fix(coderd): Ensure agent disconnect happens after timeout (coder#6600)

mafredri · web-flow · commit 348530000f76 · 2023-03-14T13:14:47.000Z
Fixes coder#6598
diff --git a/coderd/coderd.go b/coderd/coderd.go
@@ -175,6 +175,10 @@ func New(options *Options) *API {
 	if options.AgentInactiveDisconnectTimeout == 0 {
 		// Multiply the update by two to allow for some lag-time.
 		options.AgentInactiveDisconnectTimeout = options.AgentConnectionUpdateFrequency * 2
+		// Set a minimum timeout to avoid disconnecting too soon.
+		if options.AgentInactiveDisconnectTimeout < 2*time.Second {
+			options.AgentInactiveDisconnectTimeout = 2 * time.Second
+		}
 	}
 	if options.AgentStatsRefreshInterval == 0 {
 		options.AgentStatsRefreshInterval = 5 * time.Minute
diff --git a/coderd/workspaceagents.go b/coderd/workspaceagents.go
@@ -615,7 +615,7 @@ func (api *API) workspaceAgentCoordinate(rw http.ResponseWriter, r *http.Request
 
 	// We use a custom heartbeat routine here instead of `httpapi.Heartbeat`
 	// because we want to log the agent's last ping time.
-	var lastPing time.Time
+	lastPing := time.Now() // Since the agent initiated the request, assume it's alive.
 	var pingMu sync.Mutex
 	go pprof.Do(ctx, pprof.Labels("agent", workspaceAgent.ID.String()), func(ctx context.Context) {
 		// TODO(mafredri): Is this too frequent? Use separate ping disconnect timeout?
diff --git a/scaletest/reconnectingpty/run_test.go b/scaletest/reconnectingpty/run_test.go
@@ -23,7 +23,6 @@ import (
 
 func Test_Runner(t *testing.T) {
 	t.Parallel()
-	t.Skip("https://github.com/coder/coder/issues/6598")
 
 	t.Run("OK", func(t *testing.T) {
 		t.Parallel()
