From 946e23776be78510799f3f242f7b9f7acbbfd357 Mon Sep 17 00:00:00 2001 From: Stephen J Day Date: Tue, 26 Jul 2016 19:11:17 -0700 Subject: [PATCH] integration-cli: cleanup leader election tests Ensure convergence before moving on with testing leader election conditions. This reduce the flakiness of this test when run in different environments. Signed-off-by: Stephen J Day --- integration-cli/docker_api_swarm_test.go | 60 ++++++++++++++---------- 1 file changed, 36 insertions(+), 24 deletions(-) diff --git a/integration-cli/docker_api_swarm_test.go b/integration-cli/docker_api_swarm_test.go index f2b0cc0faf..481f3f2256 100644 --- a/integration-cli/docker_api_swarm_test.go +++ b/integration-cli/docker_api_swarm_test.go @@ -431,44 +431,56 @@ func (s *DockerSwarmSuite) TestApiSwarmLeaderElection(c *check.C) { c.Assert(d1.getNode(c, d2.NodeID).ManagerStatus.Leader, checker.False) c.Assert(d1.getNode(c, d3.NodeID).ManagerStatus.Leader, checker.False) - leader := d1 + d1.Stop() // stop the leader - // stop the leader - leader.Stop() + var ( + leader *SwarmDaemon // keep track of leader + followers []*SwarmDaemon // keep track of followers + ) + checkLeader := func(nodes ...*SwarmDaemon) checkF { + return func(c *check.C) (interface{}, check.CommentInterface) { + // clear these out before each run + leader = nil + followers = nil + for _, d := range nodes { + if d.getNode(c, d.NodeID).ManagerStatus.Leader { + leader = d + } else { + followers = append(followers, d) + } + } + + if leader == nil { + return false, check.Commentf("no leader elected") + } + + return true, check.Commentf("elected %v", leader.id) + } + } // wait for an election to occur - var newleader *SwarmDaemon - - for _, d := range []*SwarmDaemon{d2, d3} { - if d.getNode(c, d.NodeID).ManagerStatus.Leader { - newleader = d - break - } - } + waitAndAssert(c, defaultReconciliationTimeout, checkLeader(d2, d3), checker.True) // assert that we have a new leader - c.Assert(newleader, checker.NotNil) + c.Assert(leader, checker.NotNil) - // add the old leader back - leader.Start() + // Keep track of the current leader, since we want that to be chosen. + stableleader := leader - // clear leader and reinit the followers list - followers := make([]*SwarmDaemon, 0, 3) + // add the d1, the initial leader, back + d1.Start() + // TODO(stevvooe): may need to wait for rejoin here + + // wait for possible election + waitAndAssert(c, defaultReconciliationTimeout, checkLeader(d1, d2, d3), checker.True) // pick out the leader and the followers again - for _, d := range []*SwarmDaemon{d1, d2, d3} { - if d1.getNode(c, d.NodeID).ManagerStatus.Leader { - leader = d - } else { - followers = append(followers, d) - } - } // verify that we still only have 1 leader and 2 followers c.Assert(leader, checker.NotNil) c.Assert(followers, checker.HasLen, 2) // and that after we added d1 back, the leader hasn't changed - c.Assert(leader.NodeID, checker.Equals, newleader.NodeID) + c.Assert(leader.NodeID, checker.Equals, stableleader.NodeID) } func (s *DockerSwarmSuite) TestApiSwarmRaftQuorum(c *check.C) {