cloud/bmaas: split ShepherdAccess into Shepherd{AgentStart,Recovery}
This effectively undoes our previous attempted consolidation of all
Shepherd accesses under one tag. We now use two separate tags for the
two main Shepherd work processes, and mutually exclude them in SQL.
We do this so that we can see more clearly in work history (and in
general when processing machines) what the Shepherd is actually trying
to do to a machine.
The downside of this implementation is that we now extend the BMDB/ETP
model to being able to mutually exclude different processes. This is
easy enough to express in SQL, but might make future generic modelling
more difficult.
An alternative would be to add an extra field to work/work history that
acts as an informative field for operators to know the details of a work
item. We might still want to do that in the future. However, that field
being freeform, we could not really rely on it for machine parsing.
Change-Id: I9578ac000f6112514fe587e9fddf7e85671c6437
Reviewed-on: https://review.monogon.dev/c/monogon/+/1584
Reviewed-by: Leopold Schabel <leo@monogon.tech>
Tested-by: Jenkins CI
diff --git a/cloud/shepherd/equinix/manager/recoverer.go b/cloud/shepherd/equinix/manager/recoverer.go
index 4ec73af..85ec440 100644
--- a/cloud/shepherd/equinix/manager/recoverer.go
+++ b/cloud/shepherd/equinix/manager/recoverer.go
@@ -8,6 +8,7 @@
"k8s.io/klog/v2"
+ "source.monogon.dev/cloud/bmaas/bmdb"
"source.monogon.dev/cloud/bmaas/bmdb/model"
ecl "source.monogon.dev/cloud/shepherd/equinix/wrapngo"
)
@@ -40,6 +41,17 @@
}, nil
}
+func (r *Recoverer) getProcessInfo() processInfo {
+ return processInfo{
+ process: model.ProcessShepherdRecovery,
+ defaultBackoff: bmdb.Backoff{
+ Initial: 1 * time.Minute,
+ Maximum: 1 * time.Hour,
+ Exponent: 1.2,
+ },
+ }
+}
+
func (r *Recoverer) getMachines(ctx context.Context, q *model.Queries, limit int32) ([]model.MachineProvided, error) {
return q.GetMachineForAgentRecovery(ctx, limit)
}