OSGi requires attributes such as xconnectService to be declared volatile
to handle a dynamic reference.
Since xconnectService is now declared volatile and Karaf runtime is able
to set/unset its value, there's no longer need to use event-based
binding/unbinding with dedicated methods.
Change-Id: I5ab64fbf2bd5a76f5a14c6845127246eb7352a8f
- trying rehash of the routes for each link up
- redoing reroute only for the routes where rehash fails
Change-Id: I7495277af73d8948300f170fa92cbbfecc338d89
Trigger the driver to remove the double tagged rules (fwd classifier) when removing the flow objective and no other host is connected to the same port
Change-Id: I83d50f9b3404f2eaa733cc0c75f76bfe3562b5e8
(cherry picked from commit 7c4534ce5133538e2a6f0c95b4ecc7ea1f250e47)
Patch contains several bugfixes and improvements:
- Fixes sid retrieval when the destination leaf is down
- Fixes sid retrieval when ping goes through the spine
- Fixes MPLS deserializer
- Improves Ethernet toString
- Fixes ping to looback for dh host when bond sends to wrong leaf
Change-Id: I05963e74b2976e526826ffd377cadeb462ba0a8d
NOTE: This patch will break double tag termination support on OF-DPA.
The pipelier needs to be re-implemented to understand the new objectives.
Before:
NextObj: ETH_DST, ETH_SRC, OUTPUT, VLAN_ID (dummy)
FwdObj.EGRESS: OUTPUT, VLAN_ID (c-tag), PUSH_VLAN, VLAN_ID (s-tag)
After:
NextObj: ETH_DST, ETH_SRC, OUTPUT, VLAN_ID (c-tag), PUSH_VLAN, VLAN_ID (s-tag)
No FwdObj.EGRESS
Also remove NextObj when the host is removed
Change-Id: I4ccdfa1d20701d9b2451ea0f3b4e761006746120
filtering objectives to a device that lacks mac configuration.
However, according to Opdpa2Pipeline, device mac is not mandatory to
create proper multicast filtering flows.
It seems device mac configuration should also not be mandated by
SR for multicast filtering flows.
Change-Id: I73c381a61bedaaf99cf0951d1e939048065843d2
(cherry picked from commit 92920611c0c6cac0eb64587ee0a4bc17c253a120)
In addition,
- Deprecated old XConnectHandler
- Fixed equals() of XconnectDesc and XconnectKey
- Implemented unit tests for XconnectCodec
Change-Id: I1b5f2c1f389523e3b65a3d9acaf75bc06e39fdda
- when double tagged filtering objective an inner vlan criteria is submitted
- update to ofdpa and ovsofdpa drivers to evaluate the inner vlan criteria correctly
Change-Id: I33170c9b83482a5f26f13f7098a9b24a92da9544
Note: popVlan is effective to L2 interface group that gets created while creating L2 multicast group.
The popVlan has nothing to do with L2 multicast group itself.
Change-Id: I63baf88bfd626b19dcfdf654e4e184e48886f7fc
not the sinks that are not tied with any host.
However, one of addSinks method of MulticastRouteService of mcast
application allows adding connect points each of which is a standalone
sink. In a scenario where all connect points are used as different sinks,
processSinksToBeAdded method in McastHandler service does not return
a true list of sinks to be added;
In overall,all connect points which are associated with HostId.NONE must
be added as different sinks. In the current situation, just one of them
is selected and added.
Change-Id: I3d4eb1ff865aca4f2fe85011a6a390374028a081
In addition,
- Added a fix for host remove issue
- Added a fix for single homed (loc1) to dual homed (loc2).
Change-Id: I1a7344bd77f73d6bed8955dcdbf407b5354d0eee
FPM, STATIC and configured subnets go to high priority batch, while others (e.g. DHCP) go to low priority batch
Change-Id: I8dcef67945c31bd0eab869510bea0f1f278b2925
We should not check subnets learnt from route service. It is not only expensive but also incorrect.
This patch should significantly reduce the CPU usage when controller needs to reply ping to the gateway IP.
Change-Id: Ic245f5f136cf547cb50cabff4927da00e7f03666
Create a new class XconnecEndpoint to cover both physical port and load balancer
Also change the CLI load balancer identifier to "LB:"
Change-Id: I0b4cd6e474d8b21468d87fcadd9280fdf7d6aafa
- Listens for L2LbEvent
- Leverages a cache to wait l2lb nextid
Includes a small change in the L2LbService interface
Change-Id: Id545e2228dde013620b4b415d1d2619027ef15d2
When the next hop moves from [1A,1B] to [1A],
there should be a route on 1B pointing to 1A via spines
Change-Id: I817414fb4e9edf29357fdb5e55675537ff5f0cac
(cherry picked from commit 53eae194a8b66287855483359309597e8df2efa9)
When a host moves from [1A/1, 1B/1] to [1A/2, 1B/1], there is no need to program the pair link flow first.
Also update unit tests to verify this behavior.
Change-Id: I97c7ee5f0db891226c7ed477f5f05af5d6c1faa8
- virtual networking moved to /apps/virtual; with CLI & REST API
- tunnels and labels moved to /apps/tunnel; with CLI & REST API; UI disabled for now
- protobuf/models moved to /core/protobuf/models
- defunct grpc/rpc registry stuff left under /graveyard
- compile dependencies on /incubator moved to respective modules for compilation
- run-time dependencies will need to be re-tested for dependent apps
- /graveyard will be removed in not-too-distant future
Change-Id: I0a0b995c635487edcf95a352f50dd162186b0b39
