On 6/15/22 11:46, Xavier Simonart wrote:
> When VIF ports are claimed on a chassis, SBDB Port_Binding table is updated.
> If the SBDB IDL is still is read-only ("in transaction") when such a update
> is required, the update is not possible and recompute is triggered through
> I+P failure.
>
> This situation can happen:
> - after updating Port_Binding->chassis to SBDB for one port, in a following
> iteration, ovn-controller handles Interface:external_ids:ovn-installed
> (for the same port) while SBDB is still read-only.
> - after updating Port_Binding->chassis to SBDB for one port, in a following
> iteration, ovn-controller updates Port_Binding->chassis for another port,
> while SBDB is still read-only.
>
> This patch prevent the recompute, by having the if-status module
> updating the Port_Binding chassis (if needed) when possible.
> This does not delay Port_Binding chassis update compared to before this patch.
> - With the patch, Port_Binding chassis will be updated as soon as SBDB is
> again writable, without recompute.
> - Without the patch, Port_Binding chassis was updated as soon as SBDB was
> again writable, through a recompute.
>
> Reported-at: https://bugzilla.redhat.com/show_bug.cgi?id=2059253
> Signed-off-by: Xavier Simonart <[email protected]>
>
> ---
Hi Xavier,
Overall I think the approach is good. I do have a question related to
I-P of tracked datapaths/ports below and some more rather minor comments.
> v2: - handled Dumitru's comments.
> - handled Han's comments, mainly ensure we moved out of CLAIMED state
> only after updating pb->chassis to guarentee physical flows are
> installed
> when ovn-installed is updated in OVS.
> - slighly reorganize the code to isolate 'notify_up = false' cases in
> claim_port (i.e. ports such as virtual ports), in the idea of making
> future patch preventing recomputes when virtual ports are claimed.
> - updated test case to cause more race conditions.
> - rebased on origin/main
> - note that "additional chassis" as now supported by
> "Support LSP:options:requested-chassis as a list" might still cause
> recomputes.
> - fixed missing flows when Port_Binding chassis was updated by mgr_update
> w/o any lflow recalculation.
> ---
> controller/binding.c | 154 ++++++++++++++++++++++---------
> controller/binding.h | 15 ++-
> controller/if-status.c | 179 ++++++++++++++++++++++++++++++++----
> controller/if-status.h | 17 +++-
> controller/ovn-controller.c | 121 +++++++++++++++++++++++-
> tests/ovn.at | 113 ++++++++++++++++++++++-
> 6 files changed, 528 insertions(+), 71 deletions(-)
>
> diff --git a/controller/binding.c b/controller/binding.c
> index 2279570f9..b21577f71 100644
> --- a/controller/binding.c
> +++ b/controller/binding.c
> @@ -644,11 +644,17 @@ local_binding_get_lport_ofport(const struct shash
> *local_bindings,
> }
>
> bool
> -local_binding_is_up(struct shash *local_bindings, const char *pb_name)
> +local_binding_is_up(struct shash *local_bindings, const char *pb_name,
> + const struct sbrec_chassis *chassis_rec)
> {
> struct local_binding *lbinding =
> local_binding_find(local_bindings, pb_name);
> struct binding_lport *b_lport =
> local_binding_get_primary_lport(lbinding);
> +
> + if (b_lport && b_lport->pb->chassis != chassis_rec) {
> + return false;
> + }
> +
> if (lbinding && b_lport && lbinding->iface) {
> if (b_lport->pb->n_up && !b_lport->pb->up[0]) {
> return false;
> @@ -660,13 +666,23 @@ local_binding_is_up(struct shash *local_bindings, const
> char *pb_name)
> }
>
> bool
> -local_binding_is_down(struct shash *local_bindings, const char *pb_name)
> +local_binding_is_down(struct shash *local_bindings, const char *pb_name,
> + const struct sbrec_chassis *chassis_rec)
> {
> struct local_binding *lbinding =
> local_binding_find(local_bindings, pb_name);
>
> struct binding_lport *b_lport =
> local_binding_get_primary_lport(lbinding);
>
> + if (b_lport) {
> + if (b_lport->pb->chassis == chassis_rec) {
> + return false;
> + } else if (b_lport->pb->chassis) {
> + VLOG_DBG("lport %s already claimed by other chassis",
> + b_lport->pb->logical_port);
> + }
> + }
> +
> if (!lbinding) {
> return true;
> }
> @@ -884,37 +900,80 @@ get_lport_type_str(enum en_lport_type lport_type)
> OVS_NOT_REACHED();
> }
>
> -/* For newly claimed ports, if 'notify_up' is 'false':
> +void
> +set_pb_chassis_in_sbrec(const struct sbrec_port_binding *pb,
> + const struct sbrec_chassis *chassis_rec,
> + bool is_set)
> +{
> + if (pb->chassis != chassis_rec) {
> + if (is_set) {
> + if (pb->chassis) {
> + VLOG_INFO("Changing chassis for lport %s from %s to %s.",
> + pb->logical_port, pb->chassis->name,
> + chassis_rec->name);
> + } else {
> + VLOG_INFO("Claiming lport %s for this chassis.",
> + pb->logical_port);
> + }
> + for (int i = 0; i < pb->n_mac; i++) {
> + VLOG_INFO("%s: Claiming %s", pb->logical_port, pb->mac[i]);
> + }
> + sbrec_port_binding_set_chassis(pb, chassis_rec);
> + }
> + } else if (!is_set) {
> + sbrec_port_binding_set_chassis(pb, NULL);
> + }
> +}
> +
> +void
> +local_binding_set_pb(struct shash *local_bindings, const char *pb_name,
> + const struct sbrec_chassis *chassis_rec,
> + struct hmap *tracked_datapaths, bool is_set)
> +{
> + struct local_binding *lbinding =
> + local_binding_find(local_bindings, pb_name);
> + struct binding_lport *b_lport =
> local_binding_get_primary_lport(lbinding);
> +
> + if (b_lport) {
> + set_pb_chassis_in_sbrec(b_lport->pb, chassis_rec, is_set);
> + if (tracked_datapaths) {
> + update_lport_tracking(b_lport->pb, tracked_datapaths, true);
> + }
> + }
> +}
> +
> +/* For newly claimed ports:
> * - set the 'pb.up' field to true if 'pb' has no 'parent_pb'.
