This should already work if you use the --stop-grace-period option (or stop_grace_period in the docker compose file);

For example:

Create a service with a grace period of 30 seconds; this service is running sh, which is known to ignore a graceful "stop" signal:

$ docker service create --stop-grace-period=30s -t --replicas=2 --name shellie busybox

Scale the service:

$ docker service scale shellie=1

And notice that there's still two instances running:

$ docker ps

CONTAINER ID        IMAGE               COMMAND                  CREATED              STATUS              PORTS               NAMES
eeecfb6015f7        busybox:latest      "sh"                     About a minute ago   Up About a minute                       shellie.1.uvv48xa7duj1rtmbwgpknovy9
be890da334ca        busybox:latest      "sh"                     About a minute ago   Up About a minute                       shellie.2.spizkr8l8p3h3mejal7gl4jre

After 30 seconds, the second instance will be killed.

$ docker ps

CONTAINER ID        IMAGE               COMMAND                  CREATED              STATUS              PORTS               NAMES
be890da334ca        busybox:latest      "sh"                     About a minute ago   Up About a minute                       shellie.2.spizkr8l8p3h3mejal7gl4jre

I do notice there seems to be a bug in docker service ps; after scaling the service to 1, docker service ps immediately shows just a single task (even though there are still two tasks running). I'll open a separate issue for that and link it here.

Does the above address your use case, or is there something still missing?

Add --timeout argument to docker service scale #34963

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions