GitHub #461: TD-S056: bound every in-memory cache and every metric label-set cardinality (#124)

durable-workflow-ops · durable-workflow-ops · commit e8be6a07790c · 2026-05-08T02:17:14.000Z
diff --git a/.github/workflows/server-perf.yml b/.github/workflows/server-perf.yml
@@ -1,7 +1,77 @@
 name: Server Perf
 
 on:
+  pull_request:
+    branches: [main]
+    paths:
+      - ".github/workflows/server-perf.yml"
+      - "Dockerfile"
+      - "docker-compose.yml"
+      - "app/Support/HistoryRetentionEnforcer.php"
+      - "app/Support/BoundedMetricPolicy.php"
+      - "app/Support/LongPollSignalStore.php"
+      - "app/Support/ProjectionDriftMetrics.php"
+      - "app/Support/ServerReadiness.php"
+      - "app/Support/ServerPollingCache.php"
+      - "app/Support/TaskQueueAdmission.php"
+      - "app/Support/WorkflowQueryTaskBroker.php"
+      - "app/Support/WorkflowTaskFailureMetrics.php"
+      - "app/Support/WorkflowTaskPoller.php"
+      - "app/Support/WorkflowTaskPollRequestStore.php"
+      - "app/Http/Controllers/Api/SystemController.php"
+      - "config/dw-bounded-growth.php"
+      - "config/server.php"
+      - "docs/bounded-growth.md"
+      - "routes/api.php"
+      - "scripts/perf/**"
+      - "tests/Unit/BoundedGrowthPolicyTest.php"
+      - "tests/Unit/BoundedMetricPolicyTest.php"
+      - "tests/Unit/ServerPerfHarnessContractTest.php"
+      - "tests/Feature/SystemMetricsTest.php"
+  push:
+    branches: [main]
+    paths:
+      - ".github/workflows/server-perf.yml"
+      - "Dockerfile"
+      - "docker-compose.yml"
+      - "app/Support/HistoryRetentionEnforcer.php"
+      - "app/Support/BoundedMetricPolicy.php"
+      - "app/Support/LongPollSignalStore.php"
+      - "app/Support/ProjectionDriftMetrics.php"
+      - "app/Support/ServerReadiness.php"
+      - "app/Support/ServerPollingCache.php"
+      - "app/Support/TaskQueueAdmission.php"
+      - "app/Support/WorkflowQueryTaskBroker.php"
+      - "app/Support/WorkflowTaskFailureMetrics.php"
+      - "app/Support/WorkflowTaskPoller.php"
+      - "app/Support/WorkflowTaskPollRequestStore.php"
+      - "app/Http/Controllers/Api/SystemController.php"
+      - "config/dw-bounded-growth.php"
+      - "config/server.php"
+      - "docs/bounded-growth.md"
+      - "routes/api.php"
+      - "scripts/perf/**"
+      - "tests/Unit/BoundedGrowthPolicyTest.php"
+      - "tests/Unit/BoundedMetricPolicyTest.php"
+      - "tests/Unit/ServerPerfHarnessContractTest.php"
+      - "tests/Feature/SystemMetricsTest.php"
+  schedule:
+    - cron: "17 7 * * *"
   workflow_dispatch:
+    inputs:
+      duration_seconds:
+        description: "Soak duration in seconds"
+        required: false
+        default: "7200"
+      concurrency:
+        description: "Concurrent long-poll workers"
+        required: false
+        default: "24"
+      remote_write:
+        description: "Enable Prometheus remote_write when variables/secrets are configured"
+        required: false
+        type: boolean
+        default: true
 
 permissions:
   contents: read
@@ -13,36 +83,29 @@ concurrency:
 jobs:
   contract:
     name: Bounded-growth contract
-    if: github.event_name == 'pull_request' || github.event_name == 'push'
     runs-on: ubuntu-latest
+    if: github.event_name == 'pull_request' || github.event_name == 'push'
     timeout-minutes: 20
 
     steps:
       - name: Checkout server
         uses: actions/checkout@v6
 
       - name: Checkout workflow package
-        uses: actions/checkout@v6
-        with:
-          repository: durable-workflow/workflow
-          ref: v2
-          path: workflow-package
+        run: git clone --depth=1 --branch v2 https://github.com/durable-workflow/workflow.git workflow-package
 
