Retain IP address for container across restart

[bboreham]

from IRC

If you use docker restart then weave should arrange that the container retains its IP
address.

(Note that restart is indeed a Docker daemon command)

This could be viewed as a sub-case of #599, IP address mobility.

[rade]

Surely that's more of an IPAM issue. We free the IP when the container dies, which it does as part of restart. If it wasn't for that then this would "just work", since IPAM for a specific container & network will hand back the same address if called repeatedly.

So if we really want this feature (use case?) then IPAM needs to distinguish restarts from deaths.

Or was the idea somehow that the proxy would intercept the restart and tell IPAM to ignore the container death that follows?

[bboreham]

Yes, the proxy sees the restart command and hence could know what to do.

[rade]

IPAM cannot detect restarts from the event stream?

[rade]

Surely we should also file an issue for doing this when the container is restarted by a restart policy.

[bboreham]

If the container is restarted without weave being involved, then we can't attach an interface to the weave network. Which is certainly an issue, but probably not this issue.

That leaves the proxy, already discussed, and the weave script, which doesn't have a restart command.

[bboreham]

There is a restart event.

According to this blog it comes after the die and start events.

[rade]

yes, I saw. How very useful.

[bboreham]

Restoring the network interface on daemon-controlled restart is captured as #401

[rade]

the weave script, which doesn't have a restart command.

Hmm. So how do we make this work for containers started with weave run? I suppose we could decide not to bother. Just as we didn't with #400.

And what happens when containers have been attached to weave with weave attach? See #688.

[bboreham]

Referencing the similar Docker issue moby/moby#2801