[ovs-dev] [PATCH ovn v9 3/5] ovn-controller: Handle datapath changes incrementally for ct zone I-P engine node.

Numan Siddique numans at ovn.org
Mon Jun 14 11:14:50 UTC 2021


On Mon, Jun 14, 2021 at 3:28 AM Han Zhou <hzhou at ovn.org> wrote:
>
> On Thu, Jun 3, 2021 at 5:29 AM <numans at ovn.org> wrote:
> >
> > From: Numan Siddique <numans at ovn.org>
> >
> > After the commit [1], we do a full recompute of ct zone I-P engine for
> > any datapath binding change.  This results in physical_flow() getting
> > called.  In a highly scaled environment this can result in costly CPU
> > cycles.  This patch addressed this by handling the datapath binding
> > changes incrementally in the ct_zone engine node.  ct zone recomputation
> > is required only when a datapath binding is deleted or a logical router
> > datapath binding changes the snat-ct-zone option value.  This patch
> > handles this.
> >
> > [1] - f9cab11d5fab("Allow explicit setting of the SNAT zone on a gateway
> router.")
> >
> > Signed-off-by: Numan Siddique <numans at ovn.org>
> > ---
> >  controller/ovn-controller.c | 55 ++++++++++++++++++++++++++++++++++++-
> >  controller/physical.c       |  5 ++++
> >  tests/ovn-performance.at    | 26 ++++++++++++++++++
> >  3 files changed, 85 insertions(+), 1 deletion(-)
> >
> > diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c
> > index b49786441..5b55a45b7 100644
> > --- a/controller/ovn-controller.c
> > +++ b/controller/ovn-controller.c
> > @@ -1734,6 +1734,58 @@ en_ct_zones_run(struct engine_node *node, void
> *data)
> >      engine_set_node_state(node, EN_UPDATED);
> >  }
> >
> > +/* Handles datapath binding changes for the ct_zones engine.
> > + * Returns false if the datapath is deleted or if the requested snat
> > + * ct zone doesn't match with the ct_zones data. */
> > +static bool
> > +ct_zones_datapath_binding_handler(struct engine_node *node, void *data)
> > +{
> > +    struct ed_type_ct_zones *ct_zones_data = data;
> > +    const struct sbrec_datapath_binding *dp;
> > +    struct ed_type_runtime_data *rt_data =
> > +        engine_get_input_data("runtime_data", node);
> > +    struct sbrec_datapath_binding_table *dp_table =
> > +        (struct sbrec_datapath_binding_table *)EN_OVSDB_GET(
> > +            engine_get_input("SB_datapath_binding", node));
> > +
> > +    SBREC_DATAPATH_BINDING_TABLE_FOR_EACH_TRACKED (dp, dp_table) {
> > +        if (!get_local_datapath(&rt_data->local_datapaths,
> > +                                dp->tunnel_key)) {
> > +            continue;
> > +        }
> > +
> > +        if (sbrec_datapath_binding_is_deleted(dp)) {
> > +            /* Fall back to full recompute of ct_zones engine. */
> > +            return false;
> > +        }
>
> What if a datapath_binding is created? Should we return false as well?

Hi Han,

Thanks for the reviews.  I don't think there is a need to trigger a recompute of
ct_zones when a new datapath is created.

In the patch 0 you mentioned - " I understand it may not affect
correctness for now since new datapath always triggers recompute, but I
think it's better to be explicit that the ct_zones node cannot handle it
incrementally."

But  we don't trigger a full recompute of either runtime_data engine
or flow_output_engine
when a new datapath is created.  Probably you meant recompute the
ct_zone engine node ?

Datapath handler for the runtime data
(runtime_data_sb_datapath_binding_handler()),
returns false only if a datapath is present in the local_datapaths and
it is deleted.

When a new datapath is added to local_datapaths, there will be a full
recompute of
ct_zones since the runtime data handler is NULL.  With your suggestion we will
trigger a recompute of ct_zones when a datapath is created, but not added to the
local_datapaths.

I think there is no need to trigger recompute.  What do you think ?

