Skip to content

testing.md

Latest commit

 

History

History
408 lines (318 loc) · 26.5 KB

testing.md

File metadata and controls

408 lines (318 loc) · 26.5 KB

Testing

Running Tests

make test               # Build and run all suites (1379 tests across 41+ suites at HEAD)
./test_runner                   # Run all tests directly (after building)
./test_runner help              # Print every supported flag

# Single-suite categories
./test_runner basic             # Basic functionality (or: ./test_runner -b)
./test_runner stress            # Stress tests — 100 concurrent clients (or: ./test_runner -s)
./test_runner race              # Race condition tests (or: ./test_runner -r)
./test_runner timeout           # Connection timeout tests (or: ./test_runner -t)
./test_runner config            # Configuration tests (or: ./test_runner -c)
./test_runner http              # HTTP/1.1 internal regressions + integration (or: ./test_runner -H)
./test_runner ws                # WebSocket protocol tests (or: ./test_runner -w)
./test_runner tls               # TLS/SSL tests (or: ./test_runner -T)
./test_runner http2             # HTTP/2 internal regressions + integration (or: ./test_runner -2)
./test_runner cli               # CLI entry point tests (or: ./test_runner -C)
./test_runner route             # Route trie / pattern matching (or: ./test_runner -R)
./test_runner upstream          # Upstream connection pool tests (or: ./test_runner -U)
./test_runner rate_limit        # Rate limit tests (or: ./test_runner -L)
./test_runner kqueue            # macOS kqueue platform tests (or: ./test_runner -K)

# Feature-family umbrellas (each runs every sub-suite in the family)
./test_runner auth              # full auth feature family (or: ./test_runner -A)
./test_runner circuit_breaker   # full circuit-breaker feature family (or: ./test_runner -B)
./test_runner proxy             # full proxy feature family — internal regressions + engine (or: ./test_runner -P)
./test_runner dns               # full DNS / dual-stack feature family (or: ./test_runner -D)
./test_runner dual_stack        # sub-suite — dual-stack integration only (OS-sensitive; macOS CI subset)
./test_runner dns_resolver      # sub-suite — DnsResolver primitives only (timing-sensitive)

# Auth sub-suites (drill into one aspect)
./test_runner auth_foundation
./test_runner jwt               # JWT verifier (or: ./test_runner -J)
./test_runner jwks              # JWKS cache (or: ./test_runner -j)
./test_runner oidc              # OIDC discovery (or: ./test_runner -O)
./test_runner hrauth            # HeaderRewriter auth overlay (or: ./test_runner -W)
./test_runner auth_mgr          # AuthManager unit tests (or: ./test_runner -M)
./test_runner auth2             # Auth integration tests (or: ./test_runner -V)
./test_runner auth_fail         # Auth failure-mode tests (or: ./test_runner -F)
./test_runner auth_reload       # Auth reload tests (or: ./test_runner -X)
./test_runner auth_multi        # Auth multi-issuer tests (or: ./test_runner -I)
./test_runner auth_ws           # Auth WebSocket-upgrade tests (or: ./test_runner -G)
./test_runner auth_race         # Auth race-condition tests (or: ./test_runner -Q)
./test_runner router_async      # Router async-middleware tests (or: ./test_runner -N)
./test_runner introspection_cache  # (or: ./test_runner -Y)
./test_runner intro_client      # Introspection client + AsyncPendingState (or: ./test_runner -y)
./test_runner auth_intro        # Introspection integration (or: ./test_runner -Z)
./test_runner auth_observability  # Debug response headers + per-policy counters (or: ./test_runner -o)

# Make targets — single-suite
make test_basic
make test_stress
make test_race
make test_config
make test_http
make test_ws
make test_tls
make test_http2
make test_cli
make test_upstream
make test_rate_limit

# Make targets — feature-family umbrellas
make test_auth                  # full auth feature family
make test_circuit_breaker       # full circuit-breaker feature family
make test_proxy                 # full proxy feature family
make test_dns                   # full DNS / dual-stack feature family
make test_dual_stack            # sub-suite — dual-stack integration only (OS-sensitive)
make test_dns_resolver          # sub-suite — DnsResolver primitives only

# Make targets — auth sub-suites
make test_auth_foundation
make test_jwt
make test_jwks
make test_oidc
make test_hrauth
make test_auth_mgr
make test_auth2
make test_auth_fail
make test_auth_reload
make test_auth_multi
make test_auth_ws
make test_auth_race
make test_auth_observability

At current head, ./test_runner reports 1379 / 1379 passing (100 %).