> * - set the 'pb.up' field to true if 'parent_pb.up' is 'true' (e.g., for
> * container and virtual ports).
> - * Otherwise request a notification to be sent when the OVS flows
> - * corresponding to 'pb' have been installed.
> + *
> + * Returns false if lport is not claimed due to 'sb_readonly'.
> + * Returns true otherwise.
> *
> * Note:
> - * Updates (directly or through a notification) the 'pb->up' field only if
> - * it's explicitly set to 'false'.
> + * Updates the 'pb->up' field only if it's explicitly set to 'false'.
> * This is to ensure compatibility with older versions of ovn-northd.
> */
> -static void
> +static bool
> claimed_lport_set_up(const struct sbrec_port_binding *pb,
> const struct sbrec_port_binding *parent_pb,
> - const struct sbrec_chassis *chassis_rec,
> - bool notify_up, struct if_status_mgr *if_mgr)
> + bool sb_readonly)
> {
> - if (!notify_up) {
> - bool up = true;
> - if (!parent_pb || (parent_pb->n_up && parent_pb->up[0])) {
> + /* When notify_up is false in claim_port(), no state is created
> + * by if_status_mgr. In such cases, return false (i.e. trigger recompute)
> + * if we can't update sb (because it is readonly).
> + */
> + bool up = true;
> + if (!parent_pb || (parent_pb->n_up && parent_pb->up[0])) {
> + if (!sb_readonly) {
> if (pb->n_up) {
> sbrec_port_binding_set_up(pb, &up, 1);
> }
> + } else if (pb->n_up && !pb->up[0]) {
> + return false;
> }
> - return;
> - }
> -
> - if (pb->chassis != chassis_rec || (pb->n_up && !pb->up[0])) {
> - if_status_mgr_claim_iface(if_mgr, pb->logical_port);
> }
> + return true;
> }
>
> typedef void (*set_func)(const struct sbrec_port_binding *pb,
> @@ -1057,37 +1116,35 @@ claim_lport(const struct sbrec_port_binding *pb,
> struct hmap *tracked_datapaths,
> struct if_status_mgr *if_mgr)
> {
> - if (!sb_readonly) {
> - claimed_lport_set_up(pb, parent_pb, chassis_rec, notify_up, if_mgr);
> - }
> -
> enum can_bind can_bind = lport_can_bind_on_this_chassis(chassis_rec, pb);
> bool update_tracked = false;
>
> if (can_bind == CAN_BIND_AS_MAIN) {
> if (pb->chassis != chassis_rec) {
> - if (sb_readonly) {
> - return false;
> - }
> -
> - if (pb->chassis) {
> - VLOG_INFO("Changing chassis for lport %s from %s to %s.",
> - pb->logical_port, pb->chassis->name,
> - chassis_rec->name);
> - } else {
> - VLOG_INFO("Claiming lport %s for this chassis.",
> - pb->logical_port);
> - }
> - for (size_t i = 0; i < pb->n_mac; i++) {
> - VLOG_INFO("%s: Claiming %s", pb->logical_port, pb->mac[i]);
> - }
> -
> - sbrec_port_binding_set_chassis(pb, chassis_rec);
> if (is_additional_chassis(pb, chassis_rec)) {
> + if (sb_readonly) {
> + return false;
> + }
> remove_additional_chassis(pb, chassis_rec);
> }
> update_tracked = true;
> }
> + if (!notify_up) {
> + if (!claimed_lport_set_up(pb, parent_pb, sb_readonly)) {
> + return false;
> + }
> + if (pb->chassis != chassis_rec) {
> + if (sb_readonly) {
> + return false;
> + }
> + set_pb_chassis_in_sbrec(pb, chassis_rec, true);
> + }
> + } else {
> + if ((pb->chassis != chassis_rec) || (pb->n_up && !pb->up[0])) {
> + if_status_mgr_claim_iface(if_mgr, pb, chassis_rec,
> + sb_readonly);
> + }
> + }
> } else if (can_bind == CAN_BIND_AS_ADDITIONAL) {
> if (!is_additional_chassis(pb, chassis_rec)) {
> if (sb_readonly) {
> @@ -1132,7 +1189,8 @@ claim_lport(const struct sbrec_port_binding *pb,
> */
> static bool
> release_lport_main_chassis(const struct sbrec_port_binding *pb,
> - bool sb_readonly)
> + bool sb_readonly,
> + struct if_status_mgr *if_mgr)
> {
> if (pb->encap) {
> if (sb_readonly) {
> @@ -1141,11 +1199,13 @@ release_lport_main_chassis(const struct
> sbrec_port_binding *pb,
> sbrec_port_binding_set_encap(pb, NULL);
> }
>
> + /* If sb readonly, pb->chassis unset through if-status if present. */
> if (pb->chassis) {
> - if (sb_readonly) {
> + if (!sb_readonly) {
> + sbrec_port_binding_set_chassis(pb, NULL);
> + } else if (!if_status_mgr_iface_is_present(if_mgr,
> pb->logical_port)) {
I would really love it if we could get rid of the !notify_up special
case for ports with parents in the future. That will also allow us to
get rid of this check too, right? It can be a future patch though.