       - name: Run bounded-growth contract tests
         run: |
-          docker run --rm \
-            -u "$(id -u):$(id -g)" \
-            -v "${PWD}:/app" \
-            -v "${PWD}/workflow-package:/workflow:ro" \
-            -w /app \
-            composer:2 \
-            sh -lc 'composer install --no-interaction --no-progress --prefer-dist && vendor/bin/phpunit tests/Unit/BoundedGrowthPolicyTest.php tests/Unit/ServerPerfHarnessContractTest.php --colors=never'
+          # Stream sources into the test container so containerized runners avoid host bind-path assumptions.
+          tar --exclude=.git --exclude=vendor --exclude=build -cf - . \
+            | docker run --rm -i -w /app composer:2 \
+                sh -lc 'tar -xf - -C /app && cp -a /app/workflow-package /workflow && composer install --no-interaction --no-progress --prefer-dist && vendor/bin/phpunit tests/Unit/BoundedGrowthPolicyTest.php tests/Unit/ServerPerfHarnessContractTest.php --colors=never'
 
   smoke:
     name: Polling cache bounded-growth smoke
-    if: github.event_name == 'pull_request' || github.event_name == 'push'
     needs: contract
     runs-on: ubuntu-latest
+    if: github.event_name == 'pull_request' || github.event_name == 'push'
     timeout-minutes: 45
 
     steps:
@@ -61,13 +124,22 @@ jobs:
           DW_PERF_MAX_SERVER_MEMORY_MB: "768"
           DW_PERF_MAX_POLLING_KEYS: "512"
           DW_PERF_MAX_FINAL_POLLING_KEYS: "0"
+          DW_PERF_MIN_SAMPLE_COVERAGE: "0.75"
           DW_PERF_MAX_SERVER_CACHE_KEYS_BY_POLICY: '{"workflow_task_poll_requests":512,"long_poll_signals":512,"workflow_query_tasks":64,"task_queue_admission_locks":128,"task_queue_dispatch_counters":128,"workflow_task_expired_lease_recovery":128,"history_retention_inline":64,"readiness_probe":16}'
           DW_PERF_MAX_FINAL_SERVER_CACHE_KEYS_BY_POLICY: '{"workflow_task_poll_requests":0,"long_poll_signals":0,"workflow_query_tasks":0,"task_queue_admission_locks":0,"task_queue_dispatch_counters":0,"workflow_task_expired_lease_recovery":0,"history_retention_inline":0,"readiness_probe":0}'
           RUNNER_ENVIRONMENT: "github-hosted"
         run: scripts/perf/run-server-soak.sh
 
       - name: Upload perf artifacts
-        if: always()
+        if: always() && github.server_url == 'https://github.com'
+        uses: actions/upload-artifact@v4
+        with:
+          name: server-perf-smoke
+          path: build/perf/
+          if-no-files-found: warn
+
+      - name: Upload perf artifacts on compatible Actions servers
+        if: always() && github.server_url != 'https://github.com'
         uses: actions/upload-artifact@v3.2.2
         with:
           name: server-perf-smoke
@@ -76,8 +148,8 @@ jobs:
 
   soak:
     name: Self-hosted polling cache soak
-    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
     runs-on: [self-hosted, linux, x64, perf-soak, server-perf]
+    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
     timeout-minutes: 390
 
     steps:
@@ -105,7 +177,15 @@ jobs:
         run: scripts/perf/run-server-soak.sh
 
       - name: Upload perf artifacts
-        if: always()
+        if: always() && github.server_url == 'https://github.com'
+        uses: actions/upload-artifact@v4
+        with:
+          name: server-perf-soak
+          path: build/perf/
+          if-no-files-found: warn
+
+      - name: Upload perf artifacts on compatible Actions servers
+        if: always() && github.server_url != 'https://github.com'
         uses: actions/upload-artifact@v3.2.2
         with:
           name: server-perf-soak
diff --git a/scripts/perf/run-server-soak.sh b/scripts/perf/run-server-soak.sh
@@ -130,6 +130,7 @@ YAML
 
 server_base_url() {
   local base_url="http://127.0.0.1:${SERVER_PORT}"
+  local docker_internal_url="http://host.docker.internal:${SERVER_PORT}"
   local docker_host_url
   local docker_host_ip
   local server_id
@@ -140,6 +141,11 @@ server_base_url() {
     return
   fi
 
