fix(e2e): webhook-driven, auto-versioned CMT + fix non-PR dispatch 422

config/config-matterwick.default.json

@@ -83,5 +83,7 @@
   "E2EReleasePatternPrefix": "release-",
   "E2ENightlyTriggerWorkflowName": "E2E Nightly Trigger",
   "E2ETestWorkflowNames": ["Electron Playwright Tests", "E2E", "Compatibility Matrix Testing"],
-  "E2EInstanceMaxAge": 6
+  "E2EInstanceMaxAge": 6,
+  "CMTTriggerWorkflowName": "CMT Provisioner",
+  "CMTServerVersions": ["10.11.0", "11.7.0"]
 }

server/config.go

@@ -123,6 +123,32 @@ type MatterwickConfig struct {
 	// Set to the longest expected E2E run duration plus a small buffer.
 	// Default (0): 3 hours.
 	E2EInstanceMaxAge int
+
+	// CMTTriggerWorkflowName is the workflow name (the "name:" field) of the lightweight,
+	// scheduled CMT trigger workflow in the desktop/mobile repos. When matterwick receives
+	// a workflow_run "requested" event for this workflow it provisions one instance per
+	// version in CMTServerVersions and dispatches compatibility-matrix-testing.yml.
+	CMTTriggerWorkflowName string
+
+	// CMTServerVersions is the hardcoded set of Mattermost server versions CMT runs
+	// against (e.g. the active ESR plus the current feature release). Values must be valid
+	// Mattermost image tags (full semver, no "v" prefix, e.g. "10.11.0"). Overridable via
+	// the gitops config; when empty matterwick falls back to defaultCMTServerVersions.
+	CMTServerVersions []string
+}
+
+// defaultCMTServerVersions is the fallback CMT version set used when Config.CMTServerVersions
+// is empty. Mattermost actively supports v11.x feature releases and the v10.11 ESR, so the
+// default covers the current ESR line plus a v11 release. Update as ESR/feature lines change.
+var defaultCMTServerVersions = []string{"10.11.0", "11.7.0"}
+
+// CMTVersions returns the configured CMT server versions, or the hardcoded default set when
+// none are configured.
+func (c *MatterwickConfig) CMTVersions() []string {
+	if len(c.CMTServerVersions) > 0 {
+		return c.CMTServerVersions
+	}
+	return defaultCMTServerVersions
 }
 
 func findConfigFile(fileName string) string {

server/e2e_dryrun_test.go

@@ -809,6 +809,39 @@ func TestDryRun_InstanceTracking(t *testing.T) {
 	})
 }
 
+// ------------------------------------------------------------
+// 9b. SHA-scoped cleanup must not reap a concurrent flow on the same SHA
+// ------------------------------------------------------------
+
+func TestInstanceKeyMatchesSHA(t *testing.T) {
+	repo := "mattermost-mobile"
+	sha := "deadbeef"
+	cmtKey := fmt.Sprintf("%s-cmt-100-%s", repo, sha)             // CMT flow
+	nightlyKey := fmt.Sprintf("%s-scheduled-200-%s", repo, sha)   // nightly flow, same SHA
+	pushKey := fmt.Sprintf("%s-push-release-9.0-%s", repo, sha)   // push flow, same SHA
+	prKey := fmt.Sprintf("%s-pr-42", repo)                        // PR flow, no -sha suffix
+	otherSHAKey := fmt.Sprintf("%s-cmt-100-%s", repo, "feedface") // CMT, different SHA
+
+	t.Run("CMT completion matches only the CMT key", func(t *testing.T) {
+		assert.True(t, instanceKeyMatchesSHA(cmtKey, repo, sha, true))
+		assert.False(t, instanceKeyMatchesSHA(nightlyKey, repo, sha, true), "nightly key must survive a CMT completion")
+		assert.False(t, instanceKeyMatchesSHA(pushKey, repo, sha, true))
+		assert.False(t, instanceKeyMatchesSHA(prKey, repo, sha, true))
+		assert.False(t, instanceKeyMatchesSHA(otherSHAKey, repo, sha, true), "different SHA must not match")
+	})
+
+	t.Run("non-CMT completion matches push/scheduled but not CMT", func(t *testing.T) {
+		assert.False(t, instanceKeyMatchesSHA(cmtKey, repo, sha, false), "CMT key must survive a nightly/push completion")
+		assert.True(t, instanceKeyMatchesSHA(nightlyKey, repo, sha, false))
+		assert.True(t, instanceKeyMatchesSHA(pushKey, repo, sha, false))
+		assert.False(t, instanceKeyMatchesSHA(prKey, repo, sha, false), "PR keys have no -sha suffix")
+	})
+
+	t.Run("other repo is never matched", func(t *testing.T) {
+		assert.False(t, instanceKeyMatchesSHA("mattermost-desktop-cmt-100-"+sha, repo, sha, true))
+	})
+}
+
 // ------------------------------------------------------------
 // 10. Instance name length safety
 // ------------------------------------------------------------