Test Suites

Suite Port File Command
Basic ephemeral test/basic_test.h ./test_runner basic
Stress ephemeral test/stress_test.h ./test_runner stress
Race Condition ephemeral test/race_condition_test.h ./test_runner race
Timeout ephemeral test/timeout_test.h ./test_runner timeout
Config N/A test/config_test.h ./test_runner config
HTTP (internal + integration) ephemeral test/http_test.h, test/http_internal_test.h ./test_runner http
WebSocket ephemeral test/websocket_test.h ./test_runner ws
TLS ephemeral test/tls_test.h ./test_runner tls
HTTP/2 (internal + integration) ephemeral test/http2_test.h, test/http2_internal_test.h ./test_runner http2
CLI N/A test/cli_test.h ./test_runner cli
Route ephemeral test/route_test.h ./test_runner route
Upstream Pool ephemeral test/upstream_pool_test.h ./test_runner upstream
Proxy (engine) ephemeral test/proxy_test.h ./test_runner proxy (umbrella runs internal regressions too)
Rate Limit ephemeral test/rate_limit_test.h ./test_runner rate_limit
Circuit Breaker (state machine + window) N/A test/circuit_breaker_test.h ./test_runner circuit_breaker (umbrella)
Circuit Breaker (components) N/A test/circuit_breaker_components_test.h (bundled with circuit_breaker)
Circuit Breaker (integration) ephemeral test/circuit_breaker_integration_test.h (bundled with circuit_breaker)
Circuit Breaker (retry budget) ephemeral test/circuit_breaker_retry_budget_test.h (bundled with circuit_breaker)
Circuit Breaker (wait-queue drain) ephemeral test/circuit_breaker_wait_queue_drain_test.h (bundled with circuit_breaker)
Circuit Breaker (observability) ephemeral test/circuit_breaker_observability_test.h (bundled with circuit_breaker)
Circuit Breaker (reload) ephemeral test/circuit_breaker_reload_test.h (bundled with circuit_breaker)
Auth Foundation N/A test/auth_foundation_test.h ./test_runner auth_foundation (or via auth umbrella)
JWT Verifier N/A test/jwt_verifier_test.h ./test_runner jwt (or via auth umbrella)
JWKS Cache N/A test/jwks_cache_test.h ./test_runner jwks (or via auth umbrella)
OIDC Discovery N/A test/oidc_discovery_test.h ./test_runner oidc (or via auth umbrella)
HeaderRewriter Auth Overlay N/A test/header_rewriter_auth_test.h ./test_runner hrauth (or via auth umbrella)
AuthManager unit N/A test/auth_manager_test.h ./test_runner auth_mgr (or via auth umbrella)
Auth Integration (HTTP end-to-end) ephemeral test/auth_integration_test.h ./test_runner auth2 (or via auth umbrella)
Auth Failure Modes ephemeral test/auth_failure_mode_test.h ./test_runner auth_fail (or via auth umbrella)
Auth Reload ephemeral test/auth_reload_test.h ./test_runner auth_reload (or via auth umbrella)
Auth Multi-Issuer ephemeral test/auth_multi_issuer_test.h ./test_runner auth_multi (or via auth umbrella)
Auth WebSocket Upgrade ephemeral test/auth_websocket_upgrade_test.h ./test_runner auth_ws (or via auth umbrella)
Auth Race Conditions ephemeral test/auth_race_test.h ./test_runner auth_race (or via auth umbrella)
Router Async-Middleware N/A test/router_async_middleware_test.h ./test_runner router_async (or via auth umbrella)
Introspection Cache N/A test/introspection_cache_test.h ./test_runner introspection_cache (or via auth umbrella)
Introspection Client N/A test/introspection_client_test.h ./test_runner intro_client (or via auth umbrella)
Auth Introspection Integration ephemeral test/auth_introspection_integration_test.h ./test_runner auth_intro (or via auth umbrella)
Auth Observability ephemeral test/auth_observability_test.h ./test_runner auth_observability (or via auth umbrella)
Proxy Transaction Internal N/A test/proxy_transaction_internal_test.h (bundled with proxy)
DnsResolver N/A test/dns_resolver_test.h (bundled with dns)
DualStack ephemeral test/dual_stack_test.h (bundled with dns)
Kqueue ephemeral test/kqueue_test.h ./test_runner kqueue (macOS only, skipped on Linux)

