Even if we don't have devices, the sockets/channels must be closed.
This is particularly harmful when onos is restarted to form a cluster
and the openflow channels are left open. On the next
activation it will try to open ports that were already open and it
fails
Change-Id: I768b9db0b71ffba2805fabf631adcb2039af89e4
The preDeactivate is needed to clean up connections when the app is deactivated but needs
to be called also when the component goes down.
If the two,app and component, go down together we should not trigger the cleaning twice.
Change-Id: I66a40297c78f995c5bcd95226efc33fb135732b3
- Use execute instead of submit so that uncaught Exception will
be dealt and logged by ExecutorService's handler.
- Use component's own logger
Change-Id: I761264aea00748980929b5048e111756776dd2f6
- Using osgi_jar for all modules except apps
- Adding target for installation of core modules to .m2 repo
Change-Id: I65d9e3a9567918e275191724842d7d9b1f485128
Control message monitoring brings some overhead to controller.
In an extreme stressing environment (e.g., running Cbench),
it leads potential performance degradation.
This commit tries to mitigate the Cbench regression with two steps:
1. improve the monitoring performance by assigning more # of
threads in each thread group.
2. make the control message listening feature optional.
Change-Id: I4f7361b7c598c6de71d390eab78a20ada381d4dd
This commit bumps up all maven plugins and dependencies to their
latest version. With bumped version onos passes stc test.
Following plugins are not ready to be bumped up due to instability.
check-style 2.16 -> 2.17: compilation error
maven-compiler-plugin 2.5.1 -> 3.3: always perform recompilation
Following dependencies are not ready to be bumped up due to instability.
kryo 3.0.0 -> 3.0.3: unit test failure
slf4j 1.7.6 -> 1.7.13: fails in stc
Change-Id: Iffe8b0734c7088228ebeee1ef597f0df3dfcd63d
- Forwarding behavior added to {Device,Link}Store by ONOS-490
cauesed false update information sent from ONOS node, which has been detached from the cluster,
to be accepted by rest of the cluster after the detached node has rejoined cluster.
- Fix for periodic mastership check was left out
when MastershipService#requestRoleFor(..) return value was changed to Future.
- Fix for triggerProbe() related messages getting dropped,
right after STANDBY -> MASTER role change.
- Local state (connectedDevices) was preventing
vertical (Core -> switch) Mastership state synchronization.
- Various debug log, comment added during investigation.
Change-Id: I777beadf04db8a879830a07bfdc7ab0e2279f190
