[SFC-149] Race condition in test SfcIpv4PacketInHandlerTest.pktInPurgeBuffering

Fri, 24 Jun 2016 11:52:19 +0000

There is a road condition in the mentioned test pckInPurgeBuffering that is making the test unstable (e.g. of one failed execution at https://jenkins.opendaylight.org/releng/job/sfc-verify-boron/jdk=openjdk8,nodes=dynamic_verify/166/console):

Tests run: 5, Failures: 1, Errors: 0, Skipped: 0, Time elapsed: 0.054 sec <<< FAILURE! - in org.opendaylight.sfc.ofrenderer.SfcIpv4PacketInHandlerTest
pktInPurgeBuffering(org.opendaylight.sfc.ofrenderer.SfcIpv4PacketInHandlerTest) Time elapsed: 0.02 sec <<< FAILURE!
org.mockito.exceptions.verification.WantedButNotInvoked:
Wanted but not invoked:
sfcOfFlowProgrammerImpl.setFlowRspId(<any>);
-> at org.opendaylight.sfc.ofrenderer.SfcIpv4PacketInHandlerTest.pktInPurgeBuffering(SfcIpv4PacketInHandlerTest.java:138)

This test verifies purging after timeout in the packetIn handler in charge of creating temporary rules in the PathMapperACL table. The problem: in the test, 1ms timeout is setted in the handler:

this.pktInHandler.setMaxBufferTime(1); // 1 microsecond

The comment is incorrect: time unit is not usecs, but millis (in packetInHandler, that value is compared vs System.currentTimeMillis for determining the purge):

if((currentMillis - bufferedTime.longValue()) > maxBufferTime)

Later in the test:

// When called again, the purgeCount will be exceeded, the buffer will
// be flushed, and this packet will be added back, making a size of 1
Thread.sleep(1); // sleep 1 millisecond, to let the buffer time expire

As stated, waiting exactly 1 milisecond (== exactly the timeout time) is the road condition. The test should be corrected in order to use a value higher enough (i.e. 10 millis) than the timeout so the buffer time is always exceeded. The incorrect "microseconds" comment should also be corrected

OpenDaylight JIRA

[SFC-149] Race condition in test SfcIpv4PacketInHandlerTest.pktInPurgeBuffering