Basic Tests

  • Single client connection
  • Echo functionality
  • Multiple sequential connections (5 clients)
  • Concurrent connections (10 clients)
  • Large message transfer (512 bytes)
  • Quick connect/disconnect

Stress Tests

  • High load: 100 concurrent clients sending messages simultaneously

Race Condition Tests

Validates all fixes from the EventFD race condition investigation:

Test What It Validates
RC-1: Dispatcher Initialization Two-phase init, wake channel setup
RC-2: EnQueue No Deadlock Lock-free task execution pattern
RC-3: Double Close Prevention Atomic close guards
RC-4: Concurrent Event Handling Priority-based EPOLLRDHUP handling
RC-5: channel_map_ Race Condition Mutex-protected map access (critical -- prevents segfault)
RC-6: TOCTOU Race epoll_ctl Defense-in-depth fd validation
RC-7: Atomic Closed Flag Atomic bool operations across threads

RC-5 is the most critical test -- it directly prevents the segfault (Channel::HandleEvent(this=0x0)) that triggered the race condition investigation.

Timeout Tests

  • Custom timer configuration
  • Default timer parameters
  • Idle connection detection

HTTP Tests

  • Request parsing and routing
  • Keep-alive and pipelining
  • Middleware chain execution
  • Error responses (400, 404, 405, 413, 417, 505)
  • HEAD method handling
  • Response serialization
  • Async route middleware gating and rejection with headers
  • Async route pipeline ordering on keep-alive connections
  • Async HEAD fallback rewriting method to GET
  • Async 405 includes async methods in Allow header
  • Async HEAD body stripping
  • Async route client Connection: close header handling

WebSocket Tests

  • Handshake validation (RFC 6455)
  • Frame parsing (text, binary, control frames)
  • Fragmentation and reassembly
  • Close handshake with code validation
  • Masking requirement enforcement
  • RSV bit validation
  • UTF-8 validation

TLS Tests

  • Certificate loading and validation
  • TLS 1.2/1.3 minimum version enforcement

HTTP/2 Tests

Configuration (6 tests): Http2Config defaults, JSON parsing, RFC 9113 validation, env overrides, disabled mode, serialization round-trip.

Protocol Detection (8 tests): ALPN h2/http1.1/empty, preface detection, HTTP/1.1 fallback, short data, partial preface, MinDetectionBytes constant.

Stream Unit Tests (7 tests): Pseudo-header mapping (:method, :path, :authority), regular headers lowercase, cookie concatenation (RFC 9113 Section 8.2.3), body accumulation, stream state lifecycle, request completeness, path without query.

H2C Functional (6 tests): Simple GET via cleartext h2c, POST with body, 404 routing, middleware execution, multiple concurrent streams on one connection, large body within limit.

Error Handling (2 tests): Invalid preface (garbage bytes), body exceeding max_body_size (RST_STREAM).

Race Conditions (2 tests): Concurrent HTTP/2 clients, mixed HTTP/1.1 + HTTP/2 clients.

Uses Http2TestClient -- a test helper wrapping nghttp2 in client mode with mem_recv/mem_send for frame-level control.

Upstream Pool Tests

Unit Tests (13 tests): SocketHandler connect-refused behavior, UpstreamConnection state transitions (CONNECTING→READY→IN_USE), expiration by lifetime/request count, IsAlive checks, fd() with null transport, UpstreamLease default/move/release semantics, PoolPartition error code constants, UpstreamHostPool partition-per-dispatcher and accessors.

Integration Tests (11 tests): HasUpstream lookup, checkout from unknown service, shutdown drain, EvictExpired no-crash, CheckoutAsync valid connection, connection reuse (same fd after return), connect failure fires error callback, wait-queue overflow (POOL_EXHAUSTED), upstream drops connection while lease held, multi-dispatcher concurrency.

Uses a real HttpServer instance as the upstream backend with ephemeral ports. Tests validate the full lifecycle: checkout → use → return → reuse, including error paths and graceful shutdown.

Rate Limit Tests

TokenBucket unit (7 tests): Fresh bucket full, lazy refill, capacity limit, UpdateConfig rate/capacity change, low-rate fractional credit preservation, SecondsUntilAvailable accuracy.

