[ovs-dev] [PATCH ovn v10 5/8] ovn-controller: I-P for ct zone and OVS interface changes in flow output stage.

Dumitru Ceara dceara at redhat.com
Tue Jun 9 09:01:02 UTC 2020


On 6/9/20 10:51 AM, Numan Siddique wrote:
> 
> 
> On Mon, Jun 8, 2020 at 11:29 PM Dumitru Ceara <dceara at redhat.com
> <mailto:dceara at redhat.com>> wrote:
> 
>     On 6/8/20 3:50 PM, numans at ovn.org <mailto:numans at ovn.org> wrote:
>     > From: Numan Siddique <numans at ovn.org <mailto:numans at ovn.org>>
>     >
>     > This patch handles ct zone changes and OVS interface changes
>     incrementally
>     > in the flow output stage.
>     >
>     > Any changes to ct zone can be handled by running physical_run()
>     instead of running
>     > flow_output_run(). And any changes to OVS interfaces can be either
>     handled
>     > incrementally (for OVS interfaces representing VIFs) or just running
>     > physical_run() (for tunnel and other types of interfaces).
>     >
>     > To better handle this, a new engine node 'physical_flow_changes'
>     is added which
>     > handles changes to ct zone and OVS interfaces.
>     >
>     > Signed-off-by: Numan Siddique <numans at ovn.org <mailto:numans at ovn.org>>
> 
>     Hi Numan,
> 
>     Overall this patch looks ok to me. I do have a few
>     comments/questions below.
> 
>     Thanks,
>     Dumitru
> 
>     > ---
>     >  controller/binding.c        |  23 +-----
>     >  controller/binding.h        |  24 +++++-
>     >  controller/ovn-controller.c | 144
>     +++++++++++++++++++++++++++++++++++-
>     >  controller/physical.c       |  51 +++++++++++++
>     >  controller/physical.h       |   5 +-
>     >  5 files changed, 222 insertions(+), 25 deletions(-)
>     >
>     > diff --git a/controller/binding.c b/controller/binding.c
>     > index 71063fe9a..c2d9a4c6b 100644
>     > --- a/controller/binding.c
>     > +++ b/controller/binding.c
>     > @@ -503,7 +503,7 @@ remove_local_lport_ids(const struct
>     sbrec_port_binding *binding_rec,
>     >   * 'struct local_binding' is used. A shash of these local bindings is
>     >   * maintained with the 'external_ids:iface-id' as the key to the
>     shash.
>     >   *
>     > - * struct local_binding has 3 main fields:
>     > + * struct local_binding (defined in binding.h) has 3 main fields:
>     >   *    - type
>     >   *    - OVS interface row object
>     >   *    - Port_Binding row object
>     > @@ -554,21 +554,6 @@ remove_local_lport_ids(const struct
>     sbrec_port_binding *binding_rec,
>     >   *   - For each 'virtual' Port Binding (of type BT_VIRTUAL)
>     provided its parent
>     >   *     is bound to this chassis.
>     >   */
>     > -enum local_binding_type {
>     > -    BT_VIF,
>     > -    BT_CONTAINER,
>     > -    BT_VIRTUAL
>     > -};
>     > -
>     > -struct local_binding {
>     > -    char *name;
>     > -    enum local_binding_type type;
>     > -    const struct ovsrec_interface *iface;
>     > -    const struct sbrec_port_binding *pb;
>     > -
>     > -    /* shash of 'struct local_binding' representing children. */
>     > -    struct shash children;
>     > -};
>>     >  static struct local_binding *
>     >  local_binding_create(const char *name, const struct
>     ovsrec_interface *iface,
>     > @@ -590,12 +575,6 @@ local_binding_add(struct shash
>     *local_bindings, struct local_binding *lbinding)
>     >      shash_add(local_bindings, lbinding->name, lbinding);
>     >  }
>>     > -static struct local_binding *
>     > -local_binding_find(struct shash *local_bindings, const char *name)
>     > -{
>     > -    return shash_find_data(local_bindings, name);
>     > -}
>     > -
>     >  static void
>     >  local_binding_destroy(struct local_binding *lbinding)
>     >  {
>     > diff --git a/controller/binding.h b/controller/binding.h
>     > index f7ace6faf..21118ecd4 100644
>     > --- a/controller/binding.h
>     > +++ b/controller/binding.h
>     > @@ -18,6 +18,7 @@
>     >  #define OVN_BINDING_H 1
>>     >  #include <stdbool.h>
>     > +#include "openvswitch/shash.h"
>>     >  struct hmap;
>     >  struct ovsdb_idl;
>     > @@ -32,7 +33,6 @@ struct sbrec_chassis;
>     >  struct sbrec_port_binding_table;
>     >  struct sset;
>     >  struct sbrec_port_binding;
>     > -struct shash;
>>     >  struct binding_ctx_in {
>     >      struct ovsdb_idl_txn *ovnsb_idl_txn;
>     > @@ -60,6 +60,28 @@ struct binding_ctx_out {
>     >      struct smap *local_iface_ids;
>     >  };
>>     > +enum local_binding_type {
>     > +    BT_VIF,
>     > +    BT_CONTAINER,
>     > +    BT_VIRTUAL
>     > +};
>     > +
>     > +struct local_binding {
>     > +    char *name;
>     > +    enum local_binding_type type;
>     > +    const struct ovsrec_interface *iface;
>     > +    const struct sbrec_port_binding *pb;
>     > +
>     > +    /* shash of 'struct local_binding' representing children. */
>     > +    struct shash children;
>     > +};
>     > +
>     > +static inline struct local_binding *
