Merge pull request #19 from simsteward/feature/sentry-phase1

feat: Sentry integration Phase 1 — Replay, Transactions, Cron

Author: simsteward

deploy.ps1

@@ -434,6 +434,52 @@ function Push-SentryApi {
     }
 }
 
+function Push-SentryCheckIn {
+    param([string]$MonitorSlug, [string]$Status, [string]$CheckInId)
+    if ([string]::IsNullOrWhiteSpace($sentryAuthToken)) { return $null }
+    try {
+        $headers = @{ Authorization = "Bearer $sentryAuthToken"; 'Content-Type' = 'application/json' }
+        if ([string]::IsNullOrWhiteSpace($CheckInId)) {
+            # Initial check-in: POST with monitor_config for auto-creation
+            $url = "https://sentry.io/api/0/organizations/$sentryOrg/monitors/$MonitorSlug/checkins/"
+            $body = @{
+                status         = $Status
+                monitor_config = @{
+                    schedule        = @{ type = 'interval'; value = 1; unit = 'day' }
+                    checkin_margin  = 5
+                    max_runtime     = 10
+                    timezone        = 'UTC'
+                }
+            }
+            $json = $body | ConvertTo-Json -Compress -Depth 5
+            $resp = Invoke-RestMethod -Uri $url -Method Post -Headers $headers -Body $json -ErrorAction Stop
+            $newId = $resp.id
+            Push-LokiEvent 'deploy_sentry_checkin' 'INFO' "Sentry cron check-in started: $MonitorSlug" @{
+                monitor_slug = $MonitorSlug
+                status       = $Status
+                checkin_id   = $newId
+            }
+            return $newId
+        } else {
+            # Completion check-in: PUT to update existing
+            $url = "https://sentry.io/api/0/organizations/$sentryOrg/monitors/$MonitorSlug/checkins/$CheckInId/"
+            $body = @{ status = $Status }
+            $json = $body | ConvertTo-Json -Compress -Depth 5
+            Invoke-RestMethod -Uri $url -Method Put -Headers $headers -Body $json -ErrorAction Stop | Out-Null
+            Push-LokiEvent 'deploy_sentry_checkin' 'INFO' "Sentry cron check-in completed: $MonitorSlug ($Status)" @{
+                monitor_slug = $MonitorSlug
+                status       = $Status
+                checkin_id   = $CheckInId
+            }
+            return $null
+        }
+    } catch {
+        # Non-fatal: deploy must not fail because Sentry API is down
+        Write-Warning "Sentry CheckIn ($MonitorSlug/$Status): $($_.Exception.Message)"
+        return $null
+    }
+}
+
 if (-not [string]::IsNullOrWhiteSpace($sentryAuthToken) -and -not [string]::IsNullOrWhiteSpace($sentryRelease)) {
     Write-Host "Registering Sentry release: $sentryRelease (3 projects)"
 
@@ -509,6 +555,7 @@ if (-not $skipTests) {
                 script_count = $testScripts.Count
                 scripts      = ($testScripts | ForEach-Object { $_.Name }) -join ','
             }
+            $checkInId = Push-SentryCheckIn 'post-deploy-tests' 'in_progress'
             foreach ($ts in $testScripts) {
                 Write-Host "  Running: $($ts.Name)"
                 $tsStart = Get-Date
@@ -557,6 +604,8 @@ if (-not $skipTests) {
                     }
                 }
             }
+            $checkInStatus = if ($postDeployFailed) { 'error' } else { 'ok' }
+            Push-SentryCheckIn 'post-deploy-tests' $checkInStatus $checkInId | Out-Null
             if ($postDeployFailed) {
                 Write-Warning "Post-deploy tests failed. Deploy files are in place but integration is not fully verified."
             } else {

src/SimSteward.Dashboard/data-capture-suite.html

@@ -244,12 +244,15 @@
 .filter-clear { background: none; border: none; color: var(--muted); cursor: pointer; font-size: 0.82rem; padding: 0 4px; margin-left: -28px; margin-right: 4px; transition: color 0.15s; z-index: 1; }
 .filter-clear:hover { color: var(--text); }
 </style>
-<script src="https://browser.sentry-cdn.com/9.27.0/bundle.min.js" crossorigin="anonymous"></script>
+<script src="https://browser.sentry-cdn.com/9.27.0/bundle.tracing.replay.min.js" crossorigin="anonymous"></script>
 <script>
 Sentry.init({
   dsn: 'https://46ad94b35a51979ba4223eeffe38f5c8@o4511097126780928.ingest.us.sentry.io/4511103122210816',
   environment: 'local',
   tracesSampleRate: 1.0,
+  replaysSessionSampleRate: 0.1,
+  replaysOnErrorSampleRate: 1.0,
+  integrations: [Sentry.replayIntegration()],
 });
 Sentry.setTag('dashboard', 'data-capture-suite');
 </script>

