feat: e2e metrics reporting (#9776)

.github/workflows/ci.yml

@@ -208,6 +208,7 @@ jobs:
         uses: ./.github/ensure-tester-with-images
         env:
           USERNAME: ${{ needs.configure.outputs.username }}
+          PULL_REQUEST: ${{ github.event.pull_request.number }}
         with:
           runner_type: ${{ steps.runner_type.outputs.type }}
           builder_type: builder-x86

yarn-project/end-to-end/package.json

@@ -16,6 +16,7 @@
     "formatting": "run -T prettier --check ./src \"!src/web/main.js\" && run -T eslint ./src",
     "formatting:fix": "run -T eslint --fix ./src && run -T prettier -w ./src",
     "test": "LOG_LEVEL=${LOG_LEVEL:-verbose} DEBUG_COLORS=1 NODE_NO_WARNINGS=1 node --experimental-vm-modules ../node_modules/.bin/jest --testTimeout=300000 --forceExit",
+    "test:with-alerts": "./scripts/test-with-alerts.sh",
     "test:profile": "LOG_LEVEL=${LOG_LEVEL:-verbose} DEBUG_COLORS=1 NODE_NO_WARNINGS=1 0x --output-dir \"flame_graph/{pid}.0x\" -- node --experimental-vm-modules ../node_modules/jest/bin/jest.js --runInBand --testTimeout=300000 --forceExit",
     "serve:flames": "python3 -m http.server --directory \"flame_graph\" 8000",
     "test:debug": "LOG_LEVEL=${LOG_LEVEL:-verbose} DEBUG_COLORS=1 NODE_NO_WARNINGS=1 node --inspect --experimental-vm-modules ../node_modules/.bin/jest --testTimeout=300000 --forceExit",
@@ -99,10 +100,12 @@
     "0x": "^5.7.0",
     "@jest/globals": "^29.5.0",
     "@types/jest": "^29.5.0",
+    "@types/js-yaml": "^4.0.9",
     "@types/lodash.chunk": "^4.2.9",
     "concurrently": "^7.6.0",
     "jest": "^29.5.0",
     "jest-extended": "^4.0.2",
+    "js-yaml": "^4.1.0",
     "ts-node": "^10.9.1",
     "typescript": "^5.0.4"
   },

yarn-project/end-to-end/scripts/e2e_test.sh

@@ -50,6 +50,8 @@ fi
 # Check if the test uses docker compose
 if [ "$(echo "$test_config" | yq e '.use_compose // false' -)" = "true" ]; then
   $(dirname "$0")/e2e_compose_test.sh "$test_path" "$@" || [ "$ignore_failures" = "true" ]
+elif [ "$(echo "$test_config" | yq e '.with_alerts // false' -)" = "true" ]; then
+  $(dirname "$0")/e2e_test_with_alerts.sh "$test_path" "$@" || [ "$ignore_failures" = "true" ]
 else
   # Set environment variables
   while IFS='=' read -r key value; do

yarn-project/end-to-end/scripts/e2e_test_config.yml

@@ -85,9 +85,9 @@ tests:
   e2e_token_contract: {}
   e2e_p2p_gossip:
     test_path: 'e2e_p2p/gossip_network.test.ts'
+    with_alerts: true
   e2e_p2p_upgrade_governance_proposer:
     test_path: 'e2e_p2p/upgrade_governance_proposer.test.ts'
-  # https://github.com/AztecProtocol/aztec-packages/issues/9843
   e2e_p2p_rediscovery:
     test_path: 'e2e_p2p/rediscovery.test.ts'
   e2e_p2p_reqresp:

yarn-project/end-to-end/scripts/e2e_test_with_alerts.sh

@@ -0,0 +1,51 @@
+#! /bin/bash
+## Run an end to end test with alerts
+
+# This will run an end to end test running the otel-lgtm stack (otel-collector, grafana, prometheus, tempo and loki)
+# Then check the test against a set of alerts defined in the alerts.yaml file
+# Note: these tests must run with METRICS enabled
+
+# Usage: ./e2e_test_with_alerts.sh <test-name> <...extra-args>
+# Example: ./e2e_test_with_alerts.sh gossip_network
+
+set -e
+
+test_path=$1
+
+echo "Running otel stack"
+CONTAINER_ID=$(docker run -d -p 3000:3000 -p 4317:4317 -p 4318:4318 --rm grafana/otel-lgtm)
+
+trap "docker stop $CONTAINER_ID" EXIT SIGINT SIGTERM
+
+echo "Waiting for LGTM stack to be ready..."
+timeout=90
+while [ $timeout -gt 0 ]; do
+    if docker logs $CONTAINER_ID 2>&1 | grep -q "The OpenTelemetry collector and the Grafana LGTM stack are up and running"; then
+        echo "LGTM stack is ready!"
+        break
+    fi
+    sleep 1
+    ((timeout--))
+done
+
+if [ $timeout -eq 0 ]; then
+    echo "Timeout waiting for LGTM stack to be ready"
+    docker stop $CONTAINER_ID
+    exit 1
+fi
+
+## Pass through run the existing e2e test
+docker run \
+    --network host \
+    -e HARDWARE_CONCURRENCY="$HARDWARE_CONCURRENCY" \
+    -e FAKE_PROOFS="$FAKE_PROOFS" \
+    -e METRICS_PORT="4318" \
+    -e COLLECT_METRICS="true" \
+    -e PULL_REQUEST="$PULL_REQUEST" \
+    $env_args \
+    --rm aztecprotocol/end-to-end:$AZTEC_DOCKER_TAG \
+    "$test_path" "$@" || [ "$ignore_failures" = "true" ]
+
+
+echo "Running alert checker..."
+docker run --network host --rm aztecprotocol/end-to-end:$AZTEC_DOCKER_TAG quality_of_service/alert_checker.test.ts