>     > +local_binding_find(struct shash *local_bindings, const char *name)
>     > +{
>     > +    return shash_find_data(local_bindings, name);
>     > +}
>     > +
>     >  void binding_register_ovs_idl(struct ovsdb_idl *);
>     >  void binding_run(struct binding_ctx_in *, struct binding_ctx_out *);
>     >  bool binding_cleanup(struct ovsdb_idl_txn *ovnsb_idl_txn,
>     > diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c
>     > index 687a33c88..795a1c297 100644
>     > --- a/controller/ovn-controller.c
>     > +++ b/controller/ovn-controller.c
>     > @@ -1368,8 +1368,13 @@ static void init_physical_ctx(struct
>     engine_node *node,
>>     >      ovs_assert(br_int && chassis);
>>     > +    struct ovsrec_interface_table *iface_table =
>     > +        (struct ovsrec_interface_table *)EN_OVSDB_GET(
>     > +            engine_get_input("OVS_interface", node));
>     > +
>     >      struct ed_type_ct_zones *ct_zones_data =
>     >          engine_get_input_data("ct_zones", node);
>     > +
>     >      struct simap *ct_zones = &ct_zones_data->current;
>>     >      p_ctx->sbrec_port_binding_by_name = sbrec_port_binding_by_name;
>     > @@ -1377,12 +1382,14 @@ static void init_physical_ctx(struct
>     engine_node *node,
>     >      p_ctx->mc_group_table = multicast_group_table;
>     >      p_ctx->br_int = br_int;
>     >      p_ctx->chassis_table = chassis_table;
>     > +    p_ctx->iface_table = iface_table;
>     >      p_ctx->chassis = chassis;
>     >      p_ctx->active_tunnels = &rt_data->active_tunnels;
>     >      p_ctx->local_datapaths = &rt_data->local_datapaths;
>     >      p_ctx->local_lports = &rt_data->local_lports;
>     >      p_ctx->ct_zones = ct_zones;
>     >      p_ctx->mff_ovn_geneve = ed_mff_ovn_geneve->mff_ovn_geneve;
>     > +    p_ctx->local_bindings = &rt_data->local_bindings;
>     >  }
>>     >  static void init_lflow_ctx(struct engine_node *node,
>     > @@ -1790,6 +1797,124 @@ flow_output_port_groups_handler(struct
>     engine_node *node, void *data)
>     >      return _flow_output_resource_ref_handler(node, data,
>     REF_TYPE_PORTGROUP);
>     >  }
>>     > +/* Engine node en_physical_flow_changes indicates whether
>     > + * there is a need to
>     > + *   - recompute only physical flows or
>     > + *   - we can incrementally process the physical flows.
>     > + *
>     > + * en_physical_flow_changes is an input to flow_output engine node.
>     > + * If the engine node 'en_physical_flow_changes' gets updated during
>     > + * engine run, it means the handler for this -
>     > + * flow_output_physical_flow_changes_handler() will either
>     > + *    - recompute the physical flows by calling 'physical_run() or
>     > + *    - incrementlly process some of the changes for physical flow
>     > + *      calculation. Right now we handle OVS interfaces changes
>     > + *      for physical flow computation.
>     > + *
>     > + * When ever a port binding happens, the follow up
>     > + * activity is the zone id allocation for that port binding.
>     > + * With this intermediate engine node, we avoid full recomputation.
>     > + * Instead we do physical flow computation (either full recomputation
>     > + * by calling physical_run() or handling the changes incrementally.
>     > + *
>     > + * Hence this is an intermediate engine node to indicate the
>     > + * flow_output engine to recomputes/compute the physical flows.
>     > + *
>     > + * TODO 1. Ideally this engine node should recompute/compute the
>     physica
> 
>     Typo: s/physica/physical.
> 
>     > + *      flows instead of relegating it to the flow_output node.
>     > + *      But this requires splitting the flow_output node to
>     > + *      logical_flow_output and physical_flow_output.
>     > + *
>     > + * TODO 2. We can further optimise the en_ct_zone changes to
>     > + *         compute the phsyical flows for changed zone ids.
>     > + *
>     > + * TODO 3: physical.c has a global simap -localvif_to_ofport
>     which stores the
>     > + *         local OVS interfaces and the ofport numbers. Ideally
>     this should be
>     > + *         part of the engine data.
>     > + */
>     > +struct ed_type_pfc_data {
>     > +    bool recompute_physical_flows;
>     > +    bool ovs_ifaces_changed;
>     > +};
>     > +
>     > +static void
>     > +en_physical_flow_changes_clear_tracked_data(void *data_)
>     > +{
>     > +    struct ed_type_pfc_data *data = data_;
>     > +    data->recompute_physical_flows = false;
>     > +    data->ovs_ifaces_changed = false;
>     > +}
>     > +
> 
>     Nothing wrong here but it seems a bit confusing that we have a function
>     called *_clear_tracked_data() which just resets the node's data.
> 
>     This makes me think more about the name we used in patch 4,
>     "clear_tracked_data", and wonder if we need a less specific name for it.
>     Would "reset_data" be more accurate?
> 
> 
> I don't know. reset_data() would give the impression that it will reset
> the whole engine data.
> 

