ci: refactor Cypress sharding due to cloud limits

.github/actions/app-stop-local-env/action.yaml

@@ -1,38 +1,98 @@
 name: "App Stop Local Environment"
 description: "Stops localtest and app containers"
+inputs:
+  collect-diagnostics:
+    description: "Whether to collect local environment diagnostics before stopping it"
+    required: false
+    default: "false"
 runs:
   using: "composite"
   steps:
-    - name: Log localtest diagnostics
-      if: failure()
+    - name: Collect local environment diagnostics
+      if: ${{ inputs.collect-diagnostics == 'true' }}
       shell: bash
       run: |
-        if command -v studioctl > /dev/null 2>&1; then
-          studioctl server status || true
-          studioctl app ps || true
-
-          apps_json="$(mktemp)"
-          if studioctl app ps --json > "$apps_json"; then
-            node -e '
-              const fs = require("fs");
-              const data = JSON.parse(fs.readFileSync(process.argv[1], "utf8"));
-              for (const app of data.apps ?? []) {
-                if (app.id) console.log(app.id);
-              }
-            ' "$apps_json" | while IFS= read -r id; do
-              echo "::group::studioctl app logs --id $id"
-              studioctl app logs --id "$id" --tail 200 --follow=false || true
-              echo "::endgroup::"
-            done || true
+        diagnostics_dir="${GITHUB_WORKSPACE}/src/App/frontend"
+        diagnostics_log="${diagnostics_dir}/local-env-diagnostics.log"
+        localtest_log="${diagnostics_dir}/localtest-container.log"
+        frontend_pid_file="${diagnostics_dir}/frontend-server.pid"
+
+        mkdir -p "$diagnostics_dir"
+        set +e
+        {
+          echo "Collected at: $(date --iso-8601=seconds)"
+          echo "Runner: $(uname -a)"
+
+          run_diagnostic() {
+            echo
+            echo "### $*"
+            timeout 30 "$@"
+            echo "Exit code: $?"
+          }
+
+          run_diagnostic curl -v --max-time 10 --output /dev/null \
+            --write-out $'HTTP %{http_code}; connect %{time_connect}s; first byte %{time_starttransfer}s; total %{time_total}s\n' \
+            http://localhost:8080/
+          run_diagnostic curl -v --max-time 10 --output /dev/null \
+            --write-out $'HTTP %{http_code}; connect %{time_connect}s; first byte %{time_starttransfer}s; total %{time_total}s\n' \
+            http://app-frontend.local.altinn.cloud:8000/
+          run_diagnostic curl -v --max-time 10 --output /dev/null \
+            --write-out $'HTTP %{http_code}; connect %{time_connect}s; first byte %{time_starttransfer}s; total %{time_total}s\n' \
+            http://local.altinn.cloud:8000/ttd/frontend-test
+
+          if [[ -f "$frontend_pid_file" ]]; then
+            frontend_pid="$(<"$frontend_pid_file")"
+            run_diagnostic ps -o pid,ppid,nlwp,rss,stat,lstart,cmd -p "$frontend_pid"
+            run_diagnostic ls -l "/proc/${frontend_pid}/fd"
+            run_diagnostic cat "/proc/${frontend_pid}/limits"
+          else
+            echo
+            echo "Frontend PID file not found: $frontend_pid_file"
           fi
-          rm -f "$apps_json"
 
-          echo "::group::studioctl-server logs"
-          studioctl server logs --tail 200 --follow=false || true
-          echo "::endgroup::"
-        fi
+          run_diagnostic ss -antp
+          run_diagnostic docker ps --no-trunc
+          run_diagnostic docker inspect --format '{{json .State}}' localtest
+          run_diagnostic docker stats --no-stream
+          run_diagnostic df -h
+          run_diagnostic free -h
+          run_diagnostic ps -eLf
+          run_diagnostic sh -c 'printf "Runner thread count: "; ps -eLf --no-headers | wc -l'
+          run_diagnostic sh -c 'printf "Open system file handles: "; cat /proc/sys/fs/file-nr'
+          run_diagnostic sh -c 'printf "PID limit: "; cat /proc/sys/kernel/pid_max'
+
+          if command -v studioctl > /dev/null 2>&1; then
+            run_diagnostic studioctl doctor
+            run_diagnostic studioctl server status
+            run_diagnostic studioctl app ps
+
+            apps_json="$(mktemp)"
+            if timeout 30 studioctl app ps --json > "$apps_json"; then
+              node -e '
+                const fs = require("fs");
+                const data = JSON.parse(fs.readFileSync(process.argv[1], "utf8"));
+                for (const app of data.apps ?? []) {
+                  if (app.id) console.log(app.id);
+                }
+              ' "$apps_json" | while IFS= read -r id; do
+                echo
+                echo "### studioctl app logs --id $id"
+                timeout 30 studioctl app logs --id "$id" --tail 200 --follow=false
+              done
+            fi
+            rm -f "$apps_json"
+
+            echo
+            echo "### studioctl server logs"
+            timeout 30 studioctl server logs --tail 200 --follow=false
+          fi
+        } 2>&1 | tee "$diagnostics_log"
+
+        timeout 30 docker logs --timestamps localtest > "$localtest_log" 2>&1 || true
+        exit 0
 
