here
> so
> +                     * that a datapath shared with another reference is
> not
> +                     * released prematurely. */
> +                    if (dynamic_bitmap_is_set(&lflow->dpg_bitmap,
> +                                              sdp->index)) {
> +                        dp_refcnt_use(&lflow->dp_refcnts_map, sdp->index);
> +                    }
> +                }
> +            } else {
> +                /* A node previously upgraded to track a datapath bitmap
> is
> +                 * re-linked one datapath at a time. */
> +                if (!lrn->linked) {
> +                    /* First add of this (re-)link cycle: rebuild the
> tracked
> +                     * bitmap from scratch so it reflects the current
> datapath
> +                     * set and the current datapath array length (which
> may
> +                     * have grown or shrunk since the node was last
> linked). */
> +                    size_t len = sparse_array_len(&sdp->dps->dps_array);
> +                    bitmap_free(lrn->dpgrp_bitmap);
> +                    lrn->dpgrp_bitmap = bitmap_allocate(len);
> +                    lrn->dpgrp_bitmap_len = len;
> +                } else if (!bitmap_is_set(lrn->dpgrp_bitmap, sdp->index))
> {
> +                    /* A second or later datapath this reference
> contributes in
> +                     * the same cycle.  The first datapath was accounted
> for by
> +                     * the block below (which only runs on the first link
> of
> +                     * the cycle); account for this one too so that a
> datapath
> +                     * shared with another reference is not released
> +                     * prematurely. */
> +                    if (dynamic_bitmap_is_set(&lflow->dpg_bitmap,
> +                                              sdp->index)) {
> +                        dp_refcnt_use(&lflow->dp_refcnts_map, sdp->index);
> +                    }
> +                }
> +                bitmap_set1(lrn->dpgrp_bitmap, sdp->index);
> +            }
>          }
>
>          if (!lrn->linked) {
> -            if (lrn->dpgrp_lflow) {
> +            /* Allocate a reference counter only if the datapath(s) added
> by
> +             * this reference are already used by the lflow. */
> +            if (sdp) {
> +                if (dynamic_bitmap_is_set(&lflow->dpg_bitmap,
> sdp->index)) {
> +                    dp_refcnt_use(&lflow->dp_refcnts_map, sdp->index);
> +                }
> +            } else {
>                  ovs_assert(lrn->dpgrp_bitmap_len == dp_bitmap_len);
>                  size_t index;
>                  BITMAP_FOR_EACH_1 (index, dp_bitmap_len, dp_bitmap) {
> @@ -793,11 +904,6 @@ lflow_table_add_lflow__(struct lflow_table
> *lflow_table,
>                          dp_refcnt_use(&lflow->dp_refcnts_map, index);
>                      }
>                  }
> -            } else {
> -                /* Allocate a reference counter only if already used. */
> -                if (dynamic_bitmap_is_set(&lflow->dpg_bitmap,
> lrn->dp_index)) {
> -                    dp_refcnt_use(&lflow->dp_refcnts_map, lrn->dp_index);
> -                }
>              }
>          }
>          lrn->linked = true;
> @@ -1231,17 +1337,12 @@ sync_lflow_to_sb(struct ovn_lflow *lflow,
>                  &lflow->dpg->dpg_uuid);
>
>              if (!lflow->dpg->dp_group) {
> -                /* Ideally this should not happen.  But it can still
> happen
> -                 * due to 2 reasons:
> -                 * 1. There is a bug in the dp_group management.  We
> should
> -                 *    perhaps assert here.
> -                 * 2. A User or CMS may delete the logical_dp_groups in
> SB DB
> -                 *    or clear the SB:Logical_flow.logical_dp_groups
> column
> -                 *    (intentionally or accidentally)
> -                 *
> -                 * Because of (2) it is better to return false instead of
> -                 * assert,so that we recover from th inconsistent SB DB.
