Download the PHP package crazy-goat/workerman-bundle without Composer

Workerman runtime for Symfony applications

Workerman is a high-performance, asynchronous event-driven PHP framework written in pure PHP.
This bundle provides a Workerman integration in Symfony, allowing you to easily create an HTTP server, scheduler and supervisor all in one place. This bundle allows you to replace a traditional web application stack like php-fpm + nginx + cron + supervisord, all written in pure PHP (no Go, no external binaries). The request handler works in an event loop, which means the Symfony kernel and the dependency injection container are preserved between requests, making your application faster with fewer (or no) code changes.

Contributing

Please see CONTRIBUTING.md for information about branch protection rules and development workflow.

What's new in this fork

This section documents the differences between crazy-goat/workerman-bundle (this fork) and the upstream luzrain/workerman-bundle.

Dependencies & Compatibility

Features

1. Middleware system — composable request/response pipeline with MiddlewareInterface, MiddlewareDispatchInterface, StaticFilesMiddleware (ETag, Last-Modified, 304 support, blocked extensions, dot-file blocking, symlink control, path traversal protection, LRU realpath cache with TTL, PHAR-aware path resolution), SymfonyController (kernel boot, request conversion, response, termination, service resetter), and a zero per-request allocation pipeline built once and cached.

2. Console commands — full server lifecycle management via ServerManager: workerman:server start/stop/restart/reload/status/connections, plus workerman:build:phar and workerman:build:bin for packaging.

3. Slowloris / DoS protection — configurable connection_timeout for incomplete requests (default: 120s), keepalive_timeout for idle connections (default: 30s), and per-server body_size_cap.

4. Response conversion with strategy pattern — BinaryFileResponseStrategy (uses Workerman's withFile(), supports SplTempFileObject, offset/maxlen, deleteFileAfterSend cleanup), StreamedResponseStrategy (chunked transfer encoding), DefaultResponseStrategy (large responses via chunked transfer directly to connection), header name normalization with caching. Upstream buffers everything in memory.

5. Memory reload strategy — reloads the worker when emalloc'ed memory exceeds limit (default 128 MB); a gc_collect_cycles() is attempted once memory passes gc_limit (default 96 MB) — synchronously, before the reload decision, whenever the worker is also above limit (so a collection that frees enough memory avoids the reload), and deferred to the next event-loop tick otherwise; gc_cooldown (default 60s) limits collection frequency.

6. Trusted hosts — trusted_hosts config key with regex patterns, rejects non-matching Host header via SuspiciousOperationException (400).

7. Service state resetter integration — calls services_resetter after kernel termination to reset stateful services between requests (critical for long-running worker correctness).

8. SSL validation — validates cert/key paths, rejects symlinks for security.

9. Process inspection — /proc-based zombie detection, orphan killing, parent PID tracking.

10. PHAR/BIN runtime support — runtime_dir config key with WORKERMAN_RUNTIME_DIR env var, PharHelper for runtime path resolution, automatic runtime directory creation, skips file monitor in PHAR mode, KernelFactory with PHAR-aware getCacheDir()/getLogDir().

11. Custom exception hierarchy — 20 exception classes (plus 2 marker interfaces) under WorkermanExceptionInterface → WorkermanException → category bases (ServerException, KernelException, MiddlewareException, SchedulerException, ValidationException) with specific exceptions for the bundle's error cases. Upstream uses only generic PHP exceptions.

12. Utils::reload() — programmatic worker reload from application code with reloadAllWorkers: true param.

13. File upload validation — structural validation of uploaded files with clear error messages. Upstream has no validation.

14. Extended Request class — adds setHeader() / withHeader() methods to Workerman's Request, required by the middleware system.

15. ListenScheme enum — type-safe HTTP/HTTPS/WS/WSS scheme parsing. Upstream uses inline str_starts_with() checks.

16. Trigger factory improvement — uses CronExpression::isValidExpression() for proper cron detection. Upstream uses a fragile heuristic (count(explode(' ', $expr)) === 5 && str_contains($expr, '*')).

17. SchedulerWorker improvements — proper SIGCHLD handler that reaps children and logs exit codes/signals, file-lock-based PID management with symlink detection and inode mismatch protection. Upstream uses SIG_IGN for SIGCHLD and simple file_put_contents for PID.

18. SupervisorWorker improvements — handler returns never type (process exits with code 1 on unexpected return), logs unexpected finish, skips processes with processes <= 0.

19. Cache warmup improvements — signal-based success/failure detection (SIGKILL=success, SIGTERM=failure), configurable timeout via WORKERMAN_CACHE_WARMUP_TIMEOUT env var. Upstream uses simple pcntl_wait() with no timeout or error detection.

20. Config loader improvements — ConfigSection enum, warmUp() validates all sections before writing, setBuildConfig() / getBuildConfig() for PHAR build config.

