Include OpenTelemetry traces in rageshakes

src/analytics/OtelPosthogExporter.ts

@@ -15,9 +15,12 @@ limitations under the License.
 */
 
 import { SpanExporter, ReadableSpan } from "@opentelemetry/sdk-trace-base";
-import { ExportResult, ExportResultCode } from "@opentelemetry/core";
+import {
+  ExportResult,
+  ExportResultCode,
+  hrTimeToMilliseconds,
+} from "@opentelemetry/core";
 import { logger } from "matrix-js-sdk/src/logger";
-import { HrTime } from "@opentelemetry/api";
 
 import { PosthogAnalytics } from "./PosthogAnalytics";
 
@@ -26,10 +29,6 @@ interface PrevCall {
   hangupTs: number;
 }
 
-function hrTimeToMs(time: HrTime): number {
-  return time[0] * 1000 + time[1] * 0.000001;
-}
-
 /**
  * The maximum time between hanging up and joining the same call that we would
  * consider a 'rejoin' on the user's part.
@@ -86,14 +85,14 @@ export class PosthogSpanExporter implements SpanExporter {
     const prevCall = this.prevCall;
     const newPrevCall = (this.prevCall = {
       callId: span.attributes["matrix.confId"] as string,
-      hangupTs: hrTimeToMs(span.endTime),
+      hangupTs: hrTimeToMilliseconds(span.endTime),
     });
 
     // If the user joined the same call within a short time frame, log this as a
     // rejoin. This is interesting as a call quality metric, since rejoins may
     // indicate that users had to intervene to make the product work.
     if (prevCall !== null && newPrevCall.callId === prevCall.callId) {
-      const duration = hrTimeToMs(span.startTime) - prevCall.hangupTs;
+      const duration = hrTimeToMilliseconds(span.startTime) - prevCall.hangupTs;
       if (duration <= maxRejoinMs) {
         PosthogAnalytics.instance.trackEvent(
           {

src/analytics/RageshakeSpanExporter.ts

@@ -0,0 +1,103 @@
+import { Attributes } from "@opentelemetry/api";
+import {
+  ExportResult,
+  ExportResultCode,
+  hrTimeToMilliseconds,
+} from "@opentelemetry/core";
+import { SpanExporter, ReadableSpan } from "@opentelemetry/sdk-trace-base";
+
+const dumpAttributes = (attr: Attributes) =>
+  Object.entries(attr).map(([key, value]) => ({
+    key,
+    type: typeof value,
+    value,
+  }));
+
+/**
+ * Exports spans on demand to the Jaeger JSON format, which can be attached to
+ * rageshakes and loaded into analysis tools like Jaeger and Stalk.
+ */
+export class RageshakeSpanExporter implements SpanExporter {
+  private readonly spans: ReadableSpan[] = [];
+
+  export(
+    spans: ReadableSpan[],
+    resultCallback: (result: ExportResult) => void
+  ): void {
+    this.spans.push(...spans);
+    resultCallback({ code: ExportResultCode.SUCCESS });
+  }
+
+  /**
+   * Dumps the spans collected so far as Jaeger-compatible JSON.
+   */
+  public dump(): string {
+    const traces = new Map<string, ReadableSpan[]>();
+
+    // Organize spans by their trace IDs
+    for (const span of this.spans) {
+      const traceId = span.spanContext().traceId;
+      let trace = traces.get(traceId);
+
+      if (trace === undefined) {
+        trace = [];
+        traces.set(traceId, trace);
+      }
+
+      trace.push(span);
+    }
+
+    const processId = "p1";
+    const processes = {
+      [processId]: {
+        serviceName: "element-call",
+        tags: [],
+      },
+      warnings: null,
+    };
+
+    return JSON.stringify({
+      // Honestly not sure what some of these fields mean, I just know that
+      // they're present in Jaeger JSON exports
+      total: 0,
+      limit: 0,
+      offset: 0,
+      errors: null,
+      data: [...traces.entries()].map(([traceId, spans]) => ({
+        traceID: traceId,
+        warnings: null,
+        processes,
+        spans: spans.map((span) => {
+          const ctx = span.spanContext();
+
+          return {
+            traceID: traceId,
+            spanID: ctx.spanId,
+            operationName: span.name,
+            processID: processId,
+            warnings: null,
+            startTime: hrTimeToMilliseconds(span.startTime),
+            duration: hrTimeToMilliseconds(span.duration),
+            references:
+              span.parentSpanId === undefined
+                ? []
+                : [
+                    {
+                      refType: "CHILD_OF",
+                      traceID: traceId,
+                      spanID: span.parentSpanId,
+                    },
+                  ],
+            tags: dumpAttributes(span.attributes),
+            logs: span.events.map((event) => ({
+              timestamp: hrTimeToMilliseconds(event.time),
+              fields: dumpAttributes(event.attributes ?? {}),
+            })),
+          };
+        }),
+      })),
+    });
+  }
+
+  async shutdown(): Promise<void> {}
+}

src/otel/otel.ts

@@ -28,6 +28,7 @@ import { logger } from "matrix-js-sdk/src/logger";
 import { PosthogSpanExporter } from "../analytics/OtelPosthogExporter";
 import { Anonymity } from "../analytics/PosthogAnalytics";
 import { Config } from "../config/Config";
+import { RageshakeSpanExporter } from "../analytics/RageshakeSpanExporter";
 
 const SERVICE_NAME = "element-call";
 
