Skip to content

PostgreSQL backup tool with Point-In-Time-Recovery and restore verification

3.5 KB
Raw Blame

Container Shutdown Specification

Purpose

Defines how the Databasus container stops its application and its embedded PostgreSQL server, so that a normal stop leaves the metadata database cleanly shut down and the next start needs no crash recovery.

Requirements

Requirement: A stop signal shuts the application down before the embedded database

When the container receives SIGTERM or SIGINT while the application runs, the container SHALL pass the signal to the application and wait for the application process to exit. The embedded PostgreSQL server SHALL receive its shutdown request only after that exit. The container SHALL then shut the server down cleanly, wait for that shutdown to finish and exit with the application's exit status.

Scenario: Operator stops the container

  • WHEN an operator stops a healthy container with the default stop timeout
  • THEN the container exits with status 0 before the timeout
  • AND the stopped data directory reports the cluster state shut down, not in production

Scenario: Interrupt signal

  • WHEN a healthy container receives SIGINT
  • THEN the container exits before the stop timeout
  • AND the stopped data directory reports the cluster state shut down

Scenario: Database shutdown follows the application's exit

  • WHEN the container stops after a stop signal
  • THEN the log line in which the container reports the application's exit comes before the line in which PostgreSQL reports its shutdown request

Requirement: The container ends when either service ends

When the application exits without a stop signal, the container SHALL shut the embedded PostgreSQL server down cleanly and exit with the application's exit status. When the embedded PostgreSQL server exits while the application runs, the container SHALL stop the application and exit with a non-zero status.

Scenario: Application exits on its own

  • WHEN the application process exits while the container has received no stop signal
  • THEN the container exits with the application's exit status
  • AND the stopped data directory reports the cluster state shut down

Scenario: Embedded database exits unexpectedly

  • WHEN the embedded PostgreSQL server exits while the application runs
  • THEN the application process stops
  • AND the container exits with a non-zero status

Requirement: A stop signal during startup ends startup

When the container receives SIGTERM or SIGINT before the application has started, it SHALL NOT start the application. It SHALL stop the PostgreSQL process that is running at that moment, wait for it to exit and exit with a non-zero status. The container SHALL act on the signal while a PostgreSQL startup step is still running, not only after that step has finished.

Scenario: Stop while the regular server starts

  • WHEN the container receives a stop signal after the regular PostgreSQL server has started and before the application starts
  • THEN the container exits with a non-zero status before the stop timeout
  • AND the log has no line that the application starts
  • AND the stopped data directory reports the cluster state shut down

Scenario: Stop during a long startup step

  • WHEN the container receives a stop signal while a PostgreSQL startup administration step is still running
  • THEN the container logs that it is stopping startup before that step has ended
  • AND the container exits with a non-zero status, and the application never starts