My argument against clear_tracked_data is that
en_physical_flow_changes_clear_tracked_data() doesn't clear any tracked
data. It clears some of the generic data stored in the node.

Otoh my personal interpretation of "reset_data" would be "reset all data
that needs to be reset". But I don't have a very strong opinion about
this I guess.

>  
> 
>     > +static void *
>     > +en_physical_flow_changes_init(struct engine_node *node OVS_UNUSED,
>     > +                              struct engine_arg *arg OVS_UNUSED)
>     > +{
>     > +    struct ed_type_pfc_data *data = xzalloc(sizeof *data);
>     > +    return data;
>     > +}
>     > +
>     > +static void
>     > +en_physical_flow_changes_cleanup(void *data OVS_UNUSED)
>     > +{
>     > +}
>     > +
>     > +/* Indicate to the flow_output engine that we need to recompute
>     physical
>     > + * flows. */
>     > +static void
>     > +en_physical_flow_changes_run(struct engine_node *node, void *data)
>     > +{
>     > +    struct ed_type_pfc_data *pfc_tdata = data;
>     > +    pfc_tdata->recompute_physical_flows = true;
>     > +    engine_set_node_state(node, EN_UPDATED);
>     > +}
>     > +
>     > +/* There are OVS interface changes. Indicate to the flow_output
>     engine
>     > + * to handle these OVS interface changes for physical flow
>     computations. */
>     > +static bool
>     > +physical_flow_changes_ovs_iface_handler(struct engine_node *node,
>     void *data)
>     > +{
>     > +    struct ed_type_pfc_data *pfc_tdata = data;
>     > +    pfc_tdata->ovs_ifaces_changed = true;
>     > +    engine_set_node_state(node, EN_UPDATED);
>     > +    return true;
>     > +}
>     > +
>     > +static bool
>     > +flow_output_physical_flow_changes_handler(struct engine_node
>     *node, void *data)
>     > +{
>     > +    struct ed_type_runtime_data *rt_data =
>     > +        engine_get_input_data("runtime_data", node);
>     > +
>     > +    struct ed_type_flow_output *fo = data;
>     > +    struct physical_ctx p_ctx;
>     > +    init_physical_ctx(node, rt_data, &p_ctx);
>     > +
>     > +    engine_set_node_state(node, EN_UPDATED);
>     > +    struct ed_type_pfc_data *pfc_data =
>     > +        engine_get_input_data("physical_flow_changes", node);
>     > +
>     > +    if (pfc_data->recompute_physical_flows) {
>     > +        /* This indicates that we need to recompute the physical
>     flows. */
>     > +        physical_run(&p_ctx, &fo->flow_table);
>     > +        return true;
>     > +    }
>     > +
>     > +    if (pfc_data->ovs_ifaces_changed) {
>     > +        /* There are OVS interface changes. Try to handle them
>     > +         * incrementally. */
>     > +        return physical_handle_ovs_iface_changes(&p_ctx,
>     &fo->flow_table);
>     > +    }
>     > +
> 
>     Can it happen that we have both pfc_data->ovs_ifaces_changed and
>     pfc_data->recompute_physical_flows true? If so, shouldn't the order of
>     the two if statements above be reversed?
> 
> 
> If both are true, then  recompute_physical_flows() has to take precedence
> because there is no need to call physical_handle_ovs_iface_changes() then.
> As physical_run() is a recompute of physical flows.

