Deploy without service breakdown

[honibis]

Description

When deploying/redeploying service instances are killed causing current operations and any long running tasks to end. Also causing connection interruption from the client side. Is it possible to use draining in coordination with load balancer(like traefik)?

Steps to reproduce the issue:

1. Redeploy a service

What happens:
Error response to client while restarting tasks and killing long running tasks.

What should happen:
1- Create new tasks and redirect new traffic only to new tasks
2- Mark the old tasks as draining and prevent any new connections
3- Wait a certain time(couple minutes, hopefully configurable)
4- End old tasks

Additional information (e.g. docker version, cluster setup,...):

[honibis]

I set update config to start first and gave a bit more time which helped a bit:

      update_config:
        parallelism: 1
        order: start-first
        failure_action: rollback
        delay: 30s

Also tried to change load balancing to swarm using
traefik.docker.lbswarm=true
But this was not helpful because i could not make lbswarm work with traefik, all the requests ended up at the same task.

Then to make things a bit more robust i added health-check to traefik

        - traefik.http.services.gsb.loadbalancer.healthcheck.path=/api/service/health
        - traefik.http.services.gsb.loadbalancer.healthcheck.interval=200ms
        - traefik.http.services.gsb.loadbalancer.healthcheck.timeout=75ms
        - traefik.http.services.gsb.loadbalancer.healthcheck.scheme=http

Currently it is ok, will work on zero downtime later.
And before forgetting swarmpit is helping me a lot, thanks to all who participated in this wonderful application :)