On Fri, Aug 7, 2015 at 3:19 PM, Russell Bryant <rbry...@redhat.com> wrote:
> On 08/07/2015 03:46 AM, Alex Wang wrote: > > This commit adds the binding module to ovn-controller-vtep. The > > module will scan through the Port_Binding table in ovnsb. If there is > > a port binding entry for a logical switch on the vtep gateway chassis's > > "vtep_logical_switches", sets the port binding's chassis column to the > > vtep gateway chassis. > > > > Signed-off-by: Alex Wang <al...@nicira.com> > > I wanted to send the quick message about the test failures, but here are > a few inline comments and a question I wanted to ask ... > > > > > --- > > V4->V5: > > - rebase on top of master. > > - change to use port binding type, and options when finding bindings > > for vtep gateway chassis. > > > > V3->V4: > > - rebase to master. > > > > V2->V3: > > - since ovn-sb schema changes (removal of Gateway table), the binding > > module code needs to be adapted. > > > > PATCH->V2: > > - split into separate commit. > > - disallow and warn if more than one logical port from one 'vlan_map' > > are attached to the same logical datapath. > > --- > > ovn/controller-vtep/automake.mk | 2 + > > ovn/controller-vtep/binding.c | 247 > +++++++++++++++++++++++++++++ > > ovn/controller-vtep/binding.h | 27 ++++ > > ovn/controller-vtep/ovn-controller-vtep.c | 4 + > > tests/ovn-controller-vtep.at | 112 +++++++++++++ > > 5 files changed, 392 insertions(+) > > create mode 100644 ovn/controller-vtep/binding.c > > create mode 100644 ovn/controller-vtep/binding.h > > > > diff --git a/ovn/controller-vtep/automake.mk b/ovn/controller-vtep/ > automake.mk > > index 514cafa..33f063f 100644 > > --- a/ovn/controller-vtep/automake.mk > > +++ b/ovn/controller-vtep/automake.mk > > @@ -1,5 +1,7 @@ > > bin_PROGRAMS += ovn/controller-vtep/ovn-controller-vtep > > ovn_controller_vtep_ovn_controller_vtep_SOURCES = \ > > + ovn/controller-vtep/binding.c \ > > + ovn/controller-vtep/binding.h \ > > ovn/controller-vtep/gateway.c \ > > ovn/controller-vtep/gateway.h \ > > ovn/controller-vtep/ovn-controller-vtep.c \ > > diff --git a/ovn/controller-vtep/binding.c > b/ovn/controller-vtep/binding.c > > new file mode 100644 > > index 0000000..fc12006 > > --- /dev/null > > +++ b/ovn/controller-vtep/binding.c > > @@ -0,0 +1,247 @@ > > +/* Copyright (c) 2015 Nicira, 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 "binding.h" > > + > > +#include "lib/shash.h" > > +#include "lib/smap.h" > > +#include "lib/util.h" > > +#include "openvswitch/vlog.h" > > +#include "ovn-controller-vtep.h" > > +#include "ovn/lib/ovn-sb-idl.h" > > +#include "vtep/vtep-idl.h" > > + > > +VLOG_DEFINE_THIS_MODULE(binding); > > + > > +/* > > + * This module scans through the Port_Binding table in ovnsb. If there > is a > > + * logical port binding entry for logical switch in vtep gateway > chassis's > > + * 'vtep_logical_switches' column, sets the binding's chassis column to > the > > + * corresponding vtep gateway chassis. > > + * > > + */ > > + > > + > > +/* Returns true if the vtep logical switch specified by > 'port_binding_rec' > > + * has already been bound to another port binding entry, and resets > > + * 'port_binding_rec''s chassis column. Otherwise, updates 'ls_to_pb' > > + * and returns false. */ > > +static bool > > +check_pb_conflict(struct shash *ls_to_pb, > > + const struct sbrec_port_binding *port_binding_rec, > > + const struct sbrec_chassis *chassis_rec) > > +{ > > + const char *vtep_lswitch = > > + smap_get(&port_binding_rec->options, "vtep-logical-switch"); > > + const struct sbrec_port_binding *pb_conflict = > > + shash_find_data(ls_to_pb, vtep_lswitch); > > + > > + if (pb_conflict) { > > + VLOG_WARN("logical