How to update container without downtime?

[qupfer]

Hi, is there a "default" way to update container without (or very short downtime)?

Of cause, the easy way to upgrade is pull;stop;rm;up.
For docker-compose. there seems a ways with repicas.
1 pull
2 set repicas (--scale) to 2
3 start with --no-recreate
4 remove old
5 set scale to 1 again

Is this or something similar possible with podman?

[vrothberg]

Thanks for reaching out, @qupfer!

Regarding "no downtime": no, that is not possible with Podman. Podman does not support replicas since that would require more than one machine/node. Podman's sole focus is on single-node use cases. Everything that goes beyond that is Kubernetes' territory.

Regarding "short downtime": I recommend having a look at podman-auto-update.

[eriksjolund]

@qupfer "without downtime" could mean different things.
One interpretation could be: A client would not notice a restart of a containerized server. (The container image could be updated during the restart).

Disclaimer: I'm still learning how this all works, but this my current understanding of the topic.

If these statements are all true

• There is a containerized server application that was started by Podman via socket activation. (Let's call the containerized server application myserver).
• myserver accepted a new connection (by running accept()).
• myserver stored the accepted socket with FDSTORE.
• myserver stored the application state with FDSTORE. Application state can either be serialized to a file in /run/, or better, stored in a memory file descriptor. memfd_create(2) quote from man sd_notify (See also memfd_create(2)).
• The myserver crashed or was explicitly restarted (e.g. systemctl --user restart myserver.service). This is useful for implementing services that can restart after an explicit request or a crash without losing state quote from man sd_notify.

then the client would not notice a restart of myserver, because the listening socket FD, the accepted socket FD and the application state FD could be passed via LISTEN_FDS and LISTEN_FDNAMES (see man sd_listen_fds) to a newly started myserver. The application state could therefore be restored.

Please correct me if I've misunderstood anything.

It would be cool to see if this could be made to work!

A side-note: I wrote a demo mariadb-podman-socket-activation where Podman starts a MariaDB container with socket activation. The demo does not make use of FDSTORE, though.