> -                 */
> +                /* The SB Logical_DP_Group row referenced by this
> in-memory dp
> +                 * group no longer exists.  This can happen if a user or
> CMS
> +                 * deletes the SB:Logical_DP_Group rows or clears the
> +                 * SB:Logical_Flow.logical_dp_group column (intentionally
> or
> +                 * accidentally).  Release the stale dp group and fall
> through
> +                 * to create a fresh one. */
>                  static struct vlog_rate_limit rl =
> VLOG_RATE_LIMIT_INIT(1, 1);
>                  VLOG_WARN_RL(&rl, "SB Logical flow ["UUID_FMT"]'s "
>                              "logical_dp_group column is not set "
> @@ -1249,10 +1350,11 @@ sync_lflow_to_sb(struct ovn_lflow *lflow,
>                              "referencing the dp group ["UUID_FMT"]",
>                              UUID_ARGS(&sbflow->header_.uuid),
>                              UUID_ARGS(&lflow->dpg->dpg_uuid));
> -                lflow->sync_state = LFLOW_STALE;
> -                return false;
> +                ovn_dp_group_release(dp_groups, lflow->dpg);
> +                lflow->dpg = NULL;
>

Caught by AI:

ovn_dp_group_get() can return an in-memory group shared by multiple
lflows.  In that case ovn_dp_group_release() only drops this lflow's
reference, leaving the stale group in dp_groups, and
ovn_dp_group_create() inserts another group with the same bitmap.
Subsequent lflows can move to the new group, but sync_lflow_to_sb() only
increments the new group at the end; it does not release their
pre_sync_dpg references.  The stale group and its reference counts can
therefore remain indefinitely.


>              }
> -        } else {
> +        }
> +        if (!lflow->dpg) {
>              lflow->dpg = ovn_dp_group_create(
>                                  ovnsb_txn, dp_groups, sbrec_dp_group,
>                                  &lflow->dpg_bitmap,
> diff --git a/northd/lflow-mgr.h b/northd/lflow-mgr.h
> index 84d0b3e67..e924ffc22 100644
> --- a/northd/lflow-mgr.h
> +++ b/northd/lflow-mgr.h
> @@ -58,13 +58,7 @@ struct lflow_ref *lflow_ref_create(void);
>  void lflow_ref_destroy(struct lflow_ref *);
>  void lflow_ref_clear(struct lflow_ref *lflow_ref);
>  void lflow_ref_unlink_lflows(struct lflow_ref *);
> -bool lflow_ref_resync_flows(struct lflow_ref *,
> -                            struct lflow_table *lflow_table,
> -                            struct ovsdb_idl_txn *ovnsb_txn,
> -                            const struct ovn_synced_datapaths dps[DP_MAX],
> -                            bool ovn_internal_version_changed,
> -                            const struct sbrec_logical_flow_table *,
> -                            const struct sbrec_logical_dp_group_table *);
> +void lflow_ref_unlink_and_prune(struct lflow_ref *, struct lflow_table *);
>  bool lflow_ref_sync_lflows(struct lflow_ref *,
>                             struct lflow_table *lflow_table,
>                             struct ovsdb_idl_txn *ovnsb_txn,
> diff --git a/northd/northd.c b/northd/northd.c
> index 47f76b108..fbc0068b7 100644
> --- a/northd/northd.c
> +++ b/northd/northd.c
> @@ -20965,10 +20965,9 @@ void run_update_worker_pool(int n_threads)
>      }
>  }
>
> -/* Updates the Logical_Flow and Multicast_Group tables in the OVN_SB
> database,
> - * constructing their contents based on the OVN_NB database. */
> -void build_lflows(struct ovsdb_idl_txn *ovnsb_txn,
> -                  struct lflow_input *input_data,
> +/* Builds the in-memory logical flow table from the OVN_NB database.
> + * The flows are synced to the SB database by en_dp_group_resolved. */
> +void build_lflows(struct lflow_input *input_data,
>                    struct lflow_table *lflows)
>  {
>      struct svc_monitors_map_data svc_mons_data =
> @@ -21012,14 +21011,6 @@ void build_lflows(struct ovsdb_idl_txn *ovnsb_txn,
>      /* Parallel build may result in a suboptimal hash. Resize the
>       * lflow map to a correct size before doing lookups */
>      lflow_table_expand(lflows);
> -
> -    stopwatch_start(LFLOWS_TO_SB_STOPWATCH_NAME, time_msec());
> -    lflow_table_sync_to_sb(lflows, ovnsb_txn, input_data->dps,
> -                           input_data->ovn_internal_version_changed,
> -                           input_data->sbrec_logical_flow_table,
> -                           input_data->sbrec_logical_dp_group_table);
> -
> -    stopwatch_stop(LFLOWS_TO_SB_STOPWATCH_NAME, time_msec());
>  }
>
>  void
> @@ -21064,24 +21055,17 @@ lflow_reset_northd_refs(struct lflow_input
