Improve node management.

[dongluochen]

This change improves swarm node management according to proposal #1486.

1. Introduce pending state. Pending nodes need validation before moving to healthy state. Resolve issues of duplicate ID and no join retry for unreachable node issues.
2. Expose error and last update time in docker info.
3. Use connect success/failure to drive state transition between healthy and unhealthy.

Signed-off-by: Dong Chen dongluo.chen@docker.com

[abronan]

Design LGTM. Will review the code and test shortly.

ping @vieux for design approval.

@dongluochen This might need a set of integration tests (or improve the existing ones) to validate that the state machine works and that a node transitions correctly from a state to another. At least for Pending to Healthy and from Healthy to Unhealthy.

[dongluochen]

@abronan thanks Alex! I'll update test cases soon.

[vieux]

@dongluochen small comment, but could we please put └ Error: <none> or └ Error: (none) instead of blanc ?

[dongluochen]

@vieux I updated no error to └ Error: (none). Thanks for the comment.

[vieux]

LGTM

[dongluochen]

ping @docker/swarm-maintainers.

[abronan]

@dongluochen This is just for convenience but can we add a TODO and reference to the issue (#1508) here? Because the comment is there anyway :)

[abronan]

Tested and seems to work as expected, just very small nits but LGTM

[dongluochen]

Thanks @abronan! I've updated code accordingly.

[abronan]

Sweet! Thanks @dongluochen!

[abronan]

🎉