server/e2e_tests.go

@@ -1048,11 +1048,11 @@ func (s *Server) buildInstanceDetailsJSON(instances []*E2EInstance) (string, err
 	return string(jsonBytes), nil
 }
 
-// dispatchDesktopE2EWorkflow triggers the desktop E2E workflow via GitHub Actions API.
-// trackingKey is the s.e2eInstances map key for this run; when non-empty it is passed
-// as the "mw_tracking_key" workflow input so the workflow_run completed handler can do
-// a direct key lookup instead of fragile SHA suffix matching.
-func (s *Server) dispatchDesktopE2EWorkflow(repoOwner, repoName, ref, sha, instanceDetailsJSON, runType, trackingKey string, nightly bool) error {
+// dispatchDesktopE2EWorkflow triggers the desktop E2E workflow (e2e-functional.yml) via the
+// GitHub Actions API. Cleanup is driven by the workflow_run completed event matched on the
+// commit SHA, so no tracking key is passed as an input (e2e-functional.yml does not declare
+// one, and GitHub rejects a workflow_dispatch carrying an undeclared input with a 422).
+func (s *Server) dispatchDesktopE2EWorkflow(repoOwner, repoName, ref, sha, instanceDetailsJSON, runType string, nightly bool) error {
 	ctx := context.Background()
 	client := newGithubClient(s.Config.GithubAccessToken)
 
@@ -1085,9 +1085,6 @@ func (s *Server) dispatchDesktopE2EWorkflow(repoOwner, repoName, ref, sha, insta
 		"run_type":          runType,
 		"nightly":           fmt.Sprintf("%t", nightly),
 	}
-	if trackingKey != "" {
-		workflowInputs["mw_tracking_key"] = trackingKey
-	}
 
 	// Use REST API to trigger workflow dispatch (v32 go-github compatibility)
 	req, err := client.NewRequest("POST",
@@ -1116,11 +1113,11 @@ func (s *Server) dispatchDesktopE2EWorkflow(repoOwner, repoName, ref, sha, insta
 	return nil
 }
 
-// dispatchMobileE2EWorkflow triggers the mobile E2E workflow via GitHub Actions API.
-// trackingKey is the s.e2eInstances map key for this run; when non-empty it is passed
-// as the "mw_tracking_key" workflow input so the workflow_run completed handler can do
-// a direct key lookup instead of fragile SHA suffix matching.
-func (s *Server) dispatchMobileE2EWorkflow(repoOwner, repoName, ref, sha, site1URL, site2URL, site3URL, platform, runType, trackingKey string) error {
+// dispatchMobileE2EWorkflow triggers the mobile E2E workflow (e2e-detox-pr.yml) via the
+// GitHub Actions API. Cleanup is driven by the workflow_run completed event matched on the
+// commit SHA, so no tracking key is passed as an input (e2e-detox-pr.yml does not declare
+// one, and GitHub rejects a workflow_dispatch carrying an undeclared input with a 422).
+func (s *Server) dispatchMobileE2EWorkflow(repoOwner, repoName, ref, sha, site1URL, site2URL, site3URL, platform, runType string) error {
 	ctx := context.Background()
 	client := newGithubClient(s.Config.GithubAccessToken)
 