yarn-project/end-to-end/src/e2e_p2p/rediscovery.test.ts

@@ -3,6 +3,7 @@ import { sleep } from '@aztec/aztec.js';
 
 import fs from 'fs';
 
+import { shouldCollectMetrics } from '../fixtures/fixtures.js';
 import { type NodeContext, createNode, createNodes } from '../fixtures/setup_p2p_test.js';
 import { P2PNetworkTest, WAIT_FOR_TX_TIMEOUT } from './p2p_network.js';
 import { createPXEServiceAndSubmitTransactions } from './shared.js';
@@ -23,6 +24,8 @@ describe('e2e_p2p_rediscovery', () => {
       testName: 'e2e_p2p_rediscovery',
       numberOfNodes: NUM_NODES,
       basePort: BOOT_NODE_UDP_PORT,
+      // To collect metrics - run in aztec-packages `docker compose --profile metrics up` and set COLLECT_METRICS=true
+      metricsPort: shouldCollectMetrics(),
     });
     await t.applyBaseSnapshots();
     await t.setup();
@@ -48,6 +51,8 @@ describe('e2e_p2p_rediscovery', () => {
       NUM_NODES,
       BOOT_NODE_UDP_PORT,
       DATA_DIR,
+      // To collect metrics - run in aztec-packages `docker compose --profile metrics up`
+      shouldCollectMetrics(),
     );
 
     // wait a bit for peers to discover each other

yarn-project/end-to-end/src/e2e_p2p/reex.test.ts

@@ -7,6 +7,7 @@ import { BlockProposal, SignatureDomainSeperator, getHashedSignaturePayload } fr
 import { beforeAll, describe, it, jest } from '@jest/globals';
 import fs from 'fs';
 
+import { shouldCollectMetrics } from '../fixtures/fixtures.js';
 import { createNodes } from '../fixtures/setup_p2p_test.js';
 import { P2PNetworkTest } from './p2p_network.js';
 import { submitComplexTxsTo } from './shared.js';
@@ -28,6 +29,8 @@ describe('e2e_p2p_reex', () => {
       testName: 'e2e_p2p_reex',
       numberOfNodes: NUM_NODES,
       basePort: BOOT_NODE_UDP_PORT,
+      // To collect metrics - run in aztec-packages `docker compose --profile metrics up` and set COLLECT_METRICS=true
+      metricsPort: shouldCollectMetrics(),
     });
 
     t.logger.verbose('Setup account');
@@ -66,6 +69,9 @@ describe('e2e_p2p_reex', () => {
       t.bootstrapNodeEnr,
       NUM_NODES,
       BOOT_NODE_UDP_PORT,
+      DATA_DIR,
+      // To collect metrics - run in aztec-packages `docker compose --profile metrics up` and set COLLECT_METRICS=true
+      shouldCollectMetrics(),
     );
 
     // Hook into the node and intercept re-execution logic, ensuring that it was infact called

yarn-project/end-to-end/src/e2e_p2p/reqresp.test.ts

@@ -6,6 +6,7 @@ import { jest } from '@jest/globals';
 import fs from 'fs';
 import { getContract } from 'viem';
 
+import { shouldCollectMetrics } from '../fixtures/fixtures.js';
 import { type NodeContext, createNodes } from '../fixtures/setup_p2p_test.js';
 import { P2PNetworkTest, WAIT_FOR_TX_TIMEOUT } from './p2p_network.js';
 import { createPXEServiceAndSubmitTransactions } from './shared.js';
@@ -26,6 +27,8 @@ describe('e2e_p2p_reqresp_tx', () => {
       testName: 'e2e_p2p_reqresp_tx',
       numberOfNodes: NUM_NODES,
       basePort: BOOT_NODE_UDP_PORT,
+      // To collect metrics - run in aztec-packages `docker compose --profile metrics up`
+      metricsPort: shouldCollectMetrics(),
     });
     await t.applyBaseSnapshots();
     await t.setup();
@@ -67,6 +70,7 @@ describe('e2e_p2p_reqresp_tx', () => {
       NUM_NODES,
       BOOT_NODE_UDP_PORT,
       DATA_DIR,
+      shouldCollectMetrics(),
     );
 