> *lflow_input)
>      }
>  }
>
> -bool
> -lflow_handle_northd_lr_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                                struct tracked_dps *tracked_lrs,
> -                                struct lflow_input *lflow_input,
> -                                struct lflow_table *lflows)
> +void
> +lflow_handle_northd_lr_changes(struct tracked_dps *tracked_lrs,
> +                               struct lflow_input *lflow_input,
> +                               struct lflow_table *lflows,
> +                               struct hmapx *dirty_lflow_refs)
>  {
> -    bool handled = true;
>      struct hmapx_node *hmapx_node;
>      HMAPX_FOR_EACH (hmapx_node, &tracked_lrs->deleted) {
>          struct ovn_datapath *od = hmapx_node->data;
> -        handled = lflow_ref_resync_flows(
> -            od->datapath_lflows, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            return handled;
> -        }
> +        lflow_ref_unlink_lflows(od->datapath_lflows);
> +        hmapx_add(dirty_lflow_refs, od->datapath_lflows);
>      }
>
>      struct lswitch_flow_build_info lsi = {
> @@ -21099,34 +21083,18 @@ lflow_handle_northd_lr_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>
>          lflow_ref_unlink_lflows(od->datapath_lflows);
>          build_lswitch_and_lrouter_iterate_by_lr(od, &lsi);
> -    }
> -
> -    /* We need to make sure that all datapath groups are allocated before
> -     * trying to sync logical flows. Otherwise, we would need to recompute
> -     * those datapath groups within those flows over and over again. */
> -    HMAPX_FOR_EACH (hmapx_node, &tracked_lrs->crupdated) {
> -        struct ovn_datapath *od = hmapx_node->data;
> -
> -        handled = lflow_ref_sync_lflows(
> -            od->datapath_lflows, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            break;
> -        }
> +        hmapx_add(dirty_lflow_refs, od->datapath_lflows);
>      }
>
>      ds_destroy(&lsi.actions);
>      ds_destroy(&lsi.match);
> -    return handled;
>  }
>
> -bool
> -lflow_handle_northd_port_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                                 struct tracked_ovn_ports *trk_lsps,
> +void
> +lflow_handle_northd_port_changes(struct tracked_ovn_ports *trk_lsps,
>                                   struct lflow_input *lflow_input,
> -                                 struct lflow_table *lflows)
> +                                 struct lflow_table *lflows,
> +                                 struct hmapx *dirty_lflow_refs)
>  {
>      struct hmapx_node *hmapx_node;
>      struct ovn_port *op;
> @@ -21135,14 +21103,8 @@ lflow_handle_northd_port_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>          op = hmapx_node->data;
>          /* Make sure 'op' is an lsp and not lrp. */
>          ovs_assert(op->nbsp);
> -        bool handled = lflow_ref_resync_flows(
> -            op->lflow_ref, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            return false;
> -        }
> +        lflow_ref_unlink_lflows(op->lflow_ref);
> +        hmapx_add(dirty_lflow_refs, op->lflow_ref);
>          /* No need to update SB multicast groups, thanks to weak
>           * references. */
>      }
> @@ -21163,33 +21125,17 @@ lflow_handle_northd_port_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>                                                   &match, &actions,
>
> lflow_input->svc_monitor_mac,
>                                                   lflows);
> -        /* Sync the new flows to SB. */
> -        bool handled = lflow_ref_sync_lflows(
> -            op->lflow_ref, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (handled) {
> -            /* Now regenerate the stateful lflows for 'op' */
> -            /* Clear old lflows. */
> -            lflow_ref_unlink_lflows(op->stateful_lflow_ref);
> -            build_lbnat_lflows_iterate_by_lsp(op,
> -
> lflow_input->lr_stateful_table,
> -                                              &match, &actions, lflows);
> -            handled = lflow_ref_sync_lflows(
> -                op->stateful_lflow_ref, lflows, ovnsb_txn,
> -                lflow_input->dps,
> -                lflow_input->ovn_internal_version_changed,
> -                lflow_input->sbrec_logical_flow_table,
> -                lflow_input->sbrec_logical_dp_group_table);
> -        }
> +        hmapx_add(dirty_lflow_refs, op->lflow_ref);
> +
> +        /* Now regenerate the stateful lflows for 'op' */
