End of Docker and Kubernetes: Grace period from SIGTERM to SIGKILL

A rough understanding of SIGTERM and SIGKILL

Here's a quick overview of SIGTERM and SIGKILL:

• SIGTERM: Termination request signal. The process can still run after receiving this.
• SIGKILL: Forced termination. The process can no longer run.

Graceful shutdown

Graceful shutdown means that a process or the like closes the acceptance of new work, processes the work in process to the point where it is in process, and then finishes it safely. The image is "stores".

When a kill occurs, the system may be in a half-finished state or the data may be inconsistent. Graceful shutdown is a protective measure to prevent this from happening.

For example, considering the graceful shutdown of an HTTP server as an example, when the HTTP server enters the termination phase, it will not accept new requests, but the existing requests will be terminated after returning a proper response.

Docker, Kubernetes, etc. should send a SIGTERM signal after sending aSIGTERM signal so that the application can be gracefully shut down, and then after a certain grace period, if it still does not terminate, send a SIGKILL to force it to terminate. It has become.

Grace from SIGTERM to SIGKILL such as Docker and Kubernetes

The grace period depends on the system. The grace period for each system is listed below. Please refer to it when implementing Graceful Shutdown.

docker stop

• Send SIGTERM and after 10 seconds send SIGKILL if the process remains.
• 10 seconds can be changed with the -t and --time options.
• The same applies to docker-compose stop and docker-compose down.

docker kill

• Immediately send SIGKILL
• Same for docker rm -f, docker-compose kill.

Kubernetes Pod Termination

• Send SIGTERM and after 30 seconds send SIGKILL if the process remains.

Systemd

• SIGTERM → 90 seconds later SIGKILL

Upstart

• SIGTERM → 5 seconds later SIGKILL

Sysvinit

• SIGTERM → 5 seconds later SIGKILL