_wait_all_computers gives up too soon

It waits an hour, how long are you suggesting it waits for?

We modified the sequence of events in https://bugs.launchpad.net/landscape/+bug/1654399, maybe we should have bumped up this retry time too.

Do we have an upper bound on the autopilot deployment time? Would doubling the _wait_all_computers retry window cause us to hit that?

There is now a second example of this:
https://ci.lscape.net/job/landscape-system-tests/5215/

In 5215, the "Wait for computers to register" activity is:
    {
        "children": [],
        "schedule_after_time": null,
        "region_id": 1,
        "creator": {
            "id": 1,
            "name": "John Doe",
            "email": "<email address hidden>"
        },
        "result_text": "1 machines not yet registered",
        "activity_status": "failed",
        "creation_time": "2017-02-07T19:53:50Z",
        "schedule_before_time": null,
        "summary": "Wait for computers to register",
        "completion_time": "2017-02-07T22:06:13Z",
        "modification_time": "2017-02-07T22:06:13Z",
        "parent_id": 1,
        "type": "WaitForRegisteredComputersRequest",
        "id": 250,
        "result_code": null
    },

[from landscape-0-inner-logs/rabbitmq-server-0/var/log/juju/unit-landscape-client-20.log]
2017-02-07 22:06:26 INFO juju-log container:93: In container-relation-joined for landscape-client/20
2017-02-07 22:06:26 INFO juju-log container:93: Updating client configuration for landscape-client/20
2017-02-07 22:06:26 INFO juju-log container:93: landscape-client/20: landscape-client is not configured to run as per sysvconfig
2017-02-07 22:06:27 INFO container-relation-joined Restarting landscape-client (via systemctl): landscape-client.service.
2017-02-07 22:06:27 INFO juju-log container:93: Registering landscape-client/20

The missing client registered just seconds after the activity failed. I have not studied successful tests to determine if we are generally running close to the retry cap or if these are just outliers.