> +        lflow_ref_unlink_lflows(op->stateful_lflow_ref);
> +        build_lbnat_lflows_iterate_by_lsp(op,
> +                                          lflow_input->lr_stateful_table,
> +                                          &match, &actions, lflows);
> +        hmapx_add(dirty_lflow_refs, op->stateful_lflow_ref);
>
>          ds_destroy(&match);
>          ds_destroy(&actions);
> -
> -        if (!handled) {
> -            return false;
> -        }
>      }
>
>      HMAPX_FOR_EACH (hmapx_node, &trk_lsps->created) {
> @@ -21205,42 +21151,24 @@ lflow_handle_northd_port_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>                                                   &match, &actions,
>
> lflow_input->svc_monitor_mac,
>                                                   lflows);
> +        hmapx_add(dirty_lflow_refs, op->lflow_ref);
>
> -        /* Sync the newly added flows to SB. */
> -        bool handled = lflow_ref_sync_lflows(
> -            op->lflow_ref, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (handled) {
> -            /* Now generate the stateful lflows for 'op' */
> -            build_lbnat_lflows_iterate_by_lsp(op,
> -
> lflow_input->lr_stateful_table,
> -                                              &match, &actions, lflows);
> -            handled = lflow_ref_sync_lflows(
> -                op->stateful_lflow_ref, lflows, ovnsb_txn,
> -                lflow_input->dps,
> -                lflow_input->ovn_internal_version_changed,
> -                lflow_input->sbrec_logical_flow_table,
> -                lflow_input->sbrec_logical_dp_group_table);
> -        }
> +        /* Now generate the stateful lflows for 'op' */
> +        build_lbnat_lflows_iterate_by_lsp(op,
> +                                          lflow_input->lr_stateful_table,
> +                                          &match, &actions, lflows);
> +        hmapx_add(dirty_lflow_refs, op->stateful_lflow_ref);
>
>          ds_destroy(&match);
>          ds_destroy(&actions);
> -
> -        if (!handled) {
> -            return false;
> -        }
>      }
> -
> -    return true;
>  }
>
> -bool
> -lflow_handle_northd_lb_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                               struct tracked_lbs *trk_lbs,
> +void
> +lflow_handle_northd_lb_changes(struct tracked_lbs *trk_lbs,
>                                 struct lflow_input *lflow_input,
> -                               struct lflow_table *lflows)
> +                               struct lflow_table *lflows,
> +                               struct hmapx *dirty_lflow_refs)
>  {
>      struct ovn_lb_datapaths *lb_dps;
>      struct hmapx_node *hmapx_node;
> @@ -21252,12 +21180,8 @@ lflow_handle_northd_lb_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>
>      HMAPX_FOR_EACH (hmapx_node, &trk_lbs->deleted) {
>          lb_dps = hmapx_node->data;
> -
> -        lflow_ref_resync_flows(
> -            lb_dps->lflow_ref, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> +        lflow_ref_unlink_lflows(lb_dps->lflow_ref);
> +        hmapx_add(dirty_lflow_refs, lb_dps->lflow_ref);
>      }
>
>      HMAPX_FOR_EACH (hmapx_node, &trk_lbs->crupdated) {
> @@ -21291,31 +21215,20 @@ lflow_handle_northd_lb_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>          ds_destroy(&match);
>          ds_destroy(&actions);
>
> -        /* Sync the new flows to SB. */
> -        bool handled = lflow_ref_sync_lflows(
> -            lb_dps->lflow_ref, lflows, ovnsb_txn, lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            return false;
> -        }
> +        hmapx_add(dirty_lflow_refs, lb_dps->lflow_ref);
>      }
> -
> -    return true;
>  }
>
> -bool
> -lflow_handle_lr_stateful_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                                struct lr_stateful_tracked_data *trk_data,
> -                                struct lflow_input *lflow_input,
> -                                struct lflow_table *lflows)
> +void
> +lflow_handle_lr_stateful_changes(struct lr_stateful_tracked_data
> *trk_data,
> +                                 struct lflow_input *lflow_input,
> +                                 struct lflow_table *lflows,
> +                                 struct hmapx *dirty_lflow_refs)
>  {
>      struct lr_stateful_record *lr_stateful_rec;
>      struct ds actions = DS_EMPTY_INITIALIZER;
>      struct ds match = DS_EMPTY_INITIALIZER;
>      struct hmapx_node *hmapx_node;
> -    bool handled = true;
>
>      HMAPX_FOR_EACH (hmapx_node, &trk_data->crupdated) {