Thanks
Numan

>
> > +
> > +        int req_snat_zone = datapath_snat_ct_zone(dp);
> > +        if (req_snat_zone == -1) {
> > +            /* datapath snat ct zone is not set.  This condition will
> also hit
> > +             * when CMS clears the snat-ct-zone for the logical router.
> > +             * In this case there is no harm in using the previosly
> specified
> > +             * snat ct zone for this datapath.  Also it is hard to know
> > +             * if this option was cleared or if this option is never
> set. */
> > +            continue;
> > +        }
> > +
> > +        /* Check if the requested snat zone has changed for the datapath
> > +         * or not.  If so, then fall back to full recompute of
> > +         * ct_zone engine. */
> > +        char *snat_dp_zone_key = alloc_nat_zone_key(&dp->header_.uuid,
> "snat");
> > +        struct simap_node *simap_node =
> simap_find(&ct_zones_data->current,
> > +                                                   snat_dp_zone_key);
> > +        free(snat_dp_zone_key);
> > +        if (!simap_node || simap_node->data != req_snat_zone) {
> > +            /* There is no entry yet or the requested snat zone has
> changed.
> > +             * Trigger full recompute of ct_zones engine. */
> > +            return false;
> > +        }
> > +    }
> > +
> > +    return true;
> > +}
> > +
> >  /* The data in the ct_zones node is always valid (i.e., no stale
> pointers). */
> >  static bool
> >  en_ct_zones_is_valid(struct engine_node *node OVS_UNUSED)
> > @@ -2758,7 +2810,8 @@ main(int argc, char *argv[])
> >
> >      engine_add_input(&en_ct_zones, &en_ovs_open_vswitch, NULL);
> >      engine_add_input(&en_ct_zones, &en_ovs_bridge, NULL);
> > -    engine_add_input(&en_ct_zones, &en_sb_datapath_binding, NULL);
> > +    engine_add_input(&en_ct_zones, &en_sb_datapath_binding,
> > +                     ct_zones_datapath_binding_handler);
> >      engine_add_input(&en_ct_zones, &en_runtime_data, NULL);
> >
> >      engine_add_input(&en_runtime_data, &en_ofctrl_is_connected, NULL);
> > diff --git a/controller/physical.c b/controller/physical.c
> > index 04259d44a..e70efc71d 100644
> > --- a/controller/physical.c
> > +++ b/controller/physical.c
> > @@ -15,6 +15,7 @@
> >
> >  #include <config.h>
> >  #include "binding.h"
> > +#include "coverage.h"
> >  #include "byte-order.h"
> >  #include "encaps.h"
> >  #include "flow.h"
> > @@ -48,6 +49,8 @@
> >
> >  VLOG_DEFINE_THIS_MODULE(physical);
> >
> > +COVERAGE_DEFINE(physical_run);
> > +
> >  /* Datapath zone IDs for connection tracking and NAT */
> >  struct zone_ids {
> >      int ct;                     /* MFF_LOG_CT_ZONE. */
> > @@ -1528,6 +1531,8 @@ void
> >  physical_run(struct physical_ctx *p_ctx,
> >               struct ovn_desired_flow_table *flow_table)
> >  {
> > +    COVERAGE_INC(physical_run);
> > +
> >      if (!hc_uuid) {
> >          hc_uuid = xmalloc(sizeof(struct uuid));
> >          uuid_generate(hc_uuid);
> > diff --git a/tests/ovn-performance.at b/tests/ovn-performance.at
> > index e510c6cef..61356e16d 100644
> > --- a/tests/ovn-performance.at
> > +++ b/tests/ovn-performance.at
> > @@ -462,6 +462,32 @@ OVN_CONTROLLER_EXPECT_HIT_COND(
> >      [ovn-nbctl --wait=hv lrp-set-gateway-chassis lr1-public hv3 30 &&
> ovn-nbctl --wait=hv sync]
> >  )
> >
> > +# Test physical_run for logical router and other datapath binding
> changes.
> > +OVN_CONTROLLER_EXPECT_HIT_COND(
> > +    [hv1 hv2 hv3 hv4 hv5], [physical_run], [>0 >0 >0 =0 =0],
> > +    [ovn-nbctl --wait=hv set logical_router lr1 options:snat-ct-zone=10
> && ovn-nbctl --wait=hv sync]
> > +)
> > +
> > +OVN_CONTROLLER_EXPECT_HIT_COND(
> > +    [hv1 hv2 hv3 hv4 hv5], [physical_run], [>0 >0 >0 =0 =0],
> > +    [ovn-nbctl --wait=hv set logical_router lr1 options:snat-ct-zone=11
> && ovn-nbctl --wait=hv sync]
> > +)
> > +
> > +OVN_CONTROLLER_EXPECT_NO_HIT(
> > +    [hv1 hv2 hv3 hv4 hv5], [physical_run],
> > +    [ovn-nbctl --wait=hv remove logical_router lr1 options snat-ct-zone
> && ovn-nbctl --wait=hv sync]
> > +)
> > +
> > +OVN_CONTROLLER_EXPECT_NO_HIT(
> > +    [hv1 hv2 hv3 hv4 hv5], [physical_run],
> > +    [ovn-sbctl set datapath_binding lr1 external_ids:foo=bar &&
> ovn-nbctl --wait=hv sync]
> > +)
> > +
> > +OVN_CONTROLLER_EXPECT_NO_HIT(
> > +    [hv1 hv2 hv3 hv4 hv5], [physical_run],
> > +    [ovn-sbctl set datapath_binding ls1 external_ids:foo=bar &&
> ovn-nbctl --wait=hv sync]
> > +)
> > +
> >  # After this, BFD should be enabled from hv1 and hv2 to hv3.
> >  # So there should be lflow_run hits in hv1, hv2, hv3 and hv4
> >  OVN_CONTROLLER_EXPECT_HIT_COND(
> > --
> > 2.31.1
> >
> > _______________________________________________
> > dev mailing list
> > dev at openvswitch.org
> > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> _______________________________________________
> dev mailing list
> dev at openvswitch.org
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>


More information about the dev mailing list