Swarm Mode (with multiple nodes)

[tjhoff]

Hello! Wondering if anyone can give me some pointers on running this tool in a multi-node swarm. I've read through the docs (specifically mentioning swarm mode) but didn't find any exact answers.

Given that volumes exist only on the docker node where a service replica is running, and service replicas can be assigned to whatever nodes have availability (and meet constraints) - what could I do to use this tool to effectively scale backups of service volumes?

A minimal example of a stack that shows this behavior is below:

version: "3.8"

services:
  foo:
    image: busybox
    command: "sleep infinity"
    volumes:
      - shared_volume:/home/foo

  bar:
    image: busybox
    command: "sleep infinity"
    volumes:
      - shared_volume:/home/bar

volumes:
  shared_volume:

Assuming I have nodes a and b in the swarm, I saw that the foo service container was replicated on node a, and the bar service container was replicated on node b.

I can think of a few (non-optimal) ways to do this

• use global replication for the backup task (scaling unfortunately would be N services x M nodes)
• use node hostname constraints for both the deployed services and the backup service (not super helpful for scaling services)
• use swarm-cronjob to run global replicas of docker-volume-backup that dynamically scrapes and backs up volumes based on a whitelist (whitelist has to be maintained)
• run a system service outside of docker on each node (loses the "just works" aspect of docker swarm stack deployment and node onboarding)

I realize this gets pretty close into the evergreen issue of Docker Swarm shared volumes, but looking for ways around this for my current infrastructure.

Appreciate any help - really interested in this tool!

[m90]

How do you make sure the service that is using the volume you want to back up is placed on the correct node? I'd probably pick the same approach for the backup service.

[tjhoff]

Right, that's a great question, and something I should have considered.

Currently these stacks are using node hostname/labeling constraints which ensures the same node and therefore correct volume is being used - I could certainly use the same constraints for the backup service (with the caveat being that would be duplication of deployment constraints).

I might look into service labels with a global deployment configuration to allow for traefik-style label based backups. Thanks for the advice!