>          lr_stateful_rec = hmapx_node->data;
> @@ -21328,17 +21241,7 @@ lflow_handle_lr_stateful_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>                                  &match, &actions,
>                                  lflow_input->meter_groups,
>                                  lflow_input->features);
> -
> -        /* Sync the new flows to SB. */
> -        handled = lflow_ref_sync_lflows(
> -            lr_stateful_rec->lflow_ref, lflows, ovnsb_txn,
> -            lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            goto exit;
> -        }
> +        hmapx_add(dirty_lflow_refs, lr_stateful_rec->lflow_ref);
>
>          const struct ovn_datapath *od =
>              ovn_datapaths_find_by_index(lflow_input->lr_datapaths,
> @@ -21353,16 +21256,7 @@ lflow_handle_lr_stateful_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>                                                lflow_input->bfd_ports,
>                                                &match, &actions,
>                                                lflows);
> -
> -            handled = lflow_ref_sync_lflows(
> -                op->stateful_lflow_ref, lflows, ovnsb_txn,
> -                lflow_input->dps,
> -                lflow_input->ovn_internal_version_changed,
> -                lflow_input->sbrec_logical_flow_table,
> -                lflow_input->sbrec_logical_dp_group_table);
> -            if (!handled) {
> -                goto exit;
> -            }
> +            hmapx_add(dirty_lflow_refs, op->stateful_lflow_ref);
>
>              if (op->peer && op->peer->nbsp) {
>                  lflow_ref_unlink_lflows(op->peer->stateful_lflow_ref);
> @@ -21370,32 +21264,20 @@ lflow_handle_lr_stateful_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>                  build_lbnat_lflows_iterate_by_lsp(
>                      op->peer, lflow_input->lr_stateful_table, &match,
> &actions,
>                      lflows);
> -
> -                handled = lflow_ref_sync_lflows(
> -                    op->peer->stateful_lflow_ref, lflows, ovnsb_txn,
> -                    lflow_input->dps,
> -                    lflow_input->ovn_internal_version_changed,
> -                    lflow_input->sbrec_logical_flow_table,
> -                    lflow_input->sbrec_logical_dp_group_table);
> -                if (!handled) {
> -                    goto exit;
> -                }
> +                hmapx_add(dirty_lflow_refs, op->peer->stateful_lflow_ref);
>              }
>          }
>      }
>
> -exit:
>      ds_destroy(&match);
>      ds_destroy(&actions);
> -
> -    return handled;
>  }
>
> -bool
> -lflow_handle_ls_stateful_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                                struct ls_stateful_tracked_data *trk_data,
> -                                struct lflow_input *lflow_input,
> -                                struct lflow_table *lflows)
> +void
> +lflow_handle_ls_stateful_changes(struct ls_stateful_tracked_data
> *trk_data,
> +                                 struct lflow_input *lflow_input,
> +                                 struct lflow_table *lflows,
> +                                 struct hmapx *dirty_lflow_refs)
>  {
>      struct hmapx_node *hmapx_node;
>
> @@ -21420,39 +21302,14 @@ lflow_handle_ls_stateful_changes(struct
> ovsdb_idl_txn *ovnsb_txn,
>          build_network_function(od, lflows,
>                                 lflow_input->ls_port_groups,
>                                 ls_stateful_rec->lflow_ref);
> -    }
> -
> -    /* We need to make sure that all datapath groups are allocated before
> -     * trying to sync logical flows. Otherwise, we would need to recompute
> -     * those datapath groups within those flows over and over again. */
> -    HMAPX_FOR_EACH (hmapx_node, &trk_data->crupdated) {
> -        struct ls_stateful_record *ls_stateful_rec = hmapx_node->data;
> -        /* Sync the new flows to SB. */
> -        bool handled = lflow_ref_sync_lflows(
> -            ls_stateful_rec->lflow_ref, lflows, ovnsb_txn,
> -            lflow_input->dps,
> -            lflow_input->ovn_internal_version_changed,
> -            lflow_input->sbrec_logical_flow_table,
> -            lflow_input->sbrec_logical_dp_group_table);
> -        if (!handled) {
> -            return false;
> -        }
> +        hmapx_add(dirty_lflow_refs, ls_stateful_rec->lflow_ref);
>      }
>
>      HMAPX_FOR_EACH (hmapx_node, &trk_data->deleted) {
>          struct ls_stateful_record *ls_stateful_rec = hmapx_node->data;