RateLimitZone tests (8 tests): Key extractor types (client_ip, header, composite, empty-key skip), applies_to prefix filter with segment-boundary matching, LRU eviction after timer sweep, synchronous max_entries enforcement on insert (no timer dependency).

RateLimitManager tests (11 tests): Single-zone allow/deny, multi-zone all-pass, multi-zone one-denies (first-deny-wins), stops-debiting-after-denial, skips non-applicable zones when building headers, RateLimit/Retry-After header generation, large-policy-window overflow safety, reset-header on empty bucket, disabled short-circuit.

Hot-reload tests (6 tests): Enable/disable toggle, rate change on existing buckets, add/remove zone, status_code and dry_run changes visible to next request.

Integration tests — full HTTP (5 tests): 200 with RateLimit-* headers, 429 with Retry-After, custom status code (503), dry-run allows and strips Retry-After, middleware applies to all routes.

Configuration tests (6 tests): JSON round-trip, validation errors (rate≤0, unknown key_type, duplicate names, enabled+empty zones, empty applies_to entry rejection).

Edge cases (3 tests): Empty client_ip skip, capacity=1, very high rate (1e6) without integer overflow.

Total: 46 tests. Uses ephemeral ports via TestServerRunner<HttpServer> for integration paths.

OAuth 2.0 Auth Tests

Auth coverage is split across 12 suites (41 Phase 1a foundation + 166 Phase 1b/2 — 207 tests total):

  • Foundation (41) — token hashing, claim extraction, policy matcher, config schema / validation / reload-safety. Pure utilities; no server needed.
  • JWT Verifier (21) — jwt-cpp wrapping, algorithm allowlist, signature / exp / nbf / aud / iss failures, alg:none rejection, leeway handling.
  • JWKS Cache (20) — installs + lookups, TTL expiry, kid miss, stale-on-error, coalesced refresh CAS, hard-cap trimming, snapshot stats.
  • OIDC Discovery (14) — .well-known/openid-configuration parsing, retry-on-failure with shared-pointer cancel token, malformed-JSON containment, Cancel() idempotency.
  • HeaderRewriter Auth Overlay (18) — strip-then-inject ordering, claims_to_headers, preserve_authorization, raw_jwt_header opt-in, client-spoof defense on X-Auth-Undetermined, reserved-name enforcement.
  • AuthManager unit (20) — policy matching, bearer extraction, reload-safe snapshots, generation counter, issuer topology stability.
  • Integration (20) — end-to-end through HttpServer + AuthManager middleware with fake IdP: ALLOW / 401 / 403 / 503 paths, multi-issuer routing, on_undetermined: allow.
  • Failure modes (15) — IdP unreachable, JWKS fetch timeout, kid miss with stale-served fallback, malformed token containment.
  • Reload (14) — live-reloadable field propagation, restart-required topology warns, in-flight-request snapshot consistency across reload.
  • Multi-issuer (8) — PeekIssuer routing, allowlist enforcement, cross-issuer claim rejection.
  • WebSocket upgrade (6) — auth enforcement through the HTTP/1.1 upgrade handshake.
  • Race conditions (10) — concurrent kid-miss refreshes, reload-while-verifying, issuer destruction with in-flight OIDC retry.

Kqueue Tests (macOS only)

  • EVFILT_TIMER idle timeout verification
  • EV_EOF detected on write filter (peer close)
  • Pipe wakeup cross-thread signaling
  • Filter consolidation (read + write on same fd)
  • Channel churn stability (rapid add/remove)
  • Timer re-arm after fire
  • SO_NOSIGPIPE on accepted sockets (forked child write test)

Route Tests

  • Compressed radix trie insert/search
  • Wildcard parameter matching (:id, *path)
  • Overlapping routes, edge cases, trailing slashes

Configuration Tests

  • Default values, JSON loading, file I/O
  • Invalid JSON handling, port validation
  • TLS cert validation, environment variable overrides

Port Configuration

All test suites use ephemeral ports (port 0 + getsockname()) via the TestServerRunner<T> RAII harness. This eliminates port conflicts and startup sleeps. The OS assigns an available port, and the test harness communicates it via a promise/future.