     - name: Stop localtest
+      if: always()
       shell: bash
       run: |
         if command -v studioctl > /dev/null 2>&1; then

.github/workflows/app-frontend-cypress.yml

@@ -266,6 +266,7 @@ jobs:
         run: npx cypress install
 
       - name: Cypress run
+        id: cypress
         shell: bash
         env:
           CYPRESS_PROJECT_ID: o7mikf
@@ -277,12 +278,13 @@ jobs:
           PERCY_BRANCH: ${{ github.head_ref || github.ref_name }}
           PERCY_TARGET_BRANCH: ${{ github.base_ref || 'main' }}
           IS_PR_DRAFT: ${{ github.event.pull_request.draft }}
+          E2E_SHARD_TOTAL: 6
+          E2E_SHARD_NUMBER: ${{ matrix.containers }}
         run: |
           cypress_args=(
             run
             --record
-            --parallel
-            --group altinn-app-frontend
+            --group "altinn-app-frontend-shard-${{ matrix.containers }}"
             --tag altinn-app-frontend
             --browser "${CHROME_PATH:-chrome}"
             --ci-build-id "${{ github.run_id }}-${{ github.run_attempt }}"
@@ -296,19 +298,24 @@ jobs:
             npx cypress "${cypress_args[@]}"
           fi
 
+      - name: Stop localtest
+        if: always()
+        uses: ./.github/actions/app-stop-local-env
+        with:
+          collect-diagnostics: ${{ steps.cypress.outcome == 'failure' }}
+
       - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
-        if: failure()
+        if: ${{ always() && steps.cypress.outcome == 'failure' }}
         with:
-          name: app-frontend-logs
+          name: app-frontend-logs-${{ matrix.containers }}
           path: |
+            src/App/frontend/frontend-server.log
+            src/App/frontend/local-env-diagnostics.log
+            src/App/frontend/localtest-container.log
             src/App/frontend/test/logs/*
             src/App/frontend/test/screenshots/**
             ${{ env.STUDIOCTL_HOME }}/logs/**
 
-      - name: Stop localtest
-        if: always()
-        uses: ./.github/actions/app-stop-local-env
-
   cypress-run-external:
     name: Cypress Run External
     if: |
@@ -358,19 +365,25 @@ jobs:
         run: npx cypress install
 
       - name: Cypress run
+        id: cypress
         shell: bash
         run: npx cypress run --browser "${CHROME_PATH:-chrome}" --spec test/e2e/integration --env environment=localtest
 
+      - name: Stop localtest
+        if: always()
+        uses: ./.github/actions/app-stop-local-env
+        with:
+          collect-diagnostics: ${{ steps.cypress.outcome == 'failure' }}
+
       - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
-        if: failure()
+        if: ${{ always() && steps.cypress.outcome == 'failure' }}
         with:
           name: app-frontend-external-logs
           path: |
+            src/App/frontend/frontend-server.log
+            src/App/frontend/local-env-diagnostics.log
+            src/App/frontend/localtest-container.log
             src/App/frontend/test/logs/*
             src/App/frontend/test/screenshots/*
             src/App/frontend/test/videos/*
             ${{ env.STUDIOCTL_HOME }}/logs/**
-
-      - name: Stop localtest
-        if: always()
-        uses: ./.github/actions/app-stop-local-env

src/App/frontend/cypress.config.js

@@ -5,6 +5,7 @@ const path = require('node:path');
 const fs = require('node:fs/promises');
 const { existsSync } = require('node:fs');
 const env = require('dotenv').config();
+const configureCypressShard = require('./scripts/cypress-shard');
 
 const CYPRESS_WINDOW_WIDTH = env.parsed?.CYPRESS_WINDOW_WIDTH || 1920;
 const CYPRESS_WINDOW_HEIGHT = env.parsed?.CYPRESS_WINDOW_HEIGHT || 1080;
@@ -80,26 +81,36 @@ module.exports = defineConfig({
 
