Add --health-* flags to service create and update

[cezarsa]

- What I did

This PR is related to #25303 and adds --health-* and --no-healthcheck flags to service create and update commands.

- How I did it

The ContainerSpec was modified to include a HealthConfig instance which is them serialized to swarmkit. I also created a PR in swarmkit (moby/swarmkit#1641) which needs to be merged before this PR can be accepted.

For service create the behavior is similar to docker run with the same checks validating conflicting arguments. On service update it's also valid to set things like --health-cmd "" and --health-timeout 0 to allow inheriting these attributes from the image. Setting --no-healthcheck will reset the HealthConfig to the NONE test and remove every other attribute.

- How to verify it

Some test case were added to check if the commands are behaving correctly.

- Description for the changelog

Add --health-* and --no-healthcheck to service create and update

- A picture of a cute animal (not mandatory but encouraged)
🐱

[thaJeztah]

Looks like there's some golint errors;

00:10:17 ---> Making bundle: validate-lint (in bundles/1.13.0-dev/validate-lint)
00:10:17 Errors from golint:
00:10:17 cli/command/service/opts.go:72:6: exported type PositiveDurationOpt should have comment or be unexported
00:10:17 cli/command/service/opts.go:76:1: exported method PositiveDurationOpt.Set should have comment or be unexported
00:10:17 
00:10:17 Please fix the above errors. You can test via "golint" and commit the result.

[thaJeztah]

ping @stevvooe @aluzzardi @talex5

[cezarsa]

Thanks, I've fixed the lint errors. However, the code still doesn't compile on CI because it depends on moby/swarmkit#1641. If the swarmkit PR is approved (and updated on vendor) I'll rebase this PR and it should compile cleanly.

[thaJeztah]

@cezarsa to make CI pass during review, you can add a second commit that makes those changes in the vendored files. The "vendor" CI will fail, but all other tests can pass then. Once we're in "code review", and moby/swarmkit#1641 is merged, you can then update the PR to update the vendor.sh an get the actual changes from SwarmKit

[cezarsa]

Thanks for the tip @thaJeztah. Tests are passing now.

[thaJeztah]

oh boy, looks like this needs a rebase now. 😢

[vdemeester]

@cezarsa something is missing somehow. The healthcheck I provide for the service doesn't end-up into the container(s) created.

$ docker run -d --health-cmd="stat /etc/passwd" --health-interval=10s --health-retries=2 busybox top
$ docker service create --name test --health-cmd="stat /etc/passwd" --health-interval=10s --health-retries=2 busybox top
# … wait for a bit
$ docker ps
CONTAINER ID        IMAGE               COMMAND             CREATED              STATUS                    PORTS               NAMES
95b46e278bfb        busybox             "top"               14 seconds ago       Up 13 seconds (healthy)                       hungry_colden
7501f04e3610        busybox:latest      "top"               About a minute ago   Up About a minute                             test.1.lalhjh6envegchv54qjrkts4u
$ docker inspect --format={{.State.Health}} 95b
{healthy 0 [0xc42037e280 0xc42037e2d0 0xc42037e320 0xc42037e370 0xc42037e3c0]}
$ docker inspect --format={{.State.Health}} 750
<nil>

As shown in inspect or in ps, the container that was created as part of the service has no health check.

[icecrime]

Also ping @dongluochen and @aaronlehmann FYI!

[cezarsa]

Sorry @vdemeester, there was a bit of code missing. I just added it in 3d0132f

I made this mistake because similar code exists in swarmkit and I thought the daemon was using it, sorry. (I guess it will use in the future?)

Just tested it now and everything seems to be working now:

$ docker service create --health-cmd 'exit 1' busybox tail -f /dev/null
lb3l6fxxdehc8n65k172wpacy
$ docker ps
CONTAINER ID        IMAGE               COMMAND                  CREATED                  STATUS                                     PORTS                    NAMES
9528efa658fc        busybox:latest      "tail -f /dev/null"      Less than a second ago   Up Less than a second (health: starting)                            hungry_montalcini.1.moaaurb2ugxc2cktv4qnabpoo
$ docker inspect --format={{.State.Health}} 95
{starting 0 []}
$ docker inspect --format={{.Config.Healthcheck.Test}} 95
[CMD-SHELL exit 1]

[dongluochen]

I test this code and it looks good to me. One small problem would be the default value for healthcheck interval and retries are not shown. But we haven't done it in docker run.

[vdemeester]

LGTM 🐸
@cezarsa needs a rebase though 😅

[thaJeztah]

looks like it's rebased

[cpuguy83]

LGTM

Moving to docs review

[vdemeester]

@cezarsa could you squash all the commits that are not related to vendoring ? (this way we have one commit for vendoring, one for the actual diff)

[thaJeztah]

LGTM (after squashing)

[cezarsa]

Done. :)

[vdemeester]

@cezarsa perfect ! Thanks !
docs LGTM 🐸