> -
> -        if (!lflow_ref_resync_flows(
> -                    ls_stateful_rec->lflow_ref, lflows, ovnsb_txn,
> -                    lflow_input->dps,
> -                    lflow_input->ovn_internal_version_changed,
> -                    lflow_input->sbrec_logical_flow_table,
> -                    lflow_input->sbrec_logical_dp_group_table)) {
> -            return false;
> -        }
> +        lflow_ref_unlink_lflows(ls_stateful_rec->lflow_ref);
> +        hmapx_add(dirty_lflow_refs, ls_stateful_rec->lflow_ref);
>      }
> -
> -    return true;
>  }
>
>  static bool
> diff --git a/northd/northd.h b/northd/northd.h
> index 2e3a9e00d..df2189c97 100644
> --- a/northd/northd.h
> +++ b/northd/northd.h
> @@ -260,9 +260,7 @@ struct lflow_ref;
>  struct lr_nat_table;
>
>  struct lflow_input {
> -    /* Southbound table references */
> -    const struct sbrec_logical_flow_table *sbrec_logical_flow_table;
> -    const struct sbrec_logical_dp_group_table
> *sbrec_logical_dp_group_table;
> +    /* Southbound table references (used during lflow computation). */
>      const struct sbrec_acl_id_table *sbrec_acl_id_table;
>
>      /* Indexes */
> @@ -979,8 +977,7 @@ struct lr_stateful_tracked_data;
>  struct ls_stateful_tracked_data;
>  struct group_ecmp_datapath;
>
> -void build_lflows(struct ovsdb_idl_txn *ovnsb_txn,
> -                  struct lflow_input *input_data,
> +void build_lflows(struct lflow_input *input_data,
>                    struct lflow_table *);
>  void lflow_reset_northd_refs(struct lflow_input *);
>  void build_route_data_flows_for_lrouter(
> @@ -988,26 +985,26 @@ void build_route_data_flows_for_lrouter(
>      const struct group_ecmp_datapath *route_node,
>      const struct sset *bfd_ports);
>
> -bool lflow_handle_northd_lr_changes(struct ovsdb_idl_txn *ovnsh_txn,
> -                                     struct tracked_dps *,
> -                                     struct lflow_input *,
> -                                     struct lflow_table *lflows);
> -bool lflow_handle_northd_port_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                                      struct tracked_ovn_ports *,
> +void lflow_handle_northd_lr_changes(struct tracked_dps *,
> +                                    struct lflow_input *,
> +                                    struct lflow_table *lflows,
> +                                    struct hmapx *dirty_lflow_refs);
> +void lflow_handle_northd_port_changes(struct tracked_ovn_ports *,
>                                        struct lflow_input *,
> -                                      struct lflow_table *lflows);
> -bool lflow_handle_northd_lb_changes(struct ovsdb_idl_txn *ovnsb_txn,
> -                                    struct tracked_lbs *,
> +                                      struct lflow_table *lflows,
> +                                      struct hmapx *dirty_lflow_refs);
> +void lflow_handle_northd_lb_changes(struct tracked_lbs *,
>                                      struct lflow_input *,
> -                                    struct lflow_table *lflows);
> -bool lflow_handle_lr_stateful_changes(struct ovsdb_idl_txn *,
> -                                      struct lr_stateful_tracked_data *,
> +                                    struct lflow_table *lflows,
> +                                    struct hmapx *dirty_lflow_refs);
> +void lflow_handle_lr_stateful_changes(struct lr_stateful_tracked_data *,
>                                        struct lflow_input *,
> -                                      struct lflow_table *lflows);
> -bool lflow_handle_ls_stateful_changes(struct ovsdb_idl_txn *,
> -                                      struct ls_stateful_tracked_data *,
> +                                      struct lflow_table *lflows,
> +                                      struct hmapx *dirty_lflow_refs);
> +void lflow_handle_ls_stateful_changes(struct ls_stateful_tracked_data *,
>                                        struct lflow_input *,
> -                                      struct lflow_table *lflows);
> +                                      struct lflow_table *lflows,
> +                                      struct hmapx *dirty_lflow_refs);
>  bool northd_handle_sb_port_binding_changes(
>      const struct sbrec_port_binding_table *, struct hmap *ls_ports,
>      struct hmap *lr_ports);
> diff --git a/tests/ovn-inc-proc-graph-dump.at b/tests/
> ovn-inc-proc-graph-dump.at
> index 44bf5689e..af951da8d 100644
