fix(fleet): Fix flaky test (#30248)

DataDog · Oct 21, 2024 · 70e0419 · 70e0419
1 parent 5144429
commit 70e0419
Show file tree

Hide file tree

Showing 4 changed files with 11 additions and 5 deletions.
diff --git a/flakes.yaml b/flakes.yaml
@@ -11,6 +11,3 @@ test/new-e2e/tests/containers:
  - TestECSSuite/TestCPU/metric___container.cpu.usage{^ecs_container_name:stress-ng$}
  - TestEKSSuite/TestCPU/metric___container.cpu.usage{^kube_deployment:stress-ng$,^kube_namespace:workload-cpustress$}
  - TestKindSuite/TestCPU/metric___container.cpu.usage{^kube_deployment:stress-ng$,^kube_namespace:workload-cpustress$}
-
-test/new-e2e/tests/installer:
- - TestPackages/upgrade_scenario_ubuntu_22_04_x86_64/TestUpgradeSuccessful
diff --git a/test/new-e2e/tests/installer/host/systemd.go b/test/new-e2e/tests/installer/host/systemd.go
@@ -106,7 +106,7 @@ func (h *Host) AssertSystemdEvents(since JournaldTimestamp, events SystemdEventS
  }
  lastSearchedEvents = searchedEvents
  return j == len(events.Events)
- }, 30*time.Second, 1*time.Second)
+ }, 60*time.Second, 1*time.Second)
 
  if !success {
  logs := h.journaldLogsSince(since)

diff --git a/test/new-e2e/tests/installer/unix/all_packages_test.go b/test/new-e2e/tests/installer/unix/all_packages_test.go
@@ -229,6 +229,11 @@ func envForceVersion(pkg, version string) string {
 }
 
 func (s *packageBaseSuite) Purge() {
+ // Reset the systemctl failed counter, best effort as they may not be loaded
+ for _, service := range []string{agentUnit, agentUnitXP, traceUnit, traceUnitXP, processUnit, processUnitXP, probeUnit, probeUnitXP, securityUnit, securityUnitXP} {
+ s.Env().RemoteHost.Execute(fmt.Sprintf("sudo systemctl reset-failed %s", service))
+ }
+
  s.Env().RemoteHost.MustExecute("sudo apt-get remove -y --purge datadog-installer || sudo yum remove -y datadog-installer || sudo zypper remove -y datadog-installer")
 }
 

diff --git a/test/new-e2e/tests/installer/unix/package_agent_test.go b/test/new-e2e/tests/installer/unix/package_agent_test.go
@@ -348,6 +348,10 @@ func (s *packageAgentSuite) TestExperimentStopped() {
  s.host.Run(`sudo systemctl start datadog-agent-exp --no-block`)
 
  // ensure experiment is running
+ s.host.WaitForUnitActive(
+ "datadog-agent-trace-exp.service",
+ "datadog-agent-process-exp.service",
+ )
  s.host.AssertSystemdEvents(timestamp, host.SystemdEvents().Started(traceUnitXP))
  s.host.AssertSystemdEvents(timestamp, host.SystemdEvents().Started(processUnitXP))
  s.host.AssertSystemdEvents(timestamp, host.SystemdEvents().Skipped(securityUnitXP))
@@ -370,7 +374,7 @@ func (s *packageAgentSuite) TestExperimentStopped() {
  Unordered(host.SystemdEvents().
  Started(traceUnit).
  Started(processUnit).
- SkippedIf(probeUnitXP, s.installMethod != InstallMethodAnsible).
+ SkippedIf(probeUnit, s.installMethod != InstallMethodAnsible).
  Skipped(securityUnit),
  ),
  )