+  if curl -fsS --max-time 2 "$docker_internal_url/api/health" >/dev/null 2>&1; then
+    echo "$docker_internal_url"
+    return
+  fi
+
   docker_host_ip="$(ip route 2>/dev/null | awk '/default/ {print $3; exit}')"
   if [ -n "$docker_host_ip" ]; then
     docker_host_url="http://${docker_host_ip}:${SERVER_PORT}"
@@ -153,8 +159,11 @@ server_base_url() {
   server_ip="$(docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' "$server_id" 2>/dev/null || true)"
 
   if [ -n "$server_ip" ]; then
-    echo "http://${server_ip}:8080"
-    return
+    local server_container_url="http://${server_ip}:8080"
+    if curl -fsS --max-time 2 "$server_container_url/api/health" >/dev/null 2>&1; then
+      echo "$server_container_url"
+      return
+    fi
   fi
 
   echo "$base_url"
diff --git a/scripts/perf/server_soak.py b/scripts/perf/server_soak.py
@@ -24,8 +24,8 @@
 from typing import Any
 
 
-CONTROL_PLANE_VERSION = "2"
-WORKER_PROTOCOL_VERSION = "1.0"
+CONTROL_PLANE_VERSION = os.environ.get("DW_PERF_CONTROL_PLANE_VERSION", "2")
+WORKER_PROTOCOL_VERSION = os.environ.get("DW_PERF_WORKER_PROTOCOL_VERSION", "1.2")
 ERROR_WRITE_LOCK = threading.Lock()
 SERVER_CACHE_KEY_PATTERNS = {
     "long_poll_signals": "*server:long-poll-signal:*",
diff --git a/tests/Unit/ServerPerfHarnessContractTest.php b/tests/Unit/ServerPerfHarnessContractTest.php
@@ -187,6 +187,30 @@ public function test_ci_perf_jobs_set_runner_environment_provenance(): void
         );
     }
 
+    public function test_server_perf_jobs_keep_event_split_guards(): void
+    {
+        $workflow = file_get_contents(dirname(__DIR__, 2).'/.github/workflows/server-perf.yml');
+        $this->assertNotFalse($workflow, '.github/workflows/server-perf.yml must be readable');
+
+        $this->assertMatchesRegularExpression(
+            "/contract:\\s+name:\\s+Bounded-growth contract\\s+runs-on:\\s+ubuntu-latest\\s+if:\\s+github\\.event_name == 'pull_request' \\|\\| github\\.event_name == 'push'/s",
+            $workflow,
+            'Contract checks should only run for pull_request/push events, with runs-on before if for runner compatibility.',
+        );
+
+        $this->assertMatchesRegularExpression(
+            "/smoke:\\s+name:\\s+Polling cache bounded-growth smoke\\s+needs:\\s+contract\\s+runs-on:\\s+ubuntu-latest\\s+if:\\s+github\\.event_name == 'pull_request' \\|\\| github\\.event_name == 'push'/s",
+            $workflow,
+            'Short perf smokes should only run for pull_request/push events.',
+        );
+
+        $this->assertMatchesRegularExpression(
+            "/soak:\\s+name:\\s+Self-hosted polling cache soak\\s+runs-on:\\s+\\[self-hosted, linux, x64, perf-soak, server-perf\\]\\s+if:\\s+github\\.event_name == 'schedule' \\|\\| github\\.event_name == 'workflow_dispatch'/s",
+            $workflow,
+            'Trusted long soaks should only run for schedule/workflow_dispatch events.',
+        );
+    }
+
     public function test_self_hosted_perf_soak_requires_trusted_evidence_eligibility(): void
     {
         $workflow = file_get_contents(dirname(__DIR__, 2).'/.github/workflows/server-perf.yml');
@@ -211,6 +235,98 @@ public function test_self_hosted_perf_soak_requires_trusted_evidence_eligibility
         );
     }
 