> --- a/tests/ovn-inc-proc-graph-dump.at
> +++ b/tests/ovn-inc-proc-graph-dump.at
> @@ -143,7 +143,6 @@ digraph "Incremental-Processing-Engine" {
>         sync_meters -> sync_to_sb [[label=""]];
>         sync_to_sb_lb -> sync_to_sb [[label=""]];
>         sync_to_sb_pb -> sync_to_sb [[label=""]];
> -       SB_logical_flow [[style=filled, shape=box, fillcolor=white,
> label="SB_logical_flow"]];
>         SB_multicast_group [[style=filled, shape=box, fillcolor=white,
> label="SB_multicast_group"]];
>         NB_bfd [[style=filled, shape=box, fillcolor=white,
> label="NB_bfd"]];
>         SB_bfd [[style=filled, shape=box, fillcolor=white,
> label="SB_bfd"]];
> @@ -186,9 +185,7 @@ digraph "Incremental-Processing-Engine" {
>         SB_igmp_group -> multicast_igmp [[label=""]];
>         lflow [[style=filled, shape=box, fillcolor=white, label="lflow"]];
>         sync_meters -> lflow [[label=""]];
> -       SB_logical_flow -> lflow [[label=""]];
>         SB_multicast_group -> lflow [[label=""]];
> -       SB_logical_dp_group -> lflow [[label=""]];
>         bfd_sync -> lflow [[label=""]];
>         route_policies -> lflow [[label=""]];
>         routes -> lflow [[label="engine_noop_handler"]];
> @@ -203,6 +200,13 @@ digraph "Incremental-Processing-Engine" {
>         multicast_igmp -> lflow [[label="lflow_multicast_igmp_handler"]];
>         SB_acl_id -> lflow [[label=""]];
>         ic_learned_svc_monitors -> lflow
> [[label="lflow_ic_learned_svc_mons_handler"]];
> +       SB_logical_flow [[style=filled, shape=box, fillcolor=white,
> label="SB_logical_flow"]];
> +       dp_group_resolved [[style=filled, shape=box, fillcolor=white,
> label="dp_group_resolved"]];
> +       lflow -> dp_group_resolved
> [[label="dp_group_resolved_lflow_handler"]];
> +       SB_logical_flow -> dp_group_resolved [[label=""]];
> +       SB_logical_dp_group -> dp_group_resolved [[label=""]];
> +       datapath_sync -> dp_group_resolved [[label="engine_noop_handler"]];
> +       global_config -> dp_group_resolved [[label="engine_noop_handler"]];
>         mac_binding_aging_waker [[style=filled, shape=box,
> fillcolor=white, label="mac_binding_aging_waker"]];
>         mac_binding_aging [[style=filled, shape=box, fillcolor=white,
> label="mac_binding_aging"]];
>         SB_mac_binding -> mac_binding_aging [[label=""]];
> @@ -237,7 +241,7 @@ digraph "Incremental-Processing-Engine" {
>         acl_id -> northd_output [[label="northd_output_acl_id_handler"]];
>         sync_from_sb -> northd_output [[label=""]];
>         sync_to_sb -> northd_output
> [[label="northd_output_sync_to_sb_handler"]];
> -       lflow -> northd_output [[label="northd_output_lflow_handler"]];
> +       dp_group_resolved -> northd_output
> [[label="northd_output_dp_group_resolved_handler"]];
>         mac_binding_aging -> northd_output
> [[label="northd_output_mac_binding_aging_handler"]];
>         fdb_aging -> northd_output
> [[label="northd_output_fdb_aging_handler"]];
>         ecmp_nexthop -> northd_output
> [[label="northd_output_ecmp_nexthop_handler"]];
> diff --git a/tests/ovn-northd.at b/tests/ovn-northd.at
> index d19978a62..2a070901b 100644
> --- a/tests/ovn-northd.at
> +++ b/tests/ovn-northd.at
> @@ -13266,17 +13266,7 @@ check ovn-nbctl --wait=sb set load_balancer lb1
> options:bar=foo
>  check_engine_stats lflow norecompute compute
>  CHECK_NO_CHANGE_AFTER_RECOMPUTE
>
> -# Clear the SB:Logical_Flow.logical_dp_groups column of all the
> -# logical flows and then modify the NB:Load_balancer.  ovn-northd
> -# should resync the logical flows.
> -for l in $(ovn-sbctl --bare --columns _uuid list logical_flow)
> -do
> -    check ovn-sbctl clear logical_flow $l logical_dp_group
> -done
> -
> -check as northd ovn-appctl -t ovn-northd inc-engine/clear-stats
>

Why was this removed? We should still check if the resync happens
in those cases.

 check ovn-nbctl --wait=sb set load_balancer lb1 options:foo=bar
> -check_engine_stats lflow recompute nocompute
>  CHECK_NO_CHANGE_AFTER_RECOMPUTE
>
>  lb_lflow_uuid=$(fetch_column Logical_flow _uuid match='"ct.new &&
> ip4.dst == 10.0.0.10 && reg1[[16..23]] == 6 && reg1[[0..15]] == 80"')
> @@ -13445,16 +13435,7 @@ AT_CHECK([echo $dpgrp_dps | grep $sw3_uuid], [0],
> [ignore])
>  AT_CHECK([echo $dpgrp_dps | grep $sw4_uuid], [0], [ignore])
>  AT_CHECK([echo $dpgrp_dps | grep $sw5_uuid], [0], [ignore])
>
> -# Clear the SB:Logical_Flow.logical_dp_groups column of all the
> -# logical flows and then modify the NB:Load_balancer.  ovn-northd
> -# should resync the logical flows.