@@ -1138,9 +1135,6 @@ func (s *Server) dispatchMobileE2EWorkflow(repoOwner, repoName, ref, sha, site1U
 		"PLATFORM":       platform,
 		"run_type":       runType,
 	}
-	if trackingKey != "" {
-		workflowInputs["mw_tracking_key"] = trackingKey
-	}
 
 	// Use REST API to trigger workflow dispatch (v32 go-github compatibility)
 	req, err := client.NewRequest("POST",

server/push_events.go

@@ -140,7 +140,7 @@ func (s *Server) handlePushEventE2E(event *github.PushEvent, branch string) {
 	s.e2eInstances[key] = instances
 	s.e2eInstancesLock.Unlock()
 
-	err = s.triggerE2EWorkflowForPushEvent(repoName, instanceType, branch, sha, key, instances)
+	err = s.triggerE2EWorkflowForPushEvent(repoName, instanceType, branch, sha, instances)
 	if err != nil {
 		logger.WithError(err).Error("Failed to trigger E2E workflow")
 		s.e2eInstancesLock.Lock()
@@ -249,8 +249,8 @@ func getRunnerForPlatform(platform string) string {
 }
 
 // triggerE2EWorkflowForPushEvent routes to the desktop or mobile dispatch function.
-// trackingKey is embedded in workflow inputs as mw_tracking_key for cleanup on completion.
-func (s *Server) triggerE2EWorkflowForPushEvent(repoName, instanceType, branch, sha, trackingKey string, instances []*E2EInstance) error {
+// Cleanup is driven by the workflow_run completed event matched on the commit SHA.
+func (s *Server) triggerE2EWorkflowForPushEvent(repoName, instanceType, branch, sha string, instances []*E2EInstance) error {
 	logger := s.Logger.WithFields(logrus.Fields{
 		"repo":         repoName,
 		"instanceType": instanceType,
@@ -265,14 +265,14 @@ func (s *Server) triggerE2EWorkflowForPushEvent(repoName, instanceType, branch,
 	}
 
 	if instanceType == "desktop" {
-		return s.triggerDesktopE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha, trackingKey, instances)
+		return s.triggerDesktopE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha, instances)
 	}
 
-	return s.triggerMobileE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha, trackingKey, instances)
+	return s.triggerMobileE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha, instances)
 }
 
 // triggerDesktopE2EWorkflowForPushEvent dispatches the desktop E2E workflow.
-func (s *Server) triggerDesktopE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha, trackingKey string, instances []*E2EInstance) error {
+func (s *Server) triggerDesktopE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha string, instances []*E2EInstance) error {
 	logger := s.Logger.WithFields(logrus.Fields{
 		"repo":   repoName,
 		"branch": branch,
@@ -291,11 +291,11 @@ func (s *Server) triggerDesktopE2EWorkflowForPushEvent(repoOwner, repoName, bran
 		runType = "RELEASE"
 	}
 
-	return s.dispatchDesktopE2EWorkflow(repoOwner, repoName, branch, sha, instanceDetailsJSON, runType, trackingKey, false)
+	return s.dispatchDesktopE2EWorkflow(repoOwner, repoName, branch, sha, instanceDetailsJSON, runType, false)
 }
 
 // triggerMobileE2EWorkflowForPushEvent dispatches the mobile E2E workflow (e2e-detox-pr.yml).
-func (s *Server) triggerMobileE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha, trackingKey string, instances []*E2EInstance) error {
+func (s *Server) triggerMobileE2EWorkflowForPushEvent(repoOwner, repoName, branch, sha string, instances []*E2EInstance) error {
 	logger := s.Logger.WithFields(logrus.Fields{
 		"repo":   repoName,
 		"branch": branch,
@@ -321,6 +321,6 @@ func (s *Server) triggerMobileE2EWorkflowForPushEvent(repoOwner, repoName, branc
 		repoOwner, repoName, branch, sha,
 		instances[0].URL, instances[1].URL, instances[2].URL,
 		"both", // push events always test both iOS and Android
-		runType, trackingKey,
+		runType,
 	)
 }