     // wait a bit for peers to discover each other

yarn-project/end-to-end/src/e2e_p2p/upgrade_governance_proposer.test.ts

@@ -12,6 +12,7 @@ import {
 import fs from 'fs';
 import { getAddress, getContract } from 'viem';
 
+import { shouldCollectMetrics } from '../fixtures/fixtures.js';
 import { createNodes } from '../fixtures/setup_p2p_test.js';
 import { P2PNetworkTest } from './p2p_network.js';
 
@@ -36,6 +37,8 @@ describe('e2e_p2p_governance_proposer', () => {
       testName: 'e2e_p2p_gerousia',
       numberOfNodes: NUM_NODES,
       basePort: BOOT_NODE_UDP_PORT,
+      // To collect metrics - run in aztec-packages `docker compose --profile metrics up`
+      metricsPort: shouldCollectMetrics(),
     });
     await t.applyBaseSnapshots();
     await t.setup();
@@ -132,6 +135,7 @@ describe('e2e_p2p_governance_proposer', () => {
       NUM_NODES,
       BOOT_NODE_UDP_PORT,
       DATA_DIR,
+      shouldCollectMetrics(),
     );
 
     await sleep(4000);

yarn-project/end-to-end/src/quality_of_service/alert_checker.test.ts

@@ -0,0 +1,88 @@
+import { type DebugLogger, createDebugLogger } from '@aztec/aztec.js';
+import { fileURLToPath } from '@aztec/foundation/url';
+
+import * as fs from 'fs';
+import * as yaml from 'js-yaml';
+import { dirname, join } from 'path';
+
+const GRAFANA_ENDPOINT = 'http://localhost:3000/api/datasources/proxy/uid/prometheus/api/v1/query';
+interface AlertConfig {
+  alert: string;
+  expr: string;
+  for: string;
+  labels: Record<string, string>;
+  annotations: Record<string, string>;
+}
+// Define __dirname for ES modules
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = dirname(__filename);
+
+// Load YAML configuration
+function loadAlertsConfig(filePath: string): AlertConfig[] {
+  const fileContents = fs.readFileSync(join(__dirname, filePath), 'utf8');
+  const data = yaml.load(fileContents) as { alerts: AlertConfig[] };
+  return data.alerts;
+}
+
+// Function to query Grafana based on an expression
+async function queryGrafana(expr: string): Promise<number> {
+  // Create base64 encoded credentials for basic auth
+  const credentials = Buffer.from('admin:admin').toString('base64');
+
+  const response = await fetch(`${GRAFANA_ENDPOINT}?query=${encodeURIComponent(expr)}`, {
+    headers: {
+      Authorization: `Basic ${credentials}`,
+    },
+  });
+
+  if (!response.ok) {
+    throw new Error(`Failed to fetch data from Grafana: ${response.statusText}`);
+  }
+
+  const data = await response.json();
+  const result = data.data.result;
+  return result.length > 0 ? parseFloat(result[0].value[1]) : 0;
+}
+
+// Function to check alerts based on expressions
+async function checkAlerts(alerts: AlertConfig[], logger: DebugLogger) {
+  let alertTriggered = false;
+
+  for (const alert of alerts) {
+    logger.info(`Checking alert: ${JSON.stringify(alert)}`);
+
+    const metricValue = await queryGrafana(alert.expr);
+    logger.info(`Metric value: ${metricValue}`);
+    if (metricValue > 0) {
+      logger.error(`Alert ${alert.alert} triggered! Value: ${metricValue}`);
+      alertTriggered = true;
+    } else {
+      logger.info(`Alert ${alert.alert} passed.`);
+    }
+  }
+
+  // If any alerts have been triggered we fail the test
+  if (alertTriggered) {
+    throw new Error('Test failed due to triggered alert');
+  }
+}
+
+// Main function to run tests
+async function runAlertChecker(logger: DebugLogger) {
+  const alerts = loadAlertsConfig('alerts.yaml');
+  try {
+    await checkAlerts(alerts, logger);
+    logger.info('All alerts passed.');
+  } catch (error) {
+    logger.error(error instanceof Error ? error.message : String(error));
+    process.exit(1); // Exit with error code
+  }
+}
+
+// Running as a jest test to use existing end to end test framework
+describe('Alert Checker', () => {
+  const logger = createDebugLogger('aztec:alert-checker');
+  it('should check alerts', async () => {
+    await runAlertChecker(logger);
+  });
+});