src/SimSteward.Dashboard/index.html

@@ -475,13 +475,16 @@
 ::-webkit-scrollbar-track { background: transparent; }
 ::-webkit-scrollbar-thumb { background: var(--border); border-radius: 2px; }
 </style>
-<script src="https://browser.sentry-cdn.com/9.27.0/bundle.min.js" crossorigin="anonymous"></script>
+<script src="https://browser.sentry-cdn.com/9.27.0/bundle.tracing.replay.min.js" crossorigin="anonymous"></script>
 <script>
 Sentry.init({
   dsn: 'https://46ad94b35a51979ba4223eeffe38f5c8@o4511097126780928.ingest.us.sentry.io/4511103122210816',
   environment: 'local',
   release: '',
   tracesSampleRate: 1.0,
+  replaysSessionSampleRate: 0.1,
+  replaysOnErrorSampleRate: 1.0,
+  integrations: [Sentry.replayIntegration()],
   initialScope: { tags: { dashboard: 'index' } },
 });
 </script>

src/SimSteward.Dashboard/replay-incident-index.html

@@ -65,6 +65,18 @@
 }
 .toast.show { opacity: 1; }
 </style>
+<script src="https://browser.sentry-cdn.com/9.27.0/bundle.tracing.replay.min.js" crossorigin="anonymous"></script>
+<script>
+Sentry.init({
+  dsn: 'https://46ad94b35a51979ba4223eeffe38f5c8@o4511097126780928.ingest.us.sentry.io/4511103122210816',
+  environment: 'local',
+  tracesSampleRate: 1.0,
+  replaysSessionSampleRate: 0.1,
+  replaysOnErrorSampleRate: 1.0,
+  integrations: [Sentry.replayIntegration()],
+  initialScope: { tags: { dashboard: 'replay-incident-index' } },
+});
+</script>
 </head>
 <body>
 <div class="wrap">

src/SimSteward.Plugin/DashboardBridge.cs

@@ -278,8 +278,22 @@ private void HandleMessage(IWebSocketConnection socket, string msg)
             }
 
             var correlationId = Guid.NewGuid().ToString("N").Substring(0, 8);
-            var (success, result, error) = _dispatchAction(action, arg ?? "", correlationId);
-            SendActionResult(socket, action, success, result, error);
+            var tx = SentrySdk.StartTransaction("ws.message", "handle");
+            tx.SetExtra("action", action);
+            tx.SetExtra("correlation_id", correlationId);
+            SentrySdk.ConfigureScope(scope => scope.Transaction = tx);
+            try
+            {
+                var (success, result, error) = _dispatchAction(action, arg ?? "", correlationId);
+                SendActionResult(socket, action, success, result, error);
+                tx.Finish(success ? SpanStatus.Ok : SpanStatus.InternalError);
+            }
+            catch (Exception ex)
+            {
+                SentrySdk.CaptureException(ex);
+                tx.Finish(SpanStatus.InternalError);
+                throw;
+            }
         }
 
         private void SendActionResult(IWebSocketConnection socket, string action, bool success, string result, string error)

src/SimSteward.Plugin/SimStewardPlugin.DataCaptureSuite.cs

@@ -108,6 +108,10 @@ private enum PreflightStep
         private int  _suiteT8PollTicks;
         private bool _suiteT8BuildWasRunning;
 
+        // Sentry performance tracing
+        private ITransactionTracer _sentryTx;
+        private ISpan              _sentryCurrentSpan;
+
         // ── Public entry points (called from DataUpdate / DispatchAction) ──────
 
         private void TryStartDataCaptureSuite(string[] skipIds = null)
@@ -184,6 +188,13 @@ private void ProcessDataCaptureSuiteTick()
                     _suite60HzRecorder = null;
                     StopReplayIncidentIndexRecordModeLocked("suite_cancel");
                     EmitSuiteLifecycleEvent("sdk_capture_suite_cancelled", "Suite cancelled.", "T_cancel");
+
+                    // Sentry: finish spans/transaction as cancelled
+                    _sentryCurrentSpan?.Finish(SpanStatus.Cancelled);
+                    _sentryCurrentSpan = null;
+                    _sentryTx?.Finish(SpanStatus.Cancelled);
+                    _sentryTx = null;
+
                     _suitePhase = DataCaptureSuitePhase.Cancelled;
                 }
                 return;
@@ -683,6 +694,12 @@ private void BeginDataCaptureSuite()
             _suiteStep  = SuiteInternalStep.T0_Rewind;
             _suitePhase = DataCaptureSuitePhase.Running;
 