Code quality / DX

• Full custom exception hierarchy (20 classes plus 2 marker interfaces) instead of generic \Exception
• readonly classes where appropriate
• Extracted ConfigurationTreeBuilder, ServicesConfigurator, WorkermanCompilerPass as separate testable classes (upstream uses anonymous closures/files)
• ServerManager extracted as standalone service (testable)
• ServiceMethod value object with validation (upstream uses raw strings)
• ServiceHandlerTrait / ServiceErrorListenerTrait for shared logic
• Extensive test suite (unit + integration + e2e)
• PHPStan + Rector in CI pipeline

What was removed

• PSR-7 pipeline: WorkermanHttpMessageFactory, psr/http-factory, psr/http-message, symfony/psr-http-message-bridge dependencies — replaced by direct Workerman→Symfony conversion.

Getting started

Install composer packages

Enable the bundle

Configure the bundle

A minimal configuration might look like this.
For all available options with documentation, see the command output.

Note: The example above binds an unprivileged port (8080) so it works without sudo.

To bind a port below 1024 (e.g. 80 or 443) you must run the process as root or grant the CAP_NET_BIND_SERVICE capability on Linux.

In production, consider using the user and group config keys to drop privileges after binding, or front it with a reverse proxy (e.g. nginx, Caddy).

Note: listen is effectively required. Omitting it creates a worker that does not accept connections — no traffic reaches your application. Supported URI schemes: http://, https://, ws:// (WebSocket), wss:// (WebSocket over SSL). https:// and wss:// listeners additionally require local_cert and local_pk — see the TLS example.

Configuration reference

All top-level workerman configuration options:

servers[] options

Each entry of servers is a server definition:

The deprecated serve_files/root_dir path reads one sub-key from static_files:

reload_strategy options

reload_strategy configures five worker restart strategies. All strategies can be combined; each has an active switch:

Start application

Using the Symfony console command (the bin/console below refers to your application's Symfony console, not the bin/ directory shipped by this bundle):

Note: All bin/console workerman:* commands throughout this document refer to your application's Symfony console, not the scripts in this bundle's bin/ directory. See bin/README.md for the bundle's own development scripts.

Or using the runtime directly:

Config cache and runtime user

Since 0.25.0 the bundle refuses to load a configuration cache file that is not owned by the process that loads it: the cache file ({cacheDir}/workerman/config.cache.php, in Symfony's kernel cache directory, e.g. var/cache/prod) is a PHP file that is required at boot, so a file replaced by another user would be owned by that user. The check is enforced in the launcher process, before any worker forks — an ownership mismatch aborts workerman:server start with a RuntimeException, not a warning. See Config Cache File Protection for the full threat model.

Note: Deployments that explicitly trust the cache directory but cannot change who warms it (managed build systems, sudoless image builders, frozen base images) can opt out with the documented env var WORKERMAN_TRUST_UNSAFE_CONFIG_CACHE=1, which downgrades the refusals to warnings. This is a security downgrade — the cache file is executed PHP — and must not be enabled for directories untrusted users can write. Strict mode remains the default. See Guard downgrade.

Note: This is the upgrade-relevant change in 0.25.0. The most common containerised layout trips it: the cache is warmed at image build time (as root), the server runs as a non-root user.

Note: 0.25.0 also hardens master-process identification for stop / reload / status: with a server started by an older version still running, those commands may report "Cannot verify master process \<pid>" (the PID is alive but its identity cannot be confirmed) or "Workerman is not running" (no pid file or the process is dead). Stop the server before upgrading — see Upgrading to 0.25.

The error message names both UIDs and suggests the fix:

Warm up with the runtime user (recommended):

(COPY --chown makes www-data the owner of /app, so the runtime user can write var/cache during warm-up — a plain COPY . /app creates root-owned files that www-data cannot overwrite.)

Or re-own the cache file after warm-up:

The same applies to any deploy-user/runtime-user split (deploy scripts, CI runs, sudo): the user that warms the cache must be the user that starts workerman:server, or the cache must be re-owned by that user in between.

Manage the server

workerman:server connections output

The command lists every active TCP connection across all worker processes. Example output:

Platform note: Connection introspection relies on Workerman's internal tracking and is available on POSIX-compatible platforms (Linux, macOS). The output is generated by sending SIGIO to the master process, which collects data from each worker. Windows is not supported because the command uses posix_kill().

Note: For better performance, Workerman recommends installing the php-event extension.

Note: If you have the grpc PHP extension installed, you must set the environment variable GRPC_ENABLE_FORK_SUPPORT=1 before starting the server. The grpc extension spawns background threads that deadlock in forked child processes (e.g. scheduler tasks) unless fork support is explicitly enabled. See grpc/grpc#31241 for details. On grpc hosts the bundle additionally terminates supervised-process and task children with SIGKILL instead of exit(), because grpc_shutdown() can hang in forked children — see docs/troubleshooting.md.

Programmatic reload

You can trigger a worker reload from your application code using Utils::reload():

This sends a SIGUSR1 signal to the worker (or parent) process. It is equivalent to running bin/console workerman:server reload but can be called from any context — controllers, services, scheduled tasks, or deploy hooks.

Note: Utils::reload() requires the pcntl and posix PHP extensions. Both are always available in the Workerman runtime process.