> return false;
> }
> - sbrec_port_binding_set_chassis(pb, NULL);
> }
>
> if (pb->virtual_parent) {
> @@ -1155,7 +1215,8 @@ release_lport_main_chassis(const struct
> sbrec_port_binding *pb,
> sbrec_port_binding_set_virtual_parent(pb, NULL);
> }
>
> - VLOG_INFO("Releasing lport %s from this chassis.", pb->logical_port);
> + VLOG_INFO("Releasing lport %s from this chassis (sb_readonly=%d)",
> + pb->logical_port, sb_readonly);
> return true;
> }
>
> @@ -1189,7 +1250,7 @@ release_lport(const struct sbrec_port_binding *pb,
> struct hmap *tracked_datapaths, struct if_status_mgr *if_mgr)
> {
> if (pb->chassis == chassis_rec) {
> - if (!release_lport_main_chassis(pb, sb_readonly)) {
> + if (!release_lport_main_chassis(pb, sb_readonly, if_mgr)) {
> return false;
> }
> } else if (is_additional_chassis(pb, chassis_rec)) {
> @@ -1271,7 +1332,7 @@ consider_vif_lport_(const struct sbrec_port_binding *pb,
> b_lport->lbinding->iface,
> !b_ctx_in->ovnsb_idl_txn,
> !parent_pb, b_ctx_out->tracked_dp_bindings,
> - b_ctx_out->if_mgr)){
> + b_ctx_out->if_mgr)) {
> return false;
> }
>
> @@ -1527,7 +1588,8 @@ consider_localport(const struct sbrec_port_binding *pb,
> enum can_bind can_bind = lport_can_bind_on_this_chassis(
> b_ctx_in->chassis_rec, pb);
> if (can_bind == CAN_BIND_AS_MAIN) {
> - if (!release_lport_main_chassis(pb, !b_ctx_in->ovnsb_idl_txn)) {
> + if (!release_lport_main_chassis(pb, !b_ctx_in->ovnsb_idl_txn,
> + b_ctx_out->if_mgr)) {
> return false;
> }
> } else if (can_bind == CAN_BIND_AS_ADDITIONAL) {
> diff --git a/controller/binding.h b/controller/binding.h
> index 1fed06674..d20659b0b 100644
> --- a/controller/binding.h
> +++ b/controller/binding.h
> @@ -151,8 +151,10 @@ const struct sbrec_port_binding
> *local_binding_get_primary_pb(
> ofp_port_t local_binding_get_lport_ofport(const struct shash *local_bindings,
> const char *pb_name);
>
> -bool local_binding_is_up(struct shash *local_bindings, const char *pb_name);
> -bool local_binding_is_down(struct shash *local_bindings, const char
> *pb_name);
> +bool local_binding_is_up(struct shash *local_bindings, const char *pb_name,
> + const struct sbrec_chassis *);
> +bool local_binding_is_down(struct shash *local_bindings, const char *pb_name,
> + const struct sbrec_chassis *);
> void local_binding_set_up(struct shash *local_bindings, const char *pb_name,
> const struct sbrec_chassis *chassis_rec,
> const char *ts_now_str, bool sb_readonly,
> @@ -160,7 +162,10 @@ void local_binding_set_up(struct shash *local_bindings,
> const char *pb_name,
> void local_binding_set_down(struct shash *local_bindings, const char
> *pb_name,
> const struct sbrec_chassis *chassis_rec,
> bool sb_readonly, bool ovs_readonly);
> -
> +void local_binding_set_pb(struct shash *local_bindings, const char *pb_name,
> + const struct sbrec_chassis *chassis_rec,
> + struct hmap *tracked_datapaths,
> + bool is_set);
> 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,
> @@ -178,6 +183,10 @@ void binding_dump_local_bindings(struct
> local_binding_data *, struct ds *);
> bool is_additional_chassis(const struct sbrec_port_binding *pb,
> const struct sbrec_chassis *chassis_rec);
>
> +void set_pb_chassis_in_sbrec(const struct sbrec_port_binding *pb,
> + const struct sbrec_chassis *chassis_rec,
> + bool is_set);
> +
> /* Corresponds to each Port_Binding.type. */
> enum en_lport_type {
> LP_UNKNOWN,
> diff --git a/controller/if-status.c b/controller/if-status.c
> index ad61844d8..5b0eef859 100644
> --- a/controller/if-status.c
> +++ b/controller/if-status.c
> @@ -24,6 +24,7 @@
> #include "lib/util.h"
> #include "timeval.h"
> #include "openvswitch/vlog.h"
> +#include "lib/ovn-sb-idl.h"
>
> VLOG_DEFINE_THIS_MODULE(if_status);
>
> @@ -53,9 +54,11 @@ VLOG_DEFINE_THIS_MODULE(if_status);
> */
>
> enum if_state {
> - OIF_CLAIMED, /* Newly claimed interface. */
> - OIF_INSTALL_FLOWS, /* Already claimed interface for which flows are still
> - * being installed.
> + OIF_CLAIMED, /* Newly claimed interface. pb->chassis not yet
> updated.
> + */
> + OIF_INSTALL_FLOWS, /* Claimed interface with pb->chassis successfully
> + * updated in SB and for which flows are still being
> + * installed.