yarn-project/end-to-end/src/quality_of_service/alerts.yaml

@@ -0,0 +1,10 @@
+## A set of alerts for the quality of service of the sequencer, these are tested for in certain e2e tests
+
+## In end to end tests - page, will cause a test to fail
+## Warning will write a message to the PR
+
+alerts:
+  - alert: SequencerTimeToCollectAttestations
+    expr: aztec_sequencer_time_to_collect_attestations > 2500
+    labels:
+      severity: page

yarn-project/sequencer-client/src/sequencer/metrics.ts

@@ -21,6 +21,8 @@ export class SequencerMetrics {
   private currentBlockNumber: Gauge;
   private currentBlockSize: Gauge;
 
+  private timeToCollectAttestations: Gauge;
+
   constructor(client: TelemetryClient, getState: SequencerStateCallback, name = 'Sequencer') {
     const meter = client.getMeter(name);
     this.tracer = client.getTracer(name);
@@ -60,9 +62,26 @@ export class SequencerMetrics {
       description: 'Current block number',
     });
 
+    this.timeToCollectAttestations = meter.createGauge(Metrics.SEQUENCER_TIME_TO_COLLECT_ATTESTATIONS, {
+      description: 'The time spent collecting attestations from committee members',
+    });
+
     this.setCurrentBlock(0, 0);
   }
 
+  startCollectingAttestationsTimer(): () => void {
+    const startTime = Date.now();
+    const stop = () => {
+      const duration = Date.now() - startTime;
+      this.recordTimeToCollectAttestations(duration);
+    };
+    return stop.bind(this);
+  }
+
+  recordTimeToCollectAttestations(time: number) {
+    this.timeToCollectAttestations.record(time);
+  }
+
   recordCancelledBlock() {
     this.blockCounter.add(1, {
       [Attributes.STATUS]: 'cancelled',

yarn-project/sequencer-client/src/sequencer/sequencer.ts

@@ -633,11 +633,13 @@ export class Sequencer {
       const txHashes = validTxs.map(tx => tx.getTxHash());
 
       this.isFlushing = false;
-      this.log.info('Collecting attestations');
+      this.log.verbose('Collecting attestations');
+      const stopCollectingAttestationsTimer = this.metrics.startCollectingAttestationsTimer();
       const attestations = await this.collectAttestations(block, txHashes);
-      this.log.info('Attestations collected');
+      this.log.verbose('Attestations collected');
+      stopCollectingAttestationsTimer();
+      this.log.verbose('Collecting proof quotes');
 
-      this.log.info('Collecting proof quotes');
       const proofQuote = await this.createProofClaimForPreviousEpoch(newGlobalVariables.slotNumber.toBigInt());
       this.log.info(proofQuote ? `Using proof quote ${inspect(proofQuote.payload)}` : 'No proof quote available');
 

yarn-project/telemetry-client/src/metrics.ts

@@ -53,6 +53,7 @@ export const SEQUENCER_BLOCK_COUNT = 'aztec.sequencer.block.count';
 export const SEQUENCER_CURRENT_STATE = 'aztec.sequencer.current.state';
 export const SEQUENCER_CURRENT_BLOCK_NUMBER = 'aztec.sequencer.current.block_number';
 export const SEQUENCER_CURRENT_BLOCK_SIZE = 'aztec.sequencer.current.block_size';
+export const SEQUENCER_TIME_TO_COLLECT_ATTESTATIONS = 'aztec.sequencer.time_to_collect_attestations';
 
 export const L1_PUBLISHER_GAS_PRICE = 'aztec.l1_publisher.gas_price';
 export const L1_PUBLISHER_TX_COUNT = 'aztec.l1_publisher.tx_count';

yarn-project/telemetry-client/src/prom_otel_adapter.ts

@@ -28,7 +28,7 @@ interface IGauge<Labels extends LabelsGeneric = NoLabels> {
   set: NoLabels extends Labels ? (value: number) => void : (labels: Labels, value: number) => void;
 
   collect?(): void;
-  addCollect(fn: CollectFn<Labels>): void;
+  addCollect(collectFn: CollectFn<Labels>): void;
 }
 
 interface IHistogram<Labels extends LabelsGeneric = NoLabels> {
@@ -101,8 +101,12 @@ export class OtelGauge<Labels extends LabelsGeneric = NoLabels> implements IGaug
     this.gauge.addCallback(this.handleObservation.bind(this));
   }
 
-  addCollect(fn: CollectFn<Labels>): void {
-    this.collectFns.push(fn);
+  /**
+   * Add a collect callback
+   * @param collectFn - Callback function
+   */
+  addCollect(collectFn: CollectFn<Labels>): void {
+    this.collectFns.push(collectFn);
   }
 
   handleObservation(result: any): void {