ROX-25477: Exercise integration tests on large nodes (#1763)

A node with large number of CPU cores is an important edge case for
Collector. Run integration tests on such a setup if requested
explicitely or by schedule.

Add a new integration test, which when invoked on a machine with large
number of CPU cores will test the ring buffer adjustment, in particular
that the final adjusted value land in an allowed interval of power of
two.

.github/workflows/integration-tests.yml

@@ -36,6 +36,12 @@ on:
           tests
         type: boolean
         default: false
+      large-box:
+        description: |
+          Marker for workflows that suppose to run tests on nodes with large
+          number of CPU cores.
+        type: boolean
+        default: false
 
 jobs:
   amd64-integration-tests:
@@ -64,6 +70,25 @@ jobs:
       collector-repo: ${{ inputs.collector-repo }}
     secrets: inherit
 
+  amd64-integration-tests-large-box:
+    uses: ./.github/workflows/integration-tests-vm-type.yml
+    if: contains(github.event.pull_request.labels.*.name, 'large-box-integration-tests') || inputs.large-box
+    strategy:
+      # ensure that if one part of the matrix fails, the
+      # rest will continue
+      fail-fast: false
+      matrix:
+        vm_type:
+          - fedora-coreos-large-node
+    with:
+      vm_type: ${{ matrix.vm_type }}
+      collector-tag: ${{ inputs.collector-tag }}
+      collector-qa-tag: ${{ inputs.collector-qa-tag }}
+      collector-tests-tag: ${{ inputs.collector-tests-tag }}
+      job-tag: ${{ inputs.job-tag }}
+      collector-repo: ${{ inputs.collector-repo }}
+    secrets: inherit
+
   arm64-integration-tests:
     uses: ./.github/workflows/integration-tests-vm-type.yml
     if: |

.github/workflows/main.yml

@@ -3,6 +3,7 @@ name: Main collector CI
 on:
   schedule:
     - cron: 0 6 * * *
+    - cron: 0 0 * * 0 # for infrequent tests on a large box
   push:
     branches:
       - master
@@ -96,6 +97,7 @@ jobs:
       collector-tag: ${{ needs.init.outputs.collector-tag }}
       collector-qa-tag: ${{ needs.init.outputs.collector-qa-tag }}
       collector-tests-tag: ${{ needs.build-test-containers.outputs.collector-tests-tag }}
+      large-box: ${{ github.event.schedule == '0 0 * * 0' }}
     if: ${{ !contains(github.event.pull_request.labels.*.name, 'skip-integration-tests') }}
     needs:
     - init

ansible/group_vars/all.yml

@@ -140,6 +140,13 @@ virtual_machines:
     families:
       - fedora-coreos-stable
 
+  fedora-coreos-large-node:
+    project: fedora-coreos-cloud
+    machine_type: n2-standard-128
+    username: core
+    families:
+      - fedora-coreos-stable
+
   # This should be 'fedora-coreos-arm64', but because of
   # GCP VM name length restrictions, we need it to be shorter
   fcarm:

ansible/group_vars/platform_fedora_coreos_large_node.yml

@@ -0,0 +1,4 @@
+---
+ansible_user: core
+
+needs_selinux_permissive: true

integration-tests/integration_test.go

@@ -530,3 +530,7 @@ func TestPerfEvent(t *testing.T) {
 func TestGperftools(t *testing.T) {
 	suite.Run(t, new(suites.GperftoolsTestSuite))
 }
+
+func TestRingBuffer(t *testing.T) {
+	suite.Run(t, new(suites.RingBufferTestSuite))
+}

integration-tests/suites/ringbuf.go

@@ -0,0 +1,31 @@
+package suites
+
+import "github.com/stackrox/collector/integration-tests/pkg/collector"
+
+type RingBufferTestSuite struct {
+	IntegrationTestSuiteBase
+}
+
+func (s *RingBufferTestSuite) SetupSuite() {
+	// Set the totall allowed ring buffer size to 400Mb. On a box with 128 CPU
+	// cores this should trigger an adjustment making the singular ring buffer
+	// size 2Mb (without the power-of-two alignment it would be an invalid
+	// value of 3Mb)
+	collectorOptions := collector.StartupOptions{
+		Env: map[string]string{
+			"ROX_COLLECTOR_SINSP_TOTAL_BUFFER_SIZE": "419430400",
+		},
+	}
+
+	s.StartCollector(false, &collectorOptions)
+}
+
+func (s *RingBufferTestSuite) TearDownSuite() {
+	s.StopCollector()
+}
+
+func (s *RingBufferTestSuite) TestCollectorRunning() {
+	running, err := s.executor.IsContainerRunning("collector")
+	s.Require().NoError(err)
+	s.Require().True(running)
+}