Ah got it, thanks for the explanation!

Thanks,
Dumitru

> 
> Thanks
> Numan
> 
> 
>     > +    return true;
>     > +}
>     > +
>     > +static bool
>     > +flow_output_noop_handler(struct engine_node *node OVS_UNUSED,
>     > +                         void *data OVS_UNUSED)
>     > +{
>     > +    return true;
>     > +}
>     > +
> 
>     Please see the comment about noop_handler in patch 2.
> 
>     >  struct ovn_controller_exit_args {
>     >      bool *exiting;
>     >      bool *restart;
>     > @@ -1914,6 +2039,8 @@ main(int argc, char *argv[])
>     >      ENGINE_NODE(runtime_data, "runtime_data");
>     >      ENGINE_NODE(mff_ovn_geneve, "mff_ovn_geneve");
>     >      ENGINE_NODE(ofctrl_is_connected, "ofctrl_is_connected");
>     > +    ENGINE_NODE_WITH_CLEAR_TRACK_DATA(physical_flow_changes,
>     > +                                      "physical_flow_changes");
>     >      ENGINE_NODE(flow_output, "flow_output");
>     >      ENGINE_NODE(addr_sets, "addr_sets");
>     >      ENGINE_NODE(port_groups, "port_groups");
>     > @@ -1933,13 +2060,28 @@ main(int argc, char *argv[])
>     >      engine_add_input(&en_port_groups, &en_sb_port_group,
>     >                       port_groups_sb_port_group_handler);
>>     > +    /* Engine node physical_flow_changes indicates whether
>     > +     * we can recompute only physical flows or we can
>     > +     * incrementally process the physical flows.
>     > +     */
>     > +    engine_add_input(&en_physical_flow_changes, &en_ct_zones,
>     > +                     NULL);
>     > +    engine_add_input(&en_physical_flow_changes, &en_ovs_interface,
>     > +                     physical_flow_changes_ovs_iface_handler);
>     > +
>     >      engine_add_input(&en_flow_output, &en_addr_sets,
>     >                       flow_output_addr_sets_handler);
>     >      engine_add_input(&en_flow_output, &en_port_groups,
>     >                       flow_output_port_groups_handler);
>     >      engine_add_input(&en_flow_output, &en_runtime_data, NULL);
>     > -    engine_add_input(&en_flow_output, &en_ct_zones, NULL);
>     >      engine_add_input(&en_flow_output, &en_mff_ovn_geneve, NULL);
>     > +    engine_add_input(&en_flow_output, &en_physical_flow_changes,
>     > +                     flow_output_physical_flow_changes_handler);
>     > +
>     > +    /* We need this input nodes for only data. Hence the noop
>     handler. */
>     > +    engine_add_input(&en_flow_output, &en_ct_zones,
>     flow_output_noop_handler);
>     > +    engine_add_input(&en_flow_output, &en_ovs_interface,
>     > +                     flow_output_noop_handler);
>>     >      engine_add_input(&en_flow_output, &en_ovs_open_vswitch, NULL);
>     >      engine_add_input(&en_flow_output, &en_ovs_bridge, NULL);
>     > diff --git a/controller/physical.c b/controller/physical.c
>     > index f06313b9d..240ec158e 100644
>     > --- a/controller/physical.c
>     > +++ b/controller/physical.c
>     > @@ -1780,6 +1780,57 @@ physical_run(struct physical_ctx *p_ctx,
>     >      simap_destroy(&new_tunnel_to_ofport);
>     >  }
>>     > +bool
>     > +physical_handle_ovs_iface_changes(struct physical_ctx *p_ctx,
>     > +                                  struct ovn_desired_flow_table
>     *flow_table)
>     > +{
>     > +    const struct ovsrec_interface *iface_rec;
>     > +    OVSREC_INTERFACE_TABLE_FOR_EACH_TRACKED (iface_rec,
>     p_ctx->iface_table) {
>     > +        if (!strcmp(iface_rec->type, "geneve") ||
>     > +            !strcmp(iface_rec->type, "patch")) {
>     > +            return false;
>     > +        }