> */
> OIF_MARK_UP, /* Interface with flows successfully installed in OVS
> * but not yet marked "up" in the binding module (in
> @@ -78,6 +81,55 @@ static const char *if_state_names[] = {
> [OIF_INSTALLED] = "INSTALLED",
> };
>
> +/*
> + * +----------------------+
> + * +---> | |
> + * | +-> | NULL |
> <--------------------------------------+++-+
> + * | | +----------------------+
> |
> + * | | ^ release_iface | claim_iface
> |
> + * | | | V - sbrec_update_chassis(if sb is rw)
> |
> + * | | +----------------------+
> |
> + * | | | | <----------------------------------------+
> |
> + * | | | CLAIMED | <--------------------------------------+ |
> |
> + * | | +----------------------+ | |
> |
> + * | | | mgr_update(when sb is rw) | |
> |
> + * | | release_iface | - sbrec_update_chassis | |
> |
> + * | | | - request seqno | |
> |
> + * | | V | |
> |
> + * | | +----------------------+ | |
> |
> + * | +-- | | mgr_run(seqno not rcvd) | |
> |
> + * | | INSTALL_FLOWS | - set port down in sb | |
> |
> + * | | | mgr_update() | |
> |
> + * | +----------------------+ - sbrec_update_chassis if needed | |
> |
> + * | | | |
> |
> + * | | mgr_run(seqno rcvd) | |
> |
> + * | | - set port up in sb | |
> |
> + * | release_iface | - set ovn-installed in ovs | |
> |
> + * | V | |
> |
> + * | +----------------------+ | |
> |
> + * | | | mgr_run() | |
> |
> + * +-- | MARK_UP | - set port up in sb | |
> |
> + * | | - set ovn-installed in ovs | |
> |
> + * | | mgr_update() | |
> |
> + * +----------------------+ - sbrec_update_chassis if needed | |
> |
> + * | | |
> |
> + * | mgr_update(rcvd port up / ovn_installed & chassis set) | |
> |
> + * V | |
> |
> + * +----------------------+ | |
> |
> + * | INSTALLED | ------------> claim_iface ---------------+ |
> |
> + * +----------------------+ |
> |
> + * | |
> |
> + * | release_iface |
> |
> + * V |
> |
> + * +----------------------+ |
> |
> + * | | ------------> claim_iface -----------------+
> |
> + * | MARK_DOWN | ------> mgr_update(rcvd port down)
> ----------+
> + * | | mgr_run()
> + * | | - set port down in sb
> + * | | mgr_update()
> + * +----------------------+ - sbrec_update_chassis(NULL)
> + */
> +
Thanks for adding this, it's really useful!
> struct ovs_iface {
> char *id; /* Extracted from OVS external_ids.iface_id. */
> enum if_state state; /* State of the interface in the state machine.
> */
> @@ -85,6 +137,7 @@ struct ovs_iface {
> * be fully programmed in OVS. Only used in
> state
> * OIF_INSTALL_FLOWS.
> */
> + bool chassis_update_required; /* If true, pb->chassis must be updated.
> */
> };
>
> static uint64_t ifaces_usage;
> @@ -158,14 +211,23 @@ if_status_mgr_destroy(struct if_status_mgr *mgr)
> }
>
> void
> -if_status_mgr_claim_iface(struct if_status_mgr *mgr, const char *iface_id)
> +if_status_mgr_claim_iface(struct if_status_mgr *mgr,
> + const struct sbrec_port_binding *pb,
> + const struct sbrec_chassis *chassis_rec,
> + bool sb_readonly)
> {
> + const char *iface_id = pb->logical_port;
> struct ovs_iface *iface = shash_find_data(&mgr->ifaces, iface_id);
>
> if (!iface) {
> iface = ovs_iface_create(mgr, iface_id, OIF_CLAIMED);
> }
> -
> + if (!sb_readonly) {
> + set_pb_chassis_in_sbrec(pb, chassis_rec, true);
> + iface->chassis_update_required = false;
> + } else {
> + iface->chassis_update_required = true;
> + }
> switch (iface->state) {
> case OIF_CLAIMED:
> case OIF_INSTALL_FLOWS:
> @@ -182,6 +244,13 @@ if_status_mgr_claim_iface(struct if_status_mgr *mgr,
> const char *iface_id)
> }
> }
>
> +bool
> +if_status_mgr_iface_is_present(struct if_status_mgr *mgr, const char
> *iface_id)
> +{
> + struct ovs_iface *iface = shash_find_data(&mgr->ifaces, iface_id);
> + return (!!iface);
Nit: I'd rewrite this as:
return !!shash_find_data(&mgr->ifaces, iface_id);
> +}
> +
> void
> if_status_mgr_release_iface(struct if_status_mgr *mgr, const char *iface_id)
> {
> @@ -246,9 +315,43 @@ if_status_mgr_delete_iface(struct if_status_mgr *mgr,
> const char *iface_id)
> }
> }
>
> +bool
> +if_status_handle_claims(struct if_status_mgr *mgr,
> + struct local_binding_data *binding_data,
> + const struct sbrec_chassis *chassis_rec,
> + struct hmap *tracked_datapath,
> + bool sb_readonly)
> +{
> + if (!binding_data) {
> + return false;
> + }
> +
> + struct shash *bindings = &binding_data->bindings;
> + struct hmapx_node *node;
> +
> + bool rc = false;
> + HMAPX_FOR_EACH_SAFE (node, &mgr->ifaces_per_state[OIF_CLAIMED]) {
We don't need the _SAFE here as far as I can tell, right?