> -for l in $(ovn-sbctl --bare --columns _uuid list logical_flow)
> -do
> -    check ovn-sbctl clear logical_flow $l logical_dp_group
> -done
> -
>  check ovn-nbctl --wait=sb set load_balancer lb2 vips='{"10.0.0.10:80"="
> 10.0.0.3:80,10.0.0.4:80"}'
> -check_engine_stats lflow recompute compute


Same here.


>  CHECK_NO_CHANGE_AFTER_RECOMPUTE
>
>  lb_lflow_uuid=$(fetch_column Logical_flow _uuid match='"ct.new &&
> ip4.dst == 10.0.0.10 && reg1[[16..23]] == 6 && reg1[[0..15]] == 80"')
> @@ -23883,3 +23864,65 @@ AT_CHECK([as northd ovn-appctl -t ovn-northd
> inc-engine/enable-stopwatch nonexis
>  OVN_CLEANUP_NORTHD
>  AT_CLEANUP
>  ])
> +
> +AT_SETUP([Datapath group SB writes only when bitmap changes])
> +ovn_start
> +
> +# Create three switches, each with an ACL so they share the
> +# "has ACLs" lflows in the same datapath group {sw0, sw1, sw2}.
> +check ovn-nbctl ls-add sw0
> +check ovn-nbctl ls-add sw1
> +check ovn-nbctl ls-add sw2
> +check ovn-nbctl lsp-add sw0 sw0-p0
> +check ovn-nbctl lsp-add sw1 sw1-p0
> +check ovn-nbctl lsp-add sw2 sw2-p0
> +check ovn-nbctl acl-add sw0 to-lport 1000 ip4 allow
> +check ovn-nbctl acl-add sw1 to-lport 1000 ip6 allow
> +check ovn-nbctl --wait=sb acl-add sw2 to-lport 1000 tcp allow
> +
> +CHECK_NO_CHANGE_AFTER_RECOMPUTE
> +
> +# Record dp_group state (UUIDs + content).
> +ovn-sbctl --bare --columns=_uuid,datapaths list Logical_DP_Group \
> +    | sort > dpg_before
> +
> +# ----------------------------------------------------------
> +# Case 1: Change that triggers incremental lflow compute but
> +#         does NOT alter the dp-group bitmap.
> +#
> +# Setting a port address rebuilds the port's lflows but the
> +# set of datapaths that share the "has ACLs" lflows is still
> +# {sw0, sw1, sw2}.  The dp_group rows should not change.
> +# ----------------------------------------------------------
> +check as northd ovn-appctl -t ovn-northd inc-engine/clear-stats
> +check ovn-nbctl --wait=sb lsp-set-addresses sw0-p0 "00:00:00:00:00:01"
> +check_engine_stats lflow norecompute compute
> +
> +ovn-sbctl --bare --columns=_uuid,datapaths list Logical_DP_Group \
> +    | sort > dpg_after
> +AT_CHECK([diff dpg_before dpg_after])
> +
> +# ----------------------------------------------------------
> +# Case 2: Change that removes a datapath from a shared flow,
> +#         altering the dp-group bitmap.
> +#
> +# Move sw0's ACL to tier 1.  The egress "acl action" flows
> +# now differ for sw0, so the {sw0, sw1, sw2} group loses sw0
> +# and a {sw1, sw2} group appears (or the existing one is
> +# updated).  The dp_group rows must change.
> +# ----------------------------------------------------------
> +check as northd ovn-appctl -t ovn-northd inc-engine/clear-stats
> +acl0=$(fetch_column nb:ACL _uuid match=ip4)
> +check ovn-nbctl --wait=sb set ACL $acl0 tier=1
> +check_engine_stats lflow norecompute compute
> +
> +ovn-sbctl --bare --columns=_uuid,datapaths list Logical_DP_Group \
> +    | sort > dpg_changed
> +AT_CHECK([diff dpg_before dpg_changed], [1], [ignore])
> +
> +check ovn-nbctl --wait=sb ls-del sw0
> +
> +CHECK_NO_CHANGE_AFTER_RECOMPUTE
> +
> +OVN_CLEANUP_NORTHD
> +AT_CLEANUP
> --
> 2.55.0
>
> _______________________________________________
> dev mailing list
> [email protected]
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>
>
Regards,
Ales
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to