fleet generates stale unit state

[bcwaldon]

$ fleetctl -version
fleetctl version 0.7.0
$ fleetctl list-machines
MACHINE     IP      METADATA
3d65608d... 172.17.8.102    -
4d53f079... 172.17.8.101    -
8ab3460f... 172.17.8.103    -
$ fleetctl start examples/foo.service
Unit foo.service launched on 3d65608d.../172.17.8.102

...kill fleet on 172.17.8.101, waiting for it to disappear from the machine list

$ fleetctl list-machines
MACHINE     IP      METADATA
4d53f079... 172.17.8.101    -
8ab3460f... 172.17.8.103    -
$ fleetctl list-units
UNIT        MACHINE             ACTIVE  SUB
foo.service 3d65608d...         active  running
foo.service 4d53f079.../172.17.8.101    active  running

Notice the duplicated state for foo.service. This is mostly by design, but the old state will never not get cleaned up if the node goes down either "forever" or if units get rescheduled before it comes back up.

[bcwaldon]

/cc @jonboulle

[jonboulle]

My idea (which apparently fell by the wayside) was to just add a TTL to all of these, since we are regularly publishing them anyway. Any objection to that as the solution?