       const validEnvironments = ['localtest', 'tt02'];
       if (validEnvironments.includes(config.env.environment)) {
-        return getConfigurationByFile(config.env.environment).then((fileConfig) => ({
-          ...fileConfig,
-          env: {
-            ...config.env,
-            ...fileConfig.env,
-          },
-          expose: {
-            ...config.expose,
-            ...fileConfig.expose,
-            // Specs that assert on backend-local date/time values need the backend's timezone.
-            // Only in localtest does the app backend run on the same machine as Cypress, so only
-            // then is the machine timezone valid - read it here in the Node process, since the
-            // browser's timezone may be emulated via CDP and cannot be trusted. Against remote
-            // environments (tt02) this is deliberately left unset; specs fall back to UTC, which
-            // is what those backends run in.
-            ...(config.env.environment === 'localtest'
-              ? { machineTimezone: Intl.DateTimeFormat().resolvedOptions().timeZone }
-              : {}),
-          },
-        }));
+        return getConfigurationByFile(config.env.environment).then((fileConfig) => {
+          const configured = {
+            ...config,
+            ...fileConfig,
+            env: {
+              ...config.env,
+              ...fileConfig.env,
+            },
+            expose: {
+              ...config.expose,
+              ...fileConfig.expose,
+              // Specs that assert on backend-local date/time values need the backend's timezone.
+              // Only in localtest does the app backend run on the same machine as Cypress, so only
+              // then is the machine timezone valid - read it here in the Node process, since the
+              // browser's timezone may be emulated via CDP and cannot be trusted. Against remote
+              // environments (tt02) this is deliberately left unset; specs fall back to UTC, which
+              // is what those backends run in.
+              ...(config.env.environment === 'localtest'
+                ? { machineTimezone: Intl.DateTimeFormat().resolvedOptions().timeZone }
+                : {}),
+            },
+          };
+
+          return configureCypressShard(configured, {
+            specRoot: path.resolve(__dirname, 'test/e2e/integration'),
+            timingsFile: path.resolve(__dirname, 'test/e2e/cypress-timings.json'),
+            total: process.env.E2E_SHARD_TOTAL,
+            number: process.env.E2E_SHARD_NUMBER,
+          });
+        });
       }
 
       throw new Error(`Unknown environment "${config.env.environment}"

src/App/frontend/scripts/cypress-shard.js

@@ -0,0 +1,85 @@
+/* eslint-disable @typescript-eslint/no-require-imports, no-console */
+const fs = require('node:fs');
+const path = require('node:path');
+
+function configureCypressShard(config, { specRoot, timingsFile, total, number }) {
+  if (total === undefined && number === undefined) {
+    return config;
+  }
+
+  const shardTotal = parseInteger(total, 'E2E_SHARD_TOTAL');
+  const shardNumber = parseInteger(number, 'E2E_SHARD_NUMBER');
+  const shardIndex = shardNumber - 1;
+  if (shardTotal < 1) {
+    throw new Error('E2E_SHARD_TOTAL must be greater than zero.');
+  }
+  if (shardNumber < 1 || shardNumber > shardTotal) {
+    throw new Error(`E2E_SHARD_NUMBER must be between 1 and ${shardTotal}.`);
+  }
+
+  const specs = findSpecs(specRoot);
+  if (shardTotal > specs.length) {
+    throw new Error(`Cannot split ${specs.length} Cypress specs across ${shardTotal} shards.`);
+  }
+
+  const timingData = JSON.parse(fs.readFileSync(timingsFile, 'utf8')).durations;
+  const durations = new Map(timingData.map(({ spec, duration }) => [spec, duration]));
+  const averageDuration = timingData.reduce((sum, { duration }) => sum + duration, 0) / timingData.length;
+  const shards = Array.from({ length: shardTotal }, () => ({ duration: 0, specs: [] }));
+  let specsWithoutTimings = 0;
+
+  const weightedSpecs = specs
+    .map((spec) => {
+      const relativeSpec = toPosixPath(path.relative(config.projectRoot, spec));
+      const duration = durations.get(relativeSpec);
+      if (duration === undefined) {
+        specsWithoutTimings += 1;
+      }
+      return { spec, duration: duration ?? averageDuration };
+    })
+    .sort((left, right) => right.duration - left.duration || left.spec.localeCompare(right.spec));
+
+  for (const spec of weightedSpecs) {
+    const shard = shards.reduce((shortest, candidate) =>
+      candidate.duration < shortest.duration ? candidate : shortest,
+    );
+    shard.specs.push(spec.spec);
+    shard.duration += spec.duration;
+  }
+
+  const selectedShard = shards[shardIndex];
+  config.specPattern = selectedShard.specs.sort();
+  const missingTimingMessage =
+    specsWithoutTimings > 0
+      ? ` (${specsWithoutTimings} ${specsWithoutTimings === 1 ? 'spec uses' : 'specs use'} the historical average)`
+      : '';
+  console.log(
+    `Cypress shard ${shardNumber}/${shardTotal}: ${selectedShard.specs.length} specs, ${Math.round(selectedShard.duration / 1000)} estimated seconds${missingTimingMessage}`,
+  );
+
+  return config;
+}
+
+function findSpecs(directory) {
+  return fs
+    .readdirSync(directory, { withFileTypes: true })
+    .flatMap((entry) => {
+      const entryPath = path.join(directory, entry.name);
+      return entry.isDirectory() ? findSpecs(entryPath) : entryPath.endsWith('.ts') ? [entryPath] : [];
+    })
+    .sort();
+}
+
+function parseInteger(value, name) {
+  const parsed = Number(value);
+  if (!Number.isInteger(parsed)) {
+    throw new Error(`${name} must be an integer.`);
+  }
+  return parsed;
+}
+
+function toPosixPath(value) {
+  return value.split(path.sep).join('/');
+}
+
+module.exports = configureCypressShard;