Reload strategies

Because of the asynchronous nature of the server, the workers reuse loaded resources on each request. This means that in some cases we need to restart workers.
For example, after an exception is thrown, to prevent services from being in an unrecoverable state. Or every time you change the code in the IDE.
There are a few restart strategies that are implemented and can be enabled or disabled depending on the environment.

• exception
  Reload worker each time an exception is thrown during the request handling.
• max_requests
  Reload worker on every Nth request to prevent memory leaks.
• file_monitor
  Reload all workers each time you change the files.
• always
  Reload worker after each request.
• memory
  Reload worker when memory usage reaches a certain threshold. Four options are available: active (default: false) toggles the strategy, limit (default: 134217728 — 128 MB) is the memory threshold in bytes that triggers a worker reload, gc_limit (default: 100663296 — 96 MB) attempts gc_collect_cycles() to free memory before the reload check, and gc_cooldown (default: 60) is the minimum interval in seconds between two collection attempts. Memory is measured with memory_get_usage() (emalloc accounting), not memory_get_usage(true) (real usage): emalloc accounting drops as soon as collectable cycles are freed, whereas the allocator arena behind real usage does not shrink from gc_collect_cycles() — with real usage the post-collection re-check could never avoid a reload. When the worker is at risk of reloading (already above limit), the collection runs synchronously so the reload verdict is based on the post-collection reading; otherwise it is deferred to the next event-loop tick to keep the request path short. A collection blocked by the cooldown leaves the reload verdict on the current memory reading.

Note: It is highly recommended to install the php-inotify extension for file monitoring. Without it, monitoring will work in polling mode, which can be very CPU and disk intensive for large projects.

See all available options for each strategy in the command output.

Implement your own reload strategies

You can create a reload strategy with your own logic by implementing the RebootStrategyInterface and adding the workerman.reboot_strategy tag to the service.

Middlewares

Middlewares allow you to intercept and process requests before they reach the Symfony controller, or modify responses before they are sent to the client.

A middleware is any service implementing CrazyGoat\WorkermanBundle\Middleware\MiddlewareInterface:

Registering middlewares

Register your middleware as a service in the Symfony container, then reference its service ID under workerman.servers[].middlewares:

Static files middleware

The deprecated serve_files and root_dir server options are replaced by the StaticFilesMiddleware. To serve static files from a public directory, register the middleware with the root directory path:

The StaticFilesMiddleware resolves requests against the configured root directory, serves matching files directly, and passes through to the next handler for non-file requests. Directory traversal attacks are prevented by ensuring the resolved path stays within the root directory.

Note: The middleware's hardening is configured via constructor arguments — $allowedExtensions for the extension allowlist and $followSymlinks for symlink handling. The static_files server key (including allowed_extensions) only applies to the deprecated serve_files/root_dir path and has no effect on a service-registered middleware. See Security: Static Files Protection.

Execution order

Middlewares are executed in registration order (first registered, first executed). This means the first middleware in the middlewares list wraps the innermost layer. Using onion model terminology:

This allows outer middlewares to handle cross-cutting concerns (authentication, logging, rate limiting) before inner middlewares or the Symfony controller processes the request.

Scheduler

Periodic tasks can be configured with attributes or with tags in configuration files.
The schedule string can be formatted in several ways:

• An integer to define the frequency as a number of seconds. Example: 60
• An ISO8601 datetime format. Example: 2023-08-01T01:00:00+08:00
• An ISO8601 duration format. Example: PT1M
• A relative date format as supported by DateInterval. Example: 1 minute
• A cron expression. Example: /1 *

Note: You need to install the dragonmantank/cron-expression package if you want to use cron expressions as schedule strings.

Interval schedules (integer seconds, ISO 8601 durations, relative date strings) are fixed-rate: the task fires at multiples of the interval counted from the moment it was first scheduled. Per-run overhead (forking, dispatching, the task body itself) does not push the next run later — when a reschedule happens after the next slot has already passed (slow or lock-blocked run), the missed ticks are skipped and the task resumes on the next grid slot, so the cadence neither drifts nor produces a burst of catch-up executions. When a jitter is configured, the task keeps the same grid: jitter only decorates each grid slot with its random offset and no longer lets the schedule drift. Cron expressions are evaluated against wall-clock boundaries and self-correct in the same way. Sub-second intervals are honoured when given as a DateInterval with a fraction, e.g. DateInterval::createFromDateString('500 ms'): the delay is computed in fractional seconds instead of being truncated to whole seconds.

Supervisor

Supervisor can be configured with attributes or with tags in configuration files.
Processes are kept alive and wake up if one of them dies.

Packaging (experimental)

⚠️ Experimental: PHAR and standalone binary packaging are new features. The API may change in future releases.

The bundle provides commands to package your Symfony application as a standalone PHAR archive or a native binary:

See docs/build-packaging.md for full documentation, build configuration options, and known limitations.

For an overview of all documentation files, see docs/.

For security-related documentation including Host-header protection and trusted hosts configuration, see docs/security.md.

For long-running worker gotchas, state pollution, stale DB connections, blocking I/O, and other common issues, see docs/troubleshooting.md.

License

This bundle is open-sourced software licensed under the MIT license.