@@ -37,21 +38,24 @@ export class ElementCallOpenTelemetry {
   private _provider: WebTracerProvider;
   private _tracer: Tracer;
   private _anonymity: Anonymity;
-  private _otlpExporter: OTLPTraceExporter;
+  private otlpExporter: OTLPTraceExporter;
+  public readonly rageshakeExporter?: RageshakeSpanExporter;
 
   static globalInit(): void {
+    const config = Config.get();
     // we always enable opentelemetry in general. We only enable the OTLP
     // collector if a URL is defined (and in future if another setting is defined)
     // The posthog exporteer is always enabled, posthog reporting is enabled or disabled
     // within the posthog code.
-    const shouldEnableOtlp = Boolean(Config.get().opentelemetry?.collector_url);
+    const shouldEnableOtlp = Boolean(config.opentelemetry?.collector_url);
 
     if (!sharedInstance || sharedInstance.isOtlpEnabled !== shouldEnableOtlp) {
       logger.info("(Re)starting OpenTelemetry debug reporting");
       sharedInstance?.dispose();
 
       sharedInstance = new ElementCallOpenTelemetry(
-        Config.get().opentelemetry?.collector_url
+        config.opentelemetry?.collector_url,
+        config.rageshake?.submit_url
       );
     }
   }
@@ -60,7 +64,10 @@ export class ElementCallOpenTelemetry {
     return sharedInstance;
   }
 
-  constructor(collectorUrl: string | undefined) {
+  constructor(
+    collectorUrl: string | undefined,
+    rageshakeUrl: string | undefined
+  ) {
     // This is how we can make Jaeger show a reaonsable service in the dropdown on the left.
     const providerConfig = {
       resource: new Resource({
@@ -71,15 +78,24 @@ export class ElementCallOpenTelemetry {
 
     if (collectorUrl) {
       logger.info("Enabling OTLP collector with URL " + collectorUrl);
-      this._otlpExporter = new OTLPTraceExporter({
+      this.otlpExporter = new OTLPTraceExporter({
         url: collectorUrl,
       });
       this._provider.addSpanProcessor(
-        new SimpleSpanProcessor(this._otlpExporter)
+        new SimpleSpanProcessor(this.otlpExporter)
       );
     } else {
       logger.info("OTLP collector disabled");
     }
+
+    if (rageshakeUrl) {
+      logger.info("Enabling rageshake collector");
+      this.rageshakeExporter = new RageshakeSpanExporter();
+      this._provider.addSpanProcessor(
+        new SimpleSpanProcessor(this.rageshakeExporter)
+      );
+    }
+
     const consoleExporter = new ConsoleSpanExporter();
     const posthogExporter = new PosthogSpanExporter();
 
@@ -99,7 +115,7 @@ export class ElementCallOpenTelemetry {
   }
 
   public get isOtlpEnabled(): boolean {
-    return Boolean(this._otlpExporter);
+    return Boolean(this.otlpExporter);
   }
 
   public get tracer(): Tracer {

src/settings/submit-rageshake.ts

@@ -25,6 +25,14 @@ import { useClient } from "../ClientContext";
 import { InspectorContext } from "../room/GroupCallInspector";
 import { useModalTriggerState } from "../Modal";
 import { Config } from "../config/Config";
+import { ElementCallOpenTelemetry } from "../otel/otel";
+
+const gzip = (text: string): Blob => {
+  // encode as UTF-8
+  const buf = new TextEncoder().encode(text);
+  // compress
+  return new Blob([pako.gzip(buf)]);
+};
 
 interface RageShakeSubmitOptions {
   sendLogs: boolean;
@@ -235,14 +243,15 @@ export function useSubmitRageshake(): {
           const logs = await getLogsForReport();
 
           for (const entry of logs) {
-            // encode as UTF-8
+            body.append("compressed-log", gzip(entry.lines), entry.id);
-            let buf = new TextEncoder().encode(entry.lines);
-            // compress
-            buf = pako.gzip(buf);
-
-            body.append("compressed-log", new Blob([buf]), entry.id);
           }
 
+          body.append(
+            "file",
+            gzip(ElementCallOpenTelemetry.instance.rageshakeExporter!.dump()),
+            "traces.json"
+          );
+
           if (inspectorState) {
             body.append(
               "file",