> + struct ovs_iface *iface = node->data;
> + if (sb_readonly) {
> + return false;
> + }
> + if (iface->chassis_update_required) {
> + VLOG_INFO("if_status_handle_claims for %s", iface->id);
> + local_binding_set_pb(bindings, iface->id, chassis_rec,
> + tracked_datapath, true);
> + rc = true;
> + }
> + iface->chassis_update_required = false;
> + }
> + return rc;
> +}
> +
> void
> if_status_mgr_update(struct if_status_mgr *mgr,
> - struct local_binding_data *binding_data)
> + struct local_binding_data *binding_data,
> + const struct sbrec_chassis *chassis_rec,
> + struct hmap *tracked_datapath,
> + bool sb_readonly)
> {
> if (!binding_data) {
> return;
> @@ -257,13 +360,25 @@ if_status_mgr_update(struct if_status_mgr *mgr,
> struct shash *bindings = &binding_data->bindings;
> struct hmapx_node *node;
>
> + /* Interfaces in OIF_MARK_UP state have already set their pb->chassis.
> + * However, it might have been reset by another hv.
> + */
> /* Move all interfaces that have been confirmed "up" by the binding
> module,
> * from OIF_MARK_UP to OIF_INSTALLED.
> */
> HMAPX_FOR_EACH_SAFE (node, &mgr->ifaces_per_state[OIF_MARK_UP]) {
> struct ovs_iface *iface = node->data;
>
> - if (local_binding_is_up(bindings, iface->id)) {
> + if (iface->chassis_update_required) {
> + if (!sb_readonly) {
> + iface->chassis_update_required = false;
> + local_binding_set_pb(bindings, iface->id, chassis_rec,
> + tracked_datapath, true);
> + } else {
> + continue;
> + }
> + }
> + if (local_binding_is_up(bindings, iface->id, chassis_rec)) {
> ovs_iface_set_state(mgr, iface, OIF_INSTALLED);
> }
> }
> @@ -274,23 +389,53 @@ if_status_mgr_update(struct if_status_mgr *mgr,
> HMAPX_FOR_EACH_SAFE (node, &mgr->ifaces_per_state[OIF_MARK_DOWN]) {
> struct ovs_iface *iface = node->data;
>
> - if (local_binding_is_down(bindings, iface->id)) {
> + if (!sb_readonly) {
> + local_binding_set_pb(bindings, iface->id, chassis_rec,
> + tracked_datapath, false);
> + }
> + if (local_binding_is_down(bindings, iface->id, chassis_rec)) {
> ovs_iface_destroy(mgr, iface);
> }
> }
>
> - /* Register for a notification about flows being installed in OVS for all
> - * newly claimed interfaces.
> + if (!sb_readonly) {
> + HMAPX_FOR_EACH_SAFE (node,
> &mgr->ifaces_per_state[OIF_INSTALL_FLOWS]) {
> + struct ovs_iface *iface = node->data;
> +
> + if (iface->chassis_update_required) {
> + iface->chassis_update_required = false;
> + local_binding_set_pb(bindings, iface->id, chassis_rec,
> + tracked_datapath, true);
> + }
> + }
> + }
> +
> + /* Update Port_Binding->chassis for newly claimed interfaces
> + * Register for a notification about flows being installed in OVS for all
> + * newly claimed interfaces for which we could update pb->chassis.
> *
> * Move them from OIF_CLAIMED to OIF_INSTALL_FLOWS.
> */
> - bool new_ifaces = false;
> - HMAPX_FOR_EACH_SAFE (node, &mgr->ifaces_per_state[OIF_CLAIMED]) {
> - struct ovs_iface *iface = node->data;
>
> - ovs_iface_set_state(mgr, iface, OIF_INSTALL_FLOWS);
> - iface->install_seqno = mgr->iface_seqno + 1;
> - new_ifaces = true;
> + bool new_ifaces = false;
> + if (!sb_readonly) {
> + HMAPX_FOR_EACH_SAFE (node, &mgr->ifaces_per_state[OIF_CLAIMED]) {
> + struct ovs_iface *iface = node->data;
> + if (iface->chassis_update_required) {
> + local_binding_set_pb(bindings, iface->id, chassis_rec,
> + tracked_datapath, true);
> + }
> + ovs_iface_set_state(mgr, iface, OIF_INSTALL_FLOWS);
> + iface->install_seqno = mgr->iface_seqno + 1;
> + iface->chassis_update_required = false;
> + new_ifaces = true;
> + }
> + } else {
> + HMAPX_FOR_EACH_SAFE (node, &mgr->ifaces_per_state[OIF_CLAIMED]) {
> + struct ovs_iface *iface = node->data;
> + VLOG_INFO("Not updating pb chassis for %s now as sb is readonly",
> + iface->id);
Should we rate limit this?
Nit: indentation.
> + }
> }
>
> /* Request a seqno update when the flows for new interfaces have been
> @@ -403,7 +548,7 @@ if_status_mgr_update_bindings(struct if_status_mgr *mgr,
> struct hmapx_node *node;
>
> /* Notify the binding module to set "down" all bindings that are still
> - * in the process of being installed in OVS, i.e., are not yet
> instsalled.
> + * in the process of being installed in OVS, i.e., are not yet installed.