switch (%s), on vtep gateway chassis " > > + "(%s) has already been associated with logical " > > + "port (%s), ignore logical port (%s)", > > + vtep_lswitch, chassis_rec->name, > > + pb_conflict->logical_port, > > + port_binding_rec->logical_port); > > + sbrec_port_binding_set_chassis(port_binding_rec, NULL); > > + > > + return true; > > + } > > + > > + shash_replace(ls_to_pb, vtep_lswitch, port_binding_rec); > > + return false; > > +} > > + > > +/* Returns true if the vtep logical switch specified by > 'port_binding_rec' > > + * has already been bound to a different datapath, and resets > > + * 'port_binding_rec''s chassis column. Otherwise, updates 'ls_to_db' > and > > + * returns false. */ > > +static bool > > +check_db_conflict(struct shash *ls_to_db, > > + const struct sbrec_port_binding *port_binding_rec, > > + const struct sbrec_chassis *chassis_rec) > > +{ > > + const char *vtep_lswitch = > > + smap_get(&port_binding_rec->options, "vtep-logical-switch"); > > + const struct sbrec_datapath_binding *db_conflict = > > + shash_find_data(ls_to_db, vtep_lswitch); > > + > > + if (db_conflict && db_conflict != port_binding_rec->datapath) { > > + VLOG_WARN("logical switch (%s), on vtep gateway chassis " > > + "(%s) has already been associated with logical " > > + "datapath (with tunnel key %"PRId64"), ignore " > > + "logical port (%s) which belongs to logical " > > + "datapath (with tunnel key %"PRId64")", > > + vtep_lswitch, chassis_rec->name, > > + db_conflict->tunnel_key, > > + port_binding_rec->logical_port, > > + port_binding_rec->datapath->tunnel_key); > > + sbrec_port_binding_set_chassis(port_binding_rec, NULL); > > + > > + return true; > > + } > > + > > + shash_replace(ls_to_db, vtep_lswitch, port_binding_rec->datapath); > > + return false; > > +} > > + > > +/* Updates the 'port_binding_rec''s chassis column to 'chassis_rec'. */ > > +static void > > +update_pb_chassis(const struct sbrec_port_binding *port_binding_rec, > > + const struct sbrec_chassis *chassis_rec) > > +{ > > + if (port_binding_rec->chassis != chassis_rec) { > > + if (chassis_rec && port_binding_rec->chassis) { > > + VLOG_DBG("Changing chassis association of logical " > > + "port (%s) from (%s) to (%s)", > > + port_binding_rec->logical_port, > > + port_binding_rec->chassis->name, > > + chassis_rec->name); > > + } > > + sbrec_port_binding_set_chassis(port_binding_rec, chassis_rec); > > + } > > +} > > + > > + > > +/* Checks and updates logical port to vtep logical switch bindings for > each > > + * physical switch in VTEP. */ > > +void > > +binding_run(struct controller_vtep_ctx *ctx) > > +{ > > + const struct vteprec_physical_switch *pswitch; > > + > > + if (!ctx->ovnsb_idl_txn) { > > + return; > > + } > > + > > + /* 'ls_to_db' > > + * > > + * Maps vtep logical switch name to the datapath binding entry. > This is > > + * used to guarantee that each vtep logical switch is only included > > + * in only one ovn datapath (ovn logical switch). See > check_db_conflict() > > + * for details. > > + * > > + * 'ls_to_pb' > > + * > > + * Maps vtep logical switch name to the port binding entry. This > is used > > + * to guarantee that each vtep logical switch on a vtep physical > switch > > + * is only bound to one logical port. See check_pb_conflict() for > > + * details. > > + * > > + */ > > + struct shash ls_to_db = SHASH_INITIALIZER(&ls_to_db); > > + const struct vteprec_logical_switch *vtep_ls; > > + VTEPREC_LOGICAL_SWITCH_FOR_EACH (vtep_ls, ctx->vtep_idl) { > > + shash_add(&ls_to_db, vtep_ls->name, NULL); > > + } > > + > > + ovsdb_idl_txn_add_comment(ctx->ovnsb_idl_txn, > > + "ovn-controller-vtep: updating bindings"); > > + VTEPREC_PHYSICAL_SWITCH_FOR_EACH (pswitch, ctx->vtep_idl) { > > + const struct sbrec_chassis *chassis_rec > > + = get_chassis_by_name(ctx->ovnsb_idl, pswitch->name); > > I would make sure chassis_rec is not NULL here. It's not supposed to > happen, but just in case ... > Since we do not call ovsdb_idl_run() anywhere in the *_run() functions, and gateway_run() is called before bindnig_run(), this should never happen. How about using ovs_assert(chassis_rec); > > > + struct shash ls_to_pb = SHASH_INITIALIZER(&ls_to_pb); > > + const struct sbrec_port_binding *port_binding_rec; > > + size_t i; > > + > > + for (i = 0; i < chassis_rec->n_vtep_logical_switches; i++) { > > + shash_add(&ls_to_pb, chassis_rec->vtep_logical_switches[i], > > + NULL); > > + } > > + > > + SBREC_PORT_BINDING_FOR_EACH(port_binding_rec, ctx->ovnsb_idl) { > > How often would you have more than 1 Physical_Switch in a vtep db? I > ask because the port binding list could be quite long. Instead of > iterating the whole thing multiple times, you could iterate it once and > build a hash table of vtep port binding records hashed on the physical > switch and logical switch. > > I may be trying to micro-optimize here, though. Anyway, just a thought. > > It should be fairly common to see multiple physical switches in one vtep database... I think I know how to optimize this using shash, > > + const char *type = port_binding_rec->type; > > + const char *vtep_pswitch = > smap_get(&port_binding_rec->options, > > + "vtep-physical-switch"); > > + const char *vtep_lswitch = > smap_get(&port_binding_rec->options, > > + "vtep-logical-switch"); > > + > > + /* Port binding for vtep gateway chassis must have type > "vtep", > > + * and matched physical switch name and logical switch > name. */ > > + if (!strcmp(type, "vtep") > > + && vtep_pswitch && !strcmp(vtep_pswitch, > chassis_rec->name) > > + && vtep_lswitch && shash_find(&ls_to_pb, vtep_lswitch)) > { > > + bool pb_conflict, db_conflict; > > + > > + pb_conflict = check_pb_conflict(&ls_to_pb, > port_binding_rec, > > + chassis_rec); > > + db_conflict = check_db_conflict(&ls_to_db, > port_binding_rec, > > + chassis_rec); > > + /* Updates port binding's chassis column when there > > + * is no conflict. */ > > + if (!pb_conflict && !db_conflict) { > > + update_pb_chassis(port_binding_rec, chassis_rec); > > + } > > + } else if (port_binding_rec->chassis == chassis_rec) { > > + update_pb_chassis(port_binding_rec, NULL); > > + } > > + } > > + struct shash_node *node; > > + SHASH_FOR_EACH (node, &ls_to_pb) { > > + if (!node->data) { > > + VLOG_DBG("No port binding entry for logical switch (%s)" > > + "on vtep gateway chassis (%s)", node->name, > > + chassis_rec->name); > > This could get pretty noisy. I guess it's only a debug message, but > maybe it's worth rate limiting anyway? > > Sure, I'll rate-limit it, > + } > > + } > > + shash_destroy(&ls_to_pb); > > + } > > + shash_destroy(&ls_to_db); > > +} > > + > > +/* Removes all port binding association with vtep gateway chassis. > > + * Returns true when all done. */ > > +bool > > +binding_cleanup(struct controller_vtep_ctx *ctx) > > +{ > > + struct shash ch_to_pb = SHASH_INITIALIZER(&ch_to_pb); > > + const struct sbrec_port_binding *port_binding_rec; > > + > > + if (!ctx->ovnsb_idl_txn) { > > There's no bug here, but I thought there might be at first because > ch_to_pb is initialized but not destroyed. It turns out that doesn't > matter, but my brain just expects them to always match up. :-) > > Maybe you could move this check to the very top of the function? > > Absolutely, adopt your advice, Thanks, Alex Wang, > -- > Russell Bryant > _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
