Is it possible to gracefully restart a single Passenger process in Apache to mitigate a memory leak?

When Passenger reloads an application it rebuilds the entire application instance rather than restarting just one process. There isn’t a supported mechanism to restart only one process (or to “round robin” restart individual processes with a timeout) within Passenger.

Here’s some more detail:

• Global Restart Behavior:
  Passenger’s design philosophy is based on ensuring that all application processes run the same version of the code once an update or restart is triggered. As a side effect, using commands such as passenger-config restart-app will cause the entire application to restart rather than just one process.

• Phased Restarts:
  While Passenger supports phased restarts for zero-downtime deployments (which gradually replace old processes with new ones), it still iterates over the whole set of processes rather than target a subset by round-robin with delays. Phased restarts are intended to avoid downtime rather than isolate memory leak issues in a single process.

• Workarounds and Considerations:
  Unfortunately, there’s no built-in way to combine a timeout-based staggered restart with Passenger as it currently stands. If you’re trying to mitigate a memory leak, you might need to:

  • Consider isolating the problematic part of your app.
  • Use application-level techniques such as employing process supervision tools that can be configured to restart individual worker subprocesses (if your architecture allows it).
  • Employ a monitoring and profiling setup to identify and fix the memory leak rather than relying on restarting processes one-at-a-time.

In summary, Passenger does not support restarting just a single process or round-robin restarting with a timeout; the mechanism for restarting is designed to cover all application processes.