> */
> HMAPX_FOR_EACH (node, &mgr->ifaces_per_state[OIF_INSTALL_FLOWS]) {
> struct ovs_iface *iface = node->data;
> diff --git a/controller/if-status.h b/controller/if-status.h
> index bb8a3950d..9b1200ff0 100644
> --- a/controller/if-status.h
> +++ b/controller/if-status.h
> @@ -27,15 +27,28 @@ struct if_status_mgr *if_status_mgr_create(void);
> void if_status_mgr_clear(struct if_status_mgr *);
> void if_status_mgr_destroy(struct if_status_mgr *);
>
> -void if_status_mgr_claim_iface(struct if_status_mgr *, const char *iface_id);
> +void if_status_mgr_claim_iface(struct if_status_mgr *,
> + const struct sbrec_port_binding *pb,
> + const struct sbrec_chassis *chassis_rec,
> + bool sb_readonly);
> void if_status_mgr_release_iface(struct if_status_mgr *, const char
> *iface_id);
> void if_status_mgr_delete_iface(struct if_status_mgr *, const char
> *iface_id);
>
> -void if_status_mgr_update(struct if_status_mgr *, struct local_binding_data
> *);
> +void if_status_mgr_update(struct if_status_mgr *, struct local_binding_data
> *,
> + const struct sbrec_chassis *chassis,
> + struct hmap *tracked_dp_bindings,
> + bool sb_readonly);
> void if_status_mgr_run(struct if_status_mgr *mgr, struct local_binding_data
> *,
> const struct sbrec_chassis *,
> bool sb_readonly, bool ovs_readonly);
> void if_status_mgr_get_memory_usage(struct if_status_mgr *mgr,
> struct simap *usage);
> +bool if_status_mgr_iface_is_present(struct if_status_mgr *mgr,
> + const char *iface_id);
> +bool if_status_handle_claims(struct if_status_mgr *mgr,
> + struct local_binding_data *binding_data,
> + const struct sbrec_chassis *chassis_rec,
> + struct hmap *tracked_datapath,
> + bool sb_readonly);
>
> # endif /* controller/if-status.h */
> diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c
> index 2793c8687..1c3cac2c1 100644
> --- a/controller/ovn-controller.c
> +++ b/controller/ovn-controller.c
> @@ -1417,6 +1417,111 @@ en_runtime_data_run(struct engine_node *node, void
> *data)
> engine_set_node_state(node, EN_UPDATED);
> }
>
> +struct ed_type_sb_ro {
> + bool sb_readonly;
> +};
> +
> +static void *
> +en_sb_ro_init(struct engine_node *node OVS_UNUSED,
> + struct engine_arg *arg OVS_UNUSED)
> +{
> + struct ed_type_sb_ro *data = xzalloc(sizeof *data);
> + return data;
> +}
> +
> +static void
> +en_sb_ro_run(struct engine_node *node, void *data)
> +{
> + struct ed_type_sb_ro *sb_ro_data = data;
> + bool sb_readonly = !engine_get_context()->ovnsb_idl_txn;
> + if (sb_ro_data->sb_readonly != sb_readonly) {
> + sb_ro_data->sb_readonly = sb_readonly;
> + if (!sb_ro_data->sb_readonly) {
> + engine_set_node_state(node, EN_UPDATED);
> + }
> + }
> +}
> +
> +static void
> +en_sb_ro_cleanup(void *data OVS_UNUSED)
> +{
> +}
> +
> +static bool
> +pb_claims_sb_ro_handler(struct engine_node *node, void *data OVS_UNUSED)
> +{
> + engine_set_node_state(node, EN_UPDATED);
> + return true;
> +}
> +
> +static void *
> +en_pb_claims_init(struct engine_node *node OVS_UNUSED,
> + struct engine_arg *arg OVS_UNUSED)
> +{
> + return NULL;
> +}
> +
> +static void
> +en_pb_claims_run(struct engine_node *node OVS_UNUSED, void *data_ OVS_UNUSED)
> +{
> +}
> +
> +static void
> +en_pb_claims_cleanup(void *data OVS_UNUSED)
> +{
> +}
Not really a problem of this patch but maybe we can find a way to avoid
having to explcitly define cleanup/run callbacks that don't do much.
Maybe some sensible defaults for engine node callbacks would make sense.
Han, what do you think?
> +
> +static bool
> +pb_claims_runtime_data_handler(struct engine_node *node OVS_UNUSED,
> + void *data OVS_UNUSED)
> +{
> + return true;
> +}
> +
> +static bool
> +lflow_output_runtime_data_handler(struct engine_node *node,
> + void *data);
> +static bool
> +lflow_output_pb_claims_handler(struct engine_node *node, void *data
> OVS_UNUSED)
> +{
> + const struct sbrec_chassis *chassis = NULL;
> +
> + struct ovsrec_open_vswitch_table *ovs_table =
> + (struct ovsrec_open_vswitch_table *)EN_OVSDB_GET(
> + engine_get_input("OVS_open_vswitch", node));
> +
> + const char *chassis_id = get_ovs_chassis_id(ovs_table);
> +
> + struct ovsdb_idl_index *sbrec_chassis_by_name =
> + engine_ovsdb_node_get_index(
> + engine_get_input("SB_chassis", node),
> + "name");
> +
> + if (chassis_id) {
> + chassis = chassis_lookup_by_name(sbrec_chassis_by_name, chassis_id);
> + }
> + if (chassis) {
> + struct ed_type_runtime_data *runtime_data =
> + engine_get_input_data("runtime_data", node);
> + bool sb_readonly = !engine_get_context()->ovnsb_idl_txn;
> + struct controller_engine_ctx *ctrl_ctx =
> + engine_get_context()->client_ctx;
> +
> + if (if_status_handle_claims(ctrl_ctx->if_mgr,
> + &runtime_data->lbinding_data,
> + chassis,
> + &runtime_data->tracked_dp_bindings,
> + sb_readonly)) {
> + struct engine_node *rt_node =
> + engine_get_input("runtime_data", node);
> + if (!engine_node_changed(rt_node)) {
> + lflow_output_runtime_data_handler(node, data);
Instead of explicitly calling another input's handler here I think I
would factor out the logic of the handler into a separate helper
function which we could call from lflow_output_runtime_data_handler()
and here. Based on its return value we could set the node state, if needed.