Test Framework

Core Components

  • TestFramework (test/test_framework.h/.cc) -- Result tracking with automatic categorization
  • TestServerRunner<T> -- RAII template harness for server thread lifecycle. Works with any server type exposing SetReadyCallback/GetBoundPort/Start/Stop. Uses exception-safe constructor with promise/future synchronization to communicate the ephemeral port back to the test. Automatically stops the server and joins the thread on destruction.
  • TestHttpClient (test/http_test_client.h) -- Shared HTTP test client helpers: raw socket connect, HTTP GET/POST, response parsing, server-close detection

Test Categories

Tests are automatically grouped by category in the results output:

enum class TestCategory { BASIC, STRESS, RACE_CONDITION, TIMEOUT, CONFIG, HTTP, WEBSOCKET, TLS, OTHER };

// Recording a test result with category:
TestFramework::RecordTest("My Test", true, "", TestFramework::TestCategory::BASIC);

Results Output

Results are displayed with per-category statistics:

======================================================================
                    TEST RESULTS SUMMARY
======================================================================

Basic Tests (9/9 passed)
----------------------------------------------------------------------
  [PASS] Single Client Connection
  [PASS] Echo Functionality
  ...

Race Condition Tests (9/9 passed)
----------------------------------------------------------------------
  [PASS] RC-1: Dispatcher Initialization
  ...

======================================================================
OVERALL SUMMARY
----------------------------------------------------------------------
  Basic Tests: 9/9 (100%)
  Stress Tests: 3/3 (100%)
  Race Condition Tests: 9/9 (100%)
  ...
----------------------------------------------------------------------
Total Tests: 196 | Passed: 196 | Failed: 0
Success Rate: 100%
======================================================================

Writing New Tests

  1. Create or update a test header in test/ (e.g., test/my_test.h)
  2. Use TestServerRunner<T> for automatic server lifecycle management with ephemeral ports
  3. Set receive timeouts on all test client sockets (client.SetReceiveTimeout(5))
  4. Record results with appropriate category: TestFramework::RecordTest(name, passed, error, category)
  5. Integrate into test/run_test.cc and update the Makefile

Test Design Patterns

  • RAII TestServerRunner<T>: Automatically stops server and joins thread on destruction; uses promise/future for ephemeral port synchronization
  • Exception-safe recording: Wrap test logic in try/catch, record failure on exception
  • Atomic result tracking: Use std::atomic<int> counters for multi-threaded test result aggregation
  • Timeout on all sockets: Prevents tests from hanging indefinitely on server failure

Continuous Integration

CI workflows live in .github/workflows/ and run in three cadences. Stress and valgrind never gate PRs — both run on cron and are independent of the per-PR matrix.

Stress is also gated out of the per-PR no-arg ./test_runner invocation: RunAllTest() checks getenv("GITHUB_ACTIONS") and skips StressTests::RunStressTests() when set. GitHub Actions sets GITHUB_ACTIONS=true automatically on its runners, so the per-PR matrix runs every other suite but never stress; nightly-stress.yml invokes ./test_runner stress directly (the explicit-flag path bypasses RunAllTest). Local runs and Codespaces (which do NOT set GITHUB_ACTIONS) include stress so developers get full coverage from make test.

Stress runs with continue-on-error: true because high-concurrency suites are legitimately flake-prone on shared runners. Valgrind does NOT use continue-on-error: a memory error caught by valgrind --error-exitcode=1 should surface as a red scheduled run, otherwise the weekly job is just burning runner-minutes.

Per-PR (.github/workflows/ci.yml)

Six parallel jobs gate every PR. Cheap dimensions run all suites; the slow dimension (TSan) is sharded by hand-curated buckets so the critical path stays under ~13 minutes.

Job Runner What it runs
build-linux-gcc ubuntu-latest All suites under gcc, no sanitizers — fastest signal that the build links and the suite passes. (Stress is skipped via the GITHUB_ACTIONS gate; see above.)
build-linux-clang ubuntu-latest All suites under clang. Catches warnings / codegen-driven UB that gcc misses. (Stress is skipped via the GITHUB_ACTIONS gate.)
build-linux-asan ubuntu-latest All suites under AddressSanitizer + UndefinedBehaviorSanitizer. Catches UAF, heap/stack overflows, signed overflow, alignment, null deref. detect_leaks=0 to tolerate test-harness teardown. (Stress is skipped via the GITHUB_ACTIONS gate.)
build-linux-tsan-heavy ubuntu-latest ThreadSanitizer on the two slowest umbrellas: race and auth. (TSan amplifies runtime ~5–10x; isolating these lets the rest finish in parallel.)
build-linux-tsan-rest ubuntu-latest ThreadSanitizer on every other suite enumerated explicitly (basic, http, http2, ws, tls, cli, route, kqueue, upstream, proxy, rate_limit, circuit_breaker, dns, the obs_* family).
build-macos macos-14 OS-sensitive subset only — kqueue, race, timeout, tls, cli, upstream, proxy, http, http2, ws, dual_stack, obs_e2e. Pure-logic suites are platform-deterministic and already covered by the Linux jobs.