+    public function test_server_perf_artifact_uploads_use_github_current_action_with_compatible_fallback(): void
+    {
+        $workflow = file_get_contents(dirname(__DIR__, 2).'/.github/workflows/server-perf.yml');
+        $this->assertNotFalse($workflow, '.github/workflows/server-perf.yml must be readable');
+
+        $this->assertSame(2, substr_count($workflow, 'uses: actions/upload-artifact@v4'));
+        $this->assertSame(2, substr_count($workflow, 'uses: actions/upload-artifact@v3.2.2'));
+        $this->assertSame(2, substr_count($workflow, "github.server_url == 'https://github.com'"));
+        $this->assertSame(2, substr_count($workflow, "github.server_url != 'https://github.com'"));
+    }
+
+    public function test_server_perf_soak_uses_current_worker_protocol_default(): void
+    {
+        $source = file_get_contents(dirname(__DIR__, 2).'/scripts/perf/server_soak.py');
+        $this->assertNotFalse($source, 'scripts/perf/server_soak.py must be readable');
+
+        $this->assertStringContainsString(
+            'WORKER_PROTOCOL_VERSION = os.environ.get("DW_PERF_WORKER_PROTOCOL_VERSION", "1.2")',
+            $source,
+        );
+        $this->assertStringContainsString(
+            'headers["X-Durable-Workflow-Protocol-Version"] = WORKER_PROTOCOL_VERSION',
+            $source,
+        );
+        $this->assertStringNotContainsString('WORKER_PROTOCOL_VERSION = "1.0"', $source);
+    }
+
+    public function test_short_perf_smoke_keeps_flake_resistant_sample_coverage_floor(): void
+    {
+        $workflow = file_get_contents(dirname(__DIR__, 2).'/.github/workflows/server-perf.yml');
+        $this->assertNotFalse($workflow, '.github/workflows/server-perf.yml must be readable');
+
+        $this->assertMatchesRegularExpression(
+            '/name:\s+Polling cache bounded-growth smoke(?P<block>.*?)\n\s+soak:/s',
+            $workflow,
+            'Server Perf workflow must keep a distinct short smoke job before the long soak job.',
+        );
+        preg_match('/name:\s+Polling cache bounded-growth smoke(?P<block>.*?)\n\s+soak:/s', $workflow, $smokeMatch);
+
+        $this->assertStringContainsString(
+            'DW_PERF_MIN_SAMPLE_COVERAGE: "0.75"',
+            (string) ($smokeMatch['block'] ?? ''),
+            'Short perf smokes should tolerate one slow compose-backed sample without losing coverage signal.',
+        );
+    }
+
+    public function test_server_perf_base_url_probe_supports_containerized_actions_runners(): void
+    {
+        $source = file_get_contents(dirname(__DIR__, 2).'/scripts/perf/run-server-soak.sh');
+        $this->assertNotFalse($source, 'scripts/perf/run-server-soak.sh must be readable');
+
+        $this->assertStringContainsString('http://host.docker.internal:${SERVER_PORT}', $source);
+        $this->assertMatchesRegularExpression(
+            '/host\.docker\.internal.*?docker_host_ip.*?docker inspect/s',
+            $source,
+            'Perf smoke should prefer host-published ports before falling back to direct container addresses.',
+        );
+        $this->assertMatchesRegularExpression(
+            '/server_container_url="http:\/\/\$\{server_ip\}:8080".*?curl -fsS --max-time 2 "\$server_container_url\/api\/health"/s',
+            $source,
+            'Perf smoke should only select a direct container address after confirming the runner can reach it.',
+        );
+    }
+
+    public function test_server_perf_workflow_can_produce_trusted_long_soak_evidence(): void
+    {
+        $workflow = file_get_contents(dirname(__DIR__, 2).'/.github/workflows/server-perf.yml');
+        $this->assertNotFalse($workflow, '.github/workflows/server-perf.yml must be readable');
+
+        foreach ([
+            'schedule:',
+            'cron: "17 7 * * *"',
+            'workflow_dispatch:',
+            'duration_seconds:',
+            'default: "7200"',
+            'concurrency:',
+            'default: "24"',
+            'remote_write:',
+            'type: boolean',
+            "github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'",
+            'runs-on: [self-hosted, linux, x64, perf-soak, server-perf]',
+            'DW_PERF_REQUIRE_TRUSTED_EVIDENCE: "true"',
+            'RUNNER_ENVIRONMENT: "self-hosted"',
+        ] as $needle) {
+            $this->assertStringContainsString(
+                $needle,
+                $workflow,
+                "Server Perf workflow must retain trusted long-soak trigger support for {$needle}.",
+            );
+        }
+    }
+
     public function test_ci_perf_trigger_paths_cover_bounded_growth_runtime_surfaces(): void
     {
         $repoRoot = dirname(__DIR__, 2);