> + }
> + }
> + }
> + return true;
> +}
> +
> static bool
> runtime_data_ovs_interface_shadow_handler(struct engine_node *node, void
> *data)
> {
> @@ -3438,6 +3543,8 @@ main(int argc, char *argv[])
> stopwatch_create(VIF_PLUG_RUN_STOPWATCH_NAME, SW_MS);
>
> /* Define inc-proc-engine nodes. */
> + ENGINE_NODE(sb_ro, "sb_ro");
> + ENGINE_NODE(pb_claims, "pb_claims");
> ENGINE_NODE_WITH_CLEAR_TRACK_DATA_IS_VALID(ct_zones, "ct_zones");
> ENGINE_NODE_WITH_CLEAR_TRACK_DATA(ovs_interface_shadow,
> "ovs_interface_shadow");
> @@ -3477,6 +3584,13 @@ main(int argc, char *argv[])
> engine_add_input(&en_non_vif_data, &en_ovs_interface,
> non_vif_data_ovs_iface_handler);
>
> + /* This ensures we run after runtime data. */
> + engine_add_input(&en_pb_claims, &en_runtime_data,
> + pb_claims_runtime_data_handler);
This is neat but I think we need to at least add a comment to
lflow_output_pb_claims_handler() to explain why it's ok to access the
runtime_data input there.
Also, we can use engine_noop_handler() instead of defining an explicit one.
> +
> + /* This ensures we always run when sb becomes writable. */
> + engine_add_input(&en_pb_claims, &en_sb_ro, pb_claims_sb_ro_handler);
> +
> /* Note: The order of inputs is important, all OVS interface changes must
> * be handled before any ct_zone changes.
> */
> @@ -3518,6 +3632,8 @@ main(int argc, char *argv[])
> lflow_output_port_groups_handler);
> engine_add_input(&en_lflow_output, &en_runtime_data,
> lflow_output_runtime_data_handler);
> + engine_add_input(&en_lflow_output, &en_pb_claims,
> + lflow_output_pb_claims_handler);
> engine_add_input(&en_lflow_output, &en_non_vif_data,
> NULL);
>
> @@ -3999,7 +4115,10 @@ main(int argc, char *argv[])
> runtime_data ? &runtime_data->lbinding_data : NULL;
> stopwatch_start(IF_STATUS_MGR_UPDATE_STOPWATCH_NAME,
> time_msec());
> - if_status_mgr_update(if_mgr, binding_data);
> + if_status_mgr_update(if_mgr, binding_data, chassis,
> + (runtime_data
> + ?
> &runtime_data->tracked_dp_bindings
> + : NULL), !ovnsb_idl_txn);
I'm not completely sure that this is correct. The engine has already
run so any nodes that depend on runtime_data have already been processed.
But now we might be adding more "tracked changes" to runtime_data, e.g.:
if_status_mgr_update()
-> local_binding_set_pb()
-> update_lport_tracking()
-> tracked_datapath_lport_add(pb, TRACKED_RESOURCE_NEW, ..)
After this point these tracked changes will not be processed anymore and
will be cleared before the next run starts in the next iteration in
en_runtime_data_clear_tracked_data().
The question is: are we ever adding more "tracked changes" to
runtime_data->tracked_dp_bindings here? If the answer is "yes" then
this approach might be problematic. If the answer is "no" then we might
as well avoid passing the runtime_data->tracked_dp_bindings all the way
down and just call local_binding_set_pb() with tracked_datapaths set to
NULL where needed.
> stopwatch_stop(IF_STATUS_MGR_UPDATE_STOPWATCH_NAME,
> time_msec());
>
> diff --git a/tests/ovn.at b/tests/ovn.at
> index 59d51f3e0..e9c061939 100644
> --- a/tests/ovn.at
> +++ b/tests/ovn.at
> @@ -14990,7 +14990,10 @@ AT_CHECK([as hv1 ovs-ofctl dump-flows br-int
> table=65 | grep actions=output:1],
> echo "verifying that lsp0 binding moves when requested-chassis is changed"
>
> ovn-nbctl lsp-set-options lsp0 requested-chassis=hv2
> -OVS_WAIT_UNTIL([test 1 = $(grep -c "Releasing lport lsp0 from this chassis"
> hv1/ovn-controller.log)])
> +
> +# We might see multiple "Releasing lport ...", when sb is read only
> +OVS_WAIT_UNTIL([test 1 -le $(grep -c "Releasing lport lsp0 from this
> chassis" hv1/ovn-controller.log)])
> +
> wait_column "$hv2_uuid" Port_Binding chassis logical_port=lsp0
>
> # (6) Chassis hv2 should add flows and hv1 should not.
