Task recovery when fleet adapter get restarted

[youliangtan]

Currently, once a task is assigned to the robot, it will get "stored" in a fleet adapter internal task queue. For whatever reason that triggers a fleet adapter restart, the assigned tasks will get lost after the restart. Therefore, task recovery capability is quite crucial in a RMF system. (for example, user will often update the fleet adapter pod to a newer release, which requires a restart).

There's a pending PR which tries to implement the task backup feature: #161
In this pr, the task logs and backup are stored in a form of a db, in this case, .sqlite format.

Here, An extended proposal is to leverage on the existing rmf-web system. Since we already update the rmf-web persistence db
via WebSocket (with server_uri param), the fleet adapter can also query the same database during restart. This results in a simpler architecture. Furthermore, similar to the usage of Tortoise ORM in the api-server, the fleet adapter should also use an ORM to support various db. (eg. postgresql, sqlite...)

flowchart LR
    A[Fleet Adapter] -->|ws broadcast states| B(rmf-web: api server)
    B <-->|orm| D[database]
    A <-->|orm| D[database]

Loading