The PR matrix uses GitHub Actions concurrency: cancel-in-progress: true, so a follow-up commit on the same branch automatically cancels the in-flight run.

Compile-time caching. Every Linux job and the macOS job wraps the build step in hendrikmuhs/ccache-action. create-symlink: true puts ccache symlinks ahead of gcc / g++ / clang / clang++ on PATH so the Makefile picks them up transparently — no Makefile edit needed. Cache keys are per matrix dimension (ccache-linux-gcc, ccache-linux-clang, ccache-linux-asan, ccache-linux-tsan, ccache-macos-arm64); tsan-heavy and tsan-rest share ccache-linux-tsan since they compile identical TSan binaries. max-size is 300M for non-sanitizer dims and 500M for sanitizer dims (sanitizer instrumentation roughly doubles object size). Total per-PR cache footprint ~1.9 GB across 5 unique keys, well under GitHub's 10 GB per-repo cap. Expected warm-cache build wall-clock: ~30–60s vs ~7–13 min cold.

Nightly cron (.github/workflows/nightly-stress.yml)

Runs at 07:00 UTC (= 00:00 PT). Stress suites are noisy on shared CI runners (200 concurrent clients with an 85% success-rate threshold under CI=true; 1000 clients with 95% threshold locally) — a failure here means "investigate flake," not "investigate code". The workflow uses continue-on-error: true and does not red-X the badge. Each job invokes ./test_runner stress directly, which bypasses the RunAllTest() GITHUB_ACTIONS gate that excludes stress from the per-PR matrix.

  • stress-linux — ubuntu-latest, full stress sweep
  • stress-macos — macos-14, full stress sweep

Weekly cron (.github/workflows/weekly-valgrind.yml)

Runs Sundays at 09:00 UTC. Valgrind catches reads-of-uninitialized-memory and pointer-validity bugs that AddressSanitizer cannot, but its 10–50x runtime overhead makes it unsuitable for PR-blocking CI. The workflow has a 6-hour timeout cap and runs against a curated subset (excludes stress, timeout, race, obs_stress, obs_export — interpreter slowdown collapses their timing assertions). Unlike nightly-stress.yml, this workflow does NOT set continue-on-error: true: a memory error from valgrind --error-exitcode=1 must surface as a red scheduled run so the failure is actionable, not silently swallowed.

Adding a new test suite to CI

When adding a suite to test/run_test.cc::RunAllTest():

  1. The Linux gcc / clang / ASan jobs auto-pick it up (all suites).
  2. Add the new flag to the loop in build-linux-tsan-rest — TSan does not auto-pick-up. If the new suite is heavy (>30s base runtime) or is itself a multi-suite umbrella, add it to build-linux-tsan-heavy instead.
  3. If the suite touches OS-level primitives (sockets, signals, FDs, kqueue, TLS, DNS) add it to the macOS subset in build-macos.
  4. If it's stress-shaped, add it to nightly-stress.yml (invoked via an explicit ./test_runner <flag> step). If the suite has heavy variants you also want to gate out of the per-PR no-arg invocation, mirror the existing stress pattern: gate the call site in RunAllTest() on getenv("GITHUB_ACTIONS") so local runs include it but the GitHub Actions PR matrix skips it.
  5. If it's memory-safety-flavored and not timing-sensitive, add it to the loop in weekly-valgrind.yml.

Thread Pool Tests

The thread_pool/ subproject has its own test suite:

cd thread_pool && make clean && make && \./test_runner

Tests include:

  • Basic execution, exception propagation, cooperative cancellation
  • Stop cancels pending tasks, restartability, start validation
  • High concurrency stress test
  • Lost wakeup regression tests: NoLostWakeupOnShutdown, StopWithIdleThreads, RapidStartStop

The lost wakeup tests use timing validation (Stop must complete in < 1000ms) and multiple iterations to catch race conditions. See design-decisions.md for background.