Be more careful around scheduler state handling (#858)

src/service/containers.toit

@@ -94,7 +94,10 @@ class ContainerManager:
     return null
 
   create -> ContainerJob?
-      --name/string --id/uuid.Uuid --description/Map --state/any
+      --name/string
+      --id/uuid.Uuid
+      --description/Map
+      --state/any
       --reader/Reader?=null:
     if reader:
       writer/containers.ContainerImageWriter := ?

src/service/device.toit

@@ -206,7 +206,7 @@ class Device:
   Stores the scheduler jobs state information in memory that
     is preserved across deep sleeping.
   */
-  scheduler-job-states-update value/Map -> none:
+  scheduler-job-states-update value/Map? -> none:
     ram-store_ RAM-SCHEDULER-JOB-STATES_ value
 
   /**

src/service/jobs.toit

@@ -24,8 +24,8 @@ abstract class Job:
   // same with has_run_after_boot.
   scheduler-delayed-until_/JobTime? := null
 
-  constructor .name scheduler-state/any:
-    set-scheduler-state_ scheduler-state
+  constructor .name state/any:
+    set-scheduler-state_ state
 
   abstract is-running -> bool
   is-background -> bool: return false

src/service/scheduler.toit

@@ -33,16 +33,7 @@ class Scheduler:
         else:
           return schedule-wakeup_ now
     finally:
-      critical-do:
-        transition --runlevel=Job.RUNLEVEL-STOP
-        // For now, we only update the storage bucket when we're
-        // shutting down. This means that if hit an exceptional
-        // case, we will reschedule all jobs.
-        job-states := {:}
-        jobs_.do: | job/Job |
-          if state := job.scheduler-state:
-            job-states[job.name] = state
-        device_.scheduler-job-states-update job-states
+      critical-do: transition --runlevel=Job.RUNLEVEL-STOP
 
   add-jobs jobs/List -> none:
     jobs.do: add-job it

src/service/service.toit

@@ -42,7 +42,11 @@ run-artemis device/Device server-config/ServerConfig -> Duration
   containers ::= ContainerManager logger scheduler
   broker := BrokerService logger server-config
 
+  // Steal the job states, so if we do not shut down
+  // cleanly, we start in a fresh state.
   job-states := device.scheduler-job-states
+  device.scheduler-job-states-update null
+
   // Set up the basic jobs.
   synchronize-state := job-states.get SynchronizeJob.NAME
   synchronizer/SynchronizeJob := SynchronizeJob logger device containers broker synchronize-state
@@ -70,14 +74,14 @@ run-artemis device/Device server-config/ServerConfig -> Duration
     // so we have to be careful and clean up anyway.
     critical-do: provider.uninstall
 
-    // For now, we only update the storage bucket when we're
-    // shutting down. This means that if hit an exceptional
-    // case, we will reschedule all jobs.
-    job-states = {:}
-    scheduler.jobs_.do: | job/Job |
-      if state := job.scheduler-state:
-        job-states[job.name] = state
-    device.scheduler-job-states-update job-states
+  // For now, we only update the storage bucket when we're
+  // shutting down cleanly. This means that if hit an exceptional
+  // case, we will reschedule all jobs.
+  job-states = {:}
+  scheduler.jobs_.do: | job/Job |
+    if state := job.scheduler-state:
+      job-states[job.name] = state
+  device.scheduler-job-states-update job-states
 
   containers.setup-deep-sleep-triggers