+            // Sentry performance transaction for the entire suite run
+            _sentryTx = SentrySdk.StartTransaction("data-capture-suite", "test.run");
+            _sentryTx.SetExtra("test_run_id", _suiteTestRunId);
+            SentrySdk.ConfigureScope(scope => scope.Transaction = _sentryTx);
+            _sentryCurrentSpan = _sentryTx.StartChild("step", SuiteInternalStep.T0_Rewind.ToString());
+
             EmitSuiteLifecycleEvent(DataCaptureSuiteConstants.EventSuiteStarted,
                 $"Data capture suite started. test_run_id={_suiteTestRunId}", "T_start");
             SentrySdk.AddBreadcrumb("Data capture suite started", "lifecycle",
@@ -694,6 +711,8 @@ private void BeginDataCaptureSuite()
 
         private void TickSuiteRunning()
         {
+            var stepBefore = _suiteStep;
+
             switch (_suiteStep)
             {
                 case SuiteInternalStep.T0_Rewind:        TickT0_Rewind();        break;
@@ -724,6 +743,16 @@ private void TickSuiteRunning()
                 case SuiteInternalStep.Done:           TransitionToLoki();   break;
             }
 
+            // Sentry: finish previous span and start new one when step changes
+            if (_suiteStep != stepBefore && _sentryTx != null)
+            {
+                _sentryCurrentSpan?.Finish(SpanStatus.Ok);
+                if (_suiteStep != SuiteInternalStep.Done)
+                    _sentryCurrentSpan = _sentryTx.StartChild("step", _suiteStep.ToString());
+                else
+                    _sentryCurrentSpan = null;
+            }
+
             // 60Hz recording: every tick while running
             _suite60HzRecorder?.RecordTick(_irsdk);
         }
@@ -1638,6 +1667,12 @@ private void TransitionToLoki()
             _suiteEmitCompleteUtc = DateTime.UtcNow;
             _suitePhase = DataCaptureSuitePhase.AwaitingLoki;
 
+            // Sentry: finish any remaining span and the transaction
+            _sentryCurrentSpan?.Finish(SpanStatus.Ok);
+            _sentryCurrentSpan = null;
+            _sentryTx?.Finish(SpanStatus.Ok);
+            _sentryTx = null;
+
             var fields = BuildTestFields("T_done");
             fields["loki_wait_ms"] = DataCaptureSuiteConstants.LokiVerifyDelayMs;
             MergeSessionAndRoutingFields(fields);

src/SimSteward.Plugin/SimStewardPlugin.cs

@@ -609,6 +609,16 @@ private System.Collections.Generic.Dictionary<string, object> BuildCaptureIncide
         {
             if (string.IsNullOrEmpty(action))
                 return (false, null, "missing_action");
+
+            // Sentry: create a child span under the current transaction (if any)
+            ISpan _actionSpan = null;
+            SentrySdk.ConfigureScope(scope =>
+            {
+                var parentTx = scope.Transaction;
+                if (parentTx != null)
+                    _actionSpan = parentTx.StartChild("action", action ?? "unknown");
+            });
+
             var dispatchFields = new System.Collections.Generic.Dictionary<string, object>
             {
                 ["action"] = action,
@@ -618,6 +628,9 @@ private System.Collections.Generic.Dictionary<string, object> BuildCaptureIncide
             MergeSessionAndRoutingFields(dispatchFields);
             _logger?.Structured("INFO", "simhub-plugin", "action_dispatched", action, dispatchFields, "action", null);
 
+            try
+            {
+
             if (string.Equals(action, "replay_session", StringComparison.OrdinalIgnoreCase))
             {
                 var dir = (arg ?? "").Trim().ToLowerInvariant();
@@ -943,6 +956,18 @@ private System.Collections.Generic.Dictionary<string, object> BuildCaptureIncide
 
             LogActionResult(action, arg, correlationId, false, "not_supported");
             return (false, null, "not_supported");
+
+            }
+            catch
+            {
+                _actionSpan?.Finish(SpanStatus.InternalError);
+                _actionSpan = null;
+                throw;
+            }
+            finally
+            {
+                _actionSpan?.Finish(SpanStatus.Ok);
+            }
         }
 
         private void OnLog(string level, string message, string source)