Fix initial_replicas for scheduled services

Author: r4victor

src/dstack/_internal/server/services/prometheus/client_metrics.py

@@ -5,6 +5,9 @@ class RunMetrics:
     """Wrapper class for run-related Prometheus metrics."""
 
     def __init__(self):
+        # submit_to_provision_duration reflects real provisioning time
+        # but does not reflect how quickly provisioning processing works
+        # since it includes scheduling time, retrying, etc.
         self._submit_to_provision_duration = Histogram(
             "dstack_submit_to_provision_duration_seconds",
             "Time from when a run has been submitted and first job provisioning",

src/dstack/_internal/server/services/runs.py

@@ -517,6 +517,8 @@ async def submit_run(
         if run_spec.merged_profile.schedule is not None:
             initial_status = RunStatus.PENDING
             initial_replicas = 0
+        elif run_spec.configuration.type == "service":
+            initial_replicas = run_spec.configuration.replicas.min
 
         run_model = RunModel(
             id=uuid.uuid4(),
@@ -537,7 +539,6 @@ async def submit_run(
         session.add(run_model)
 
         if run_spec.configuration.type == "service":
-            initial_replicas = run_spec.configuration.replicas.min
             await services.register_service(session, run_model, run_spec)
 
         for replica_num in range(initial_replicas):