Hi Mansi, On 10/7/24 06:38, Mansi Sharma via dev wrote: > Ping for review. >
Sorry for the delay. > From: Mansi Sharma <[email protected]> > Date: Monday, 30 September 2024 at 12:41 PM > To: [email protected] <[email protected]>, [email protected] > <[email protected]>, [email protected] <[email protected]> > Cc: Mansi Sharma <[email protected]> > Subject: [PATCH ovn v8] ovn-controller: Reserve zones for upcoming ports. > This change will be useful for migration cases, where it can be > used to sync ct-entries before port is up on new chassis, I know it's not the purpose of the patch but I'm curious: how are the conntrack entries synced up between chassis? > resulting in reduced network package drops. > It also fulfills the need of any other service which might need > advance ct-zone reservation in future. > > Signed-off-by: Mansi Sharma <[email protected]> > --- > controller/ct-zone.c | 53 ++++++++++++++- > controller/ct-zone.h | 4 +- > controller/ovn-controller.8.xml | 15 +++- > controller/ovn-controller.c | 3 +- > tests/ovn-controller.at | 117 +++++++++++++++++++++++++++++--- > 5 files changed, 180 insertions(+), 12 deletions(-) > > diff --git a/controller/ct-zone.c b/controller/ct-zone.c > index 77eb16ac9..a47a1e829 100644 > --- a/controller/ct-zone.c > +++ b/controller/ct-zone.c > @@ -183,6 +183,31 @@ ct_zones_update(const struct sset *local_lports, > sset_add(&all_users, local_lport); > } > > + /* Add local_lport name which are supposed to come up on the > + * chassis and might need ct-zone reservation in advance. > + * The data is picked up from ovs_vswitch table. */ > + const struct ovsrec_open_vswitch *cfg; > + cfg = ovsrec_open_vswitch_table_first(ovs_table); > + > + if (cfg) { > + const char *reserve_ct_zone_request_list = smap_get( > + &cfg->external_ids, "reserve_ct_zones"); > + > + if (reserve_ct_zone_request_list) { > + char *dup_reserve = xstrdup(reserve_ct_zone_request_list); > + char *reserve_port; > + char *save_ptr = NULL; > + > + for (reserve_port = strtok_r(dup_reserve, ",", &save_ptr); > + reserve_port != NULL; > + reserve_port = strtok_r(NULL, ",", &save_ptr)) { > + sset_add(&all_users, reserve_port); > + } > + > + free(dup_reserve); > + } > + } > + > /* Local patched datapath (gateway routers) need zones assigned. */ > const struct local_datapath *ld; > HMAP_FOR_EACH (ld, hmap_node, local_datapaths) { > @@ -396,9 +421,33 @@ bool > ct_zone_handle_port_update(struct ct_zone_ctx *ctx, > const struct sbrec_port_binding *pb, > bool updated, int *scan_start, > - int min_ct_zone, int max_ct_zone) > + int min_ct_zone, int max_ct_zone, > + const struct ovsrec_open_vswitch_table *ovs_table) > { > struct shash_node *node = shash_find(&ctx->current, pb->logical_port); > + bool is_reserved = false; > + const struct ovsrec_open_vswitch *cfg; > + cfg = ovsrec_open_vswitch_table_first(ovs_table); > + > + if (cfg) { > + const char *reserve_ct_zone_request_list = smap_get( > + &cfg->external_ids, "reserve_ct_zones"); > + > + if (reserve_ct_zone_request_list) { > + char *dup_reserve = xstrdup(reserve_ct_zone_request_list); > + char *reserve_port; > + char *save_ptr = NULL; > + > + for (reserve_port = strtok_r(dup_reserve, ",", &save_ptr); > + reserve_port != NULL; > + reserve_port = strtok_r(NULL, ",", &save_ptr)) { > + is_reserved = true; > + break; > + } > + > + free(dup_reserve); This function and parsing happens for each port update and there might be multiple updates being processed in a single incremental processing run. I'd suggest we create a new function that builds the set of ports that have reserved zones and call it before all ct_zone_handle_port_update() calls. We can pass the set to ct_zone_handle_port_update() and also to ct_zones_update(). > + } > + } > > if (node) { > struct ct_zone *ct_zone = node->data; > @@ -415,6 +464,8 @@ ct_zone_handle_port_update(struct ct_zone_ctx *ctx, > } > ct_zone_limit_update(ctx, pb->logical_port, > ct_zone_get_pb_limit(pb)); > return true; > + } else if (node && is_reserved) { > + return true; > } else if (node && ct_zone_remove(ctx, node->name)) { > return true; > } > diff --git a/controller/ct-zone.h b/controller/ct-zone.h > index 6df03975c..d54ba861b 100644 > --- a/controller/ct-zone.h > +++ b/controller/ct-zone.h > @@ -85,7 +85,9 @@ bool ct_zone_handle_dp_update(struct ct_zone_ctx *ctx, > bool ct_zone_handle_port_update(struct ct_zone_ctx *ctx, > const struct sbrec_port_binding *pb, > bool updated, int *scan_start, > - int min_ct_zone, int max_ct_zone); > + int min_ct_zone, int max_ct_zone, > + const struct ovsrec_open_vswitch_table > + *ovs_table); Nit: the indentation seems a bit weird to me. I think I'd write this as: bool ct_zone_handle_port_update( struct ct_zone_ctx *ctx, const struct sbrec_port_binding *pb, bool updated, int *scan_start, int min_ct_zone, int max_ct_zone, const struct ovsrec_open_vswitch_table *ovs_table); > uint16_t ct_zone_find_zone(const struct shash *ct_zones, const char *name); > void ct_zones_limits_sync(struct ct_zone_ctx *ctx, > const struct hmap *local_datapaths, > diff --git a/controller/ovn-controller.8.xml b/controller/ovn-controller.8.xml > index faefa77b9..b070b724c 100644 > --- a/controller/ovn-controller.8.xml > +++ b/controller/ovn-controller.8.xml > @@ -626,7 +626,20 @@ > <code>external_ids:ovn-installed-ts</code>. > </p> > </dd> > - </dl> > + > + <dt> > + <code>external-ids:reserve_ct_zones</code> in the <code>Bridge</code> > + table > + </dt> > + > + <dd> > + <p> > + This key represents list of ports which are supposed to come up on > + the chassis, and hence need advance reservation of ct-zones. > + It is comma seprated list of port names. > + </p> > + </dd> > + </dl> > > <h1>OVN Southbound Database Usage</h1> > > diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c > index 6a7cca673..694f2983d 100644 > --- a/controller/ovn-controller.c > +++ b/controller/ovn-controller.c > @@ -2321,7 +2321,8 @@ ct_zones_runtime_data_handler(struct engine_node *node, > void *data) > updated |= ct_zone_handle_port_update(&ct_zones_data->ctx, > t_lport->pb, > port_updated, &scan_start, > - min_ct_zone, max_ct_zone); > + min_ct_zone, max_ct_zone, > + ovs_table); This needs a slight rebase, the current main branch has another invocation of ct_zone_handle_port_update(). > } > } > > diff --git a/tests/ovn-controller.at b/tests/ovn-controller.at > index 74bff9035..c2fd35a9c 100644 > --- a/tests/ovn-controller.at > +++ b/tests/ovn-controller.at > @@ -2533,8 +2533,14 @@ check_ovsdb_zone() { > test $ct_zone -eq $db_zone > } > > +check_duplicates() { > + output=$1 > + AT_CHECK([printf "$output" | tr ' ' '\n' | sort | uniq -d | grep .], [1]) Nit: s/printf/echo/ ? > +} > + > check ovs-vsctl add-port br-int ls0-hv1 -- set Interface ls0-hv1 > external-ids:iface-id=ls0-hv1 > check ovs-vsctl add-port br-int ls0-hv2 -- set Interface ls0-hv2 > external-ids:iface-id=ls0-hv2 > +check ovs-vsctl set Open_vSwitch . > external_ids:reserve_ct_zones=ls0-req-hv3,ls0-req-hv4 > > check ovn-nbctl lr-add lr0 > > @@ -2560,17 +2566,18 @@ echo "$ct_zones" > > port1_zone=$(get_zone_num "$ct_zones" ls0-hv1) > port2_zone=$(get_zone_num "$ct_zones" ls0-hv2) > - > +req_port3_zone=$(get_zone_num "$ct_zones" ls0-req-hv3) > +req_port4_zone=$(get_zone_num "$ct_zones" ls0-req-hv4) > snat_zone=$(get_zone_num "$ct_zones" lr0_snat) > echo "snat_zone is $snat_zone" > > -check test "$port1_zone" -ne "$port2_zone" > -check test "$port2_zone" -ne "$snat_zone" > -check test "$port1_zone" -ne "$snat_zone" > - > OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv1 $port1_zone]) > OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv2 $port2_zone]) > OVS_WAIT_UNTIL([check_ovsdb_zone lr0_snat $snat_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv3 $req_port3_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv4 $req_port4_zone]) > + > +check_duplicates "$ct_zones" > > # Now purposely request an SNAT zone for lr0 that conflicts with a zone > # currently assigned to a logical port > @@ -2585,15 +2592,109 @@ echo "$ct_zones" > port1_zone=$(get_zone_num "$ct_zones" ls0-hv1) > port2_zone=$(get_zone_num "$ct_zones" ls0-hv2) > snat_zone=$(get_zone_num "$ct_zones" lr0_snat) > +req_port3_zone=$(get_zone_num "$ct_zones" ls0-req-hv3) > +req_port4_zone=$(get_zone_num "$ct_zones" ls0-req-hv4) > > check test "$snat_zone" -eq "$snat_req_zone" > -check test "$port1_zone" -ne "$port2_zone" > -check test "$port2_zone" -ne "$snat_zone" > -check test "$port1_zone" -ne "$snat_zone" > + > +check_duplicates "$ct_zones" > > OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv1 $port1_zone]) > OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv2 $port2_zone]) > OVS_WAIT_UNTIL([check_ovsdb_zone lr0_snat $snat_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv3 $req_port3_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv4 $req_port4_zone]) > + > +# Add port named ls0-req-hv3 and check if same zone assigned > +# previously get assigned to it this time as well. > + > +check ovn-nbctl lsp-add ls0 ls0-req-hv3 > +check ovs-vsctl -- add-port br-int hv3-vif3 -- \ > + set interface hv3-vif3 external-ids:iface-id=ls0-req-hv3 > +ct_zones=$(ovn-appctl -t ovn-controller ct-zone-list) > +echo "$ct_zones" Debug print? Can it be removed? This comment applies to multiple instances below. > + > +check ovn-nbctl --wait=hv sync > + > +req_port3_zone_new=$(get_zone_num "$ct_zones" ls0-req-hv3) > +check test "$req_port3_zone" -eq "$req_port3_zone_new" > +check_duplicates "$ct_zones" > + > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv1 $port1_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv2 $port2_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone lr0_snat $snat_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv3 $req_port3_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv4 $req_port4_zone]) > + > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv3 > +check test $? -eq 0 Please use AT_CHECK([ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv3], [0], [ignore]) instead. This comment applies to multiple instances below. > + > +# Checks for two cases after removing entry from ovs_vswitch table - > +# 1. If port is already up, ct-zone should be reserved. > +# 2. If port is not up yet, ct-zone should not be reserved. > + > +check ovs-vsctl remove Open_vSwitch . external_ids reserve_ct_zones > + > +ct_zones=$(ovn-appctl -t ovn-controller ct-zone-list) > +echo "$ct_zones" > + > +req_port3_zone_after_delete=$(get_zone_num "$ct_zones" ls0-req-hv3) > +req_port4_zone_after_delete=$(get_zone_num "$ct_zones" ls0-req-hv4) > + > +check test "$req_port3_zone_new" -eq "$req_port3_zone_after_delete" > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv3 > +check test $? -eq 0 > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv4 > +check test $? -eq 1 > +check test "$req_port4_zone_after_delete" == "" > + > +# Checks for case when a ct-zone is reserved it comes up on that chassis, and > +# gets deleted, but its persisted in ovs_vswitch table, it should persist the > +# same zone throughout. > + > +check ovs-vsctl set Open_vSwitch . external_ids:reserve_ct_zones=ls0-req-hv5 > +check ovn-nbctl lsp-add ls0 ls0-req-hv5 > +check ovs-vsctl -- add-port br-int hv5-vif5 -- \ > + set interface hv5-vif5 external-ids:iface-id=ls0-req-hv5 > +ct_zones=$(ovn-appctl -t ovn-controller ct-zone-list) > +echo "$ct_zones" > +req_port5_zone=$(get_zone_num "$ct_zones" ls0-req-hv5) > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv5 > +check test $? -eq 0 > + > +check ovs-vsctl remove interface hv5-vif5 external_ids iface-id > +ct_zones=$(ovn-appctl -t ovn-controller ct-zone-list) > +echo "$ct_zones" > +req_port5_zone_new=$(get_zone_num "$ct_zones" ls0-req-hv5) > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv5 > +check test $? -eq 0 > +check test "$req_port5_zone" -eq "$req_port5_zone_new" > +check_duplicates "$ct_zones" > + > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv1 $port1_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-hv2 $port2_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone lr0_snat $snat_zone]) > +OVS_WAIT_UNTIL([check_ovsdb_zone ls0-req-hv3 $req_port3_zone]) > + > +check ovs-vsctl set Open_vSwitch . external_ids:reserve_ct_zones=ls0-req-hv6 > +check ovn-nbctl lsp-add ls0 ls0-req-hv6 > +check ovs-vsctl -- add-port br-int hv6-vif6 -- \ > + set interface hv6-vif6 external-ids:iface-id=ls0-req-hv6 > +ct_zones=$(ovn-appctl -t ovn-controller ct-zone-list) > +echo "$ct_zones" > +req_port6_zone=$(get_zone_num "$ct_zones" ls0-req-hv6) > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv6 > +check test $? -eq 0 > + > +ovn-nbctl --wait=hv lsp-del ls0-req-hv6 > +ct_zones=$(ovn-appctl -t ovn-controller ct-zone-list) > +echo "$ct_zones" > +ovs-vsctl get bridge br-int external_ids:ct-zone-ls0-req-hv6 > +check test $? -eq 0 > +req_port6_zone_new=$(get_zone_num "$ct_zones" ls0-req-hv6) > +check_duplicates "$ct_zones" > +check test "$req_port6_zone" -eq "$req_port6_zone_new" > +check_duplicates "$ct_zones" > > # Now create a conflict in the OVSDB and restart ovn-controller. > Regards, Dumitru _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