>     > +    }
>     > +
>     > +    struct ofpbuf ofpacts;
>     > +    ofpbuf_init(&ofpacts, 0);
>     > +
>     > +    OVSREC_INTERFACE_TABLE_FOR_EACH_TRACKED (iface_rec,
>     p_ctx->iface_table) {
>     > +        const char *iface_id = smap_get(&iface_rec->external_ids,
>     "iface-id");
>     > +        if (!iface_id) {
>     > +            continue;
>     > +        }
>     > +
>     > +        const struct local_binding *lb =
>     > +            local_binding_find(p_ctx->local_bindings, iface_id);
>     > +
>     > +        if (!lb || !lb->pb) {
>     > +            continue;
>     > +        }
>     > +
>     > +        int64_t ofport = iface_rec->n_ofport ? *iface_rec->ofport
>     : 0;
>     > +        if (ovsrec_interface_is_deleted(iface_rec)) {
>     > +            ofctrl_remove_flows(flow_table, &lb->pb->header_.uuid);
>     > +            simap_find_and_delete(&localvif_to_ofport, iface_id);
>     > +        } else {
>     > +            if (!ovsrec_interface_is_new(iface_rec)) {
>     > +                ofctrl_remove_flows(flow_table,
>     &lb->pb->header_.uuid);
>     > +            }
>     > +
>     > +            simap_put(&localvif_to_ofport, iface_id, ofport);
>     > +            consider_port_binding(p_ctx->sbrec_port_binding_by_name,
>     > +                                  p_ctx->mff_ovn_geneve,
>     p_ctx->ct_zones,
>     > +                                  p_ctx->active_tunnels,
>     > +                                  p_ctx->local_datapaths,
>     > +                                  lb->pb, p_ctx->chassis,
>     > +                                  flow_table, &ofpacts);
>     > +        }
>     > +    }
>     > +
>     > +    ofpbuf_uninit(&ofpacts);
>     > +    return true;
>     > +}
>     > +
>     >  bool
>     >  get_tunnel_ofport(const char *chassis_name, char *encap_ip,
>     ofp_port_t *ofport)
>     >  {
>     > diff --git a/controller/physical.h b/controller/physical.h
>     > index dadf84ea1..9ca34436a 100644
>     > --- a/controller/physical.h
>     > +++ b/controller/physical.h
>     > @@ -49,11 +49,13 @@ struct physical_ctx {
>     >      const struct ovsrec_bridge *br_int;
>     >      const struct sbrec_chassis_table *chassis_table;
>     >      const struct sbrec_chassis *chassis;
>     > +    const struct ovsrec_interface_table *iface_table;
>     >      const struct sset *active_tunnels;
>     >      struct hmap *local_datapaths;
>     >      struct sset *local_lports;
>     >      const struct simap *ct_zones;
>     >      enum mf_field_id mff_ovn_geneve;
>     > +    struct shash *local_bindings;
>     >  };
>>     >  void physical_register_ovs_idl(struct ovsdb_idl *);
>     > @@ -63,7 +65,8 @@ void physical_handle_port_binding_changes(struct
>     physical_ctx *,
>     >                                            struct
>     ovn_desired_flow_table *);
>     >  void physical_handle_mc_group_changes(struct physical_ctx *,
>     >                                        struct
>     ovn_desired_flow_table *);
>     > -
>     > +bool physical_handle_ovs_iface_changes(struct physical_ctx *,
>     > +                                       struct
>     ovn_desired_flow_table *);
>     >  bool get_tunnel_ofport(const char *chassis_name, char *encap_ip,
>     >                         ofp_port_t *ofport);
>     >  #endif /* controller/physical.h */
>     >
> 
>     _______________________________________________
>     dev mailing list
>     dev at openvswitch.org <mailto:dev at openvswitch.org>
>     https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> 



More information about the dev mailing list