> @@ -15036,7 +15039,7 @@ AT_CHECK([as hv1 ovs-ofctl dump-flows br-int table=0
> | grep in_port=1], [0], [ig
> AT_CHECK([as hv1 ovs-ofctl dump-flows br-int table=65 | grep
> actions=output:1], [0], [ignore])
>
> check ovn-nbctl --wait=hv lsp-set-options lsp0
> requested-chassis=non-existant-chassis
> -OVS_WAIT_UNTIL([test 1 = $(grep -c "Releasing lport lsp0 from this chassis"
> hv1/ovn-controller.log)])
> +OVS_WAIT_UNTIL([test 1 -le $(grep -c "Releasing lport lsp0 from this
> chassis" hv1/ovn-controller.log)])
> check ovn-nbctl --wait=hv sync
> wait_column '' Port_Binding chasssi logical_port=lsp0
> AT_CHECK([as hv1 ovs-ofctl dump-flows br-int table=0 | grep in_port=1], [1],
> [])
> @@ -32041,3 +32044,109 @@ AT_CHECK([test $(ovn-sbctl list fdb | grep -c
> "00:00:00:00:10:30") = 0])
> OVN_CLEANUP([hv1])
> AT_CLEANUP
> ])
> +
> +# RUN_OVN_NBCTL() executes list of commands built by the OVN_NBCTL() macro.
> +m4_define([OVN_NBCTL], [
> + command="${command} -- $1"
> +])
> +
> +m4_define([RUN_OVN_NBCTL], [
> + check ovn-nbctl ${command}
> + unset command
> +])
Why can't we factor this out to ovn-macros.at (and remove the version
from perf-northd.at too)?
> +
> +OVN_FOR_EACH_NORTHD([
> +AT_SETUP([recomputes])
As this is a test that is also performance related I think we could add
AT_KEYWORDS([performance]). We could also add that to the tests in
perf-northd.at. But we can do all this as a follow up, I guess, to
allow us to run all performance related tests separately if needed.
> +ovn_start
> +
> +n_hv=4
> +
> +# Add chassis
> +net_add n1
> +for i in $(seq 1 $n_hv); do
> + sim_add hv$i
> + as hv$i
> + check ovs-vsctl add-br br-phys
> + ovs-vsctl set open . external-ids:ovn-bridge-mappings=phys:br-phys
> + ovn_attach n1 br-phys 192.168.0.$i 24 geneve
> +done
> +
> +add_switch_ports() {
> + start_port=$1
> + end_port=$2
> + nb_hv=$3
> + bulk_size=$4
> + for ((i=start_port; i<end_port; )) do
> + start_bulk=$i
> + for hv in $(seq 1 $nb_hv); do
> + end_bulk=$((start_bulk+bulk_size-1))
> + for port in $(seq $start_bulk $end_bulk); do
> + logical_switch_port=lsp${port}
> + OVN_NBCTL(lsp-add ls1 $logical_switch_port)
> + OVN_NBCTL(lsp-set-addresses $logical_switch_port dynamic)
> + done
> + start_bulk=$((end_bulk+1))
> + done
> + RUN_OVN_NBCTL()
> +
> + start_bulk=$i
> + for hv in $(seq 1 $nb_hv); do
> + end_bulk=$((start_bulk+bulk_size-1))
> + for port in $(seq $start_bulk $end_bulk); do
> + logical_switch_port=lsp${port}
> + as hv$hv ovs-vsctl \
> + --no-wait -- add-port br-int vif${port} \
> + -- set Interface vif${port}
> external_ids:iface-id=$logical_switch_port
> + done
> + start_bulk=$((end_bulk+1))
> + done
> + i=$((end_bulk+1))
> + done
> +
> + # Claiming still WIP here but check for recomputes so test can fail
> faster
> + AT_CHECK([test $(ovn-appctl -t ovn-controller coverage/read-counter
> lflow_run) == $lflow_run])
> +}
> +check ovn-nbctl ls-add ls1
> +check ovn-nbctl set Logical_Switch ls1 other_config:subnet=10.1.0.0/16
> +check ovn-nbctl set Logical_Switch ls1 other_config:exclude_ips=10.1.255.254
> +
> +check ovn-nbctl lr-add lr1
> +check ovn-nbctl lsp-add ls1 lsp0 -- set Logical_Switch_Port lsp0 type=router
> options:router-port=lrp0 addresses=dynamic
> +check ovn-nbctl lrp-add lr1 lrp0 "f0:00:00:01:00:01" 10.1.255.254/16
> +check ovn-nbctl lr-nat-add lr1 snat 10.2.0.1 10.1.0.0/16
> +
> +check ovn-nbctl --wait=hv sync
> +lflow_run=$(ovn-appctl -t ovn-controller coverage/read-counter lflow_run)
> +
> +add_switch_ports 1 1000 $n_hv 5
> +
> +wait_for_ports_up
> +check ovn-nbctl --wait=hv sync
> +
> +for i in $(seq 1 $n_hv); do
> + pid=$(cat hv${i}/ovn-controller.pid)
> + u=$((u+$(cat "/proc/$pid/stat" | awk '{print $14}')))
> + s=$((s+$(cat "/proc/$pid/stat" | awk '{print $15}')))
> +done
> +
> +n_pid=$(cat northd/ovn-northd.pid)
> +n_u=$(cat "/proc/$pid/stat" | awk '{print $14}')
> +n_s=$(cat "/proc/$pid/stat" | awk '{print $15}')
> +
> +echo "Total Northd User Time: $n_u"
> +echo "Total Northd System Time: $n_s"
> +echo "Total Controller User Time: $u"
> +echo "Total Controller System Time: $s"
> +
> +lflow_run1=$(ovn-appctl -t ovn-controller coverage/read-counter lflow_run)
> +n_recomputes=`expr $lflow_run1 - $lflow_run`
> +echo "$n_recomputes recomputes"
> +
> +AT_CHECK([test $lflow_run1 == $lflow_run])
> +
> +for i in $(seq 2 $n_hv); do
> + OVN_CLEANUP_SBOX([hv$i])
> +done
> +OVN_CLEANUP([hv1])
> +AT_CLEANUP
> +])
Thanks,
Dumitru
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
