Introduce job_submissions_limit for /api/runs/list

Author: r4victor

src/dstack/_internal/core/models/runs.py

@@ -556,6 +556,10 @@ def _get_error(termination_reason: Optional[RunTerminationReason]) -> Optional[s
 
     @root_validator
     def _status_message(cls, values) -> Dict:
+        # FIXME: status_message should not require all job submissions for status calculation
+        # since it's very expensive and is not required for anything else.
+        # May return a different status if not all job submissions requested.
+        # TODO: Calculate status_message by looking at job models directly instead job submissions.
         try:
             status = values["status"]
             jobs: List[Job] = values["jobs"]

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

@@ -54,6 +54,7 @@ async def list_runs(
         repo_id=body.repo_id,
         username=body.username,
         only_active=body.only_active,
+        job_submissions_limit=body.job_submissions_limit,
         prev_submitted_at=body.prev_submitted_at,
         prev_run_id=body.prev_run_id,
         limit=body.limit,

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

@@ -9,12 +9,20 @@
 
 
 class ListRunsRequest(CoreModel):
-    project_name: Optional[str]
-    repo_id: Optional[str]
-    username: Optional[str]
+    project_name: Optional[str] = None
+    repo_id: Optional[str] = None
+    username: Optional[str] = None
     only_active: bool = False
-    prev_submitted_at: Optional[datetime]
-    prev_run_id: Optional[UUID]
+    job_submissions_limit: Optional[int] = Field(
+        None,
+        ge=0,
+        description=(
+            "Limit number of job submissions returned per job to avoid large responses."
+            "Drops old job submissions"
+        ),
+    )
+    prev_submitted_at: Optional[datetime] = None
+    prev_run_id: Optional[UUID] = None
     limit: int = Field(100, ge=0, le=100)
     ascending: bool = False
 

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

@@ -105,6 +105,7 @@ async def list_user_runs(
     repo_id: Optional[str],
     username: Optional[str],
     only_active: bool,
+    job_submissions_limit: Optional[int],
     prev_submitted_at: Optional[datetime],
     prev_run_id: Optional[uuid.UUID],
     limit: int,
@@ -148,7 +149,11 @@ async def list_user_runs(
     runs = []
     for r in run_models:
         try:
-            runs.append(run_model_to_run(r, return_in_api=True))
+            runs.append(
+                run_model_to_run(
+                    r, return_in_api=True, job_submissions_limit=job_submissions_limit
+                )
+            )
         except pydantic.ValidationError:
             pass
     if len(run_models) > len(runs):
@@ -653,6 +658,7 @@ async def delete_runs(
 def run_model_to_run(
     run_model: RunModel,
     include_job_submissions: bool = True,
+    job_submissions_limit: Optional[int] = None,
     return_in_api: bool = False,
     include_sensitive: bool = False,
 ) -> Run:
@@ -666,6 +672,11 @@ def run_model_to_run(
         ):
             submissions = []
             job_model = None
+            if job_submissions_limit is not None:
+                if job_submissions_limit == 0:
+                    job_submissions = []
+                else:
+                    job_submissions = list(job_submissions)[-job_submissions_limit:]
             for job_model in job_submissions:
                 if include_job_submissions:
                     job_submission = job_model_to_job_submission(job_model)

src/dstack/api/_public/runs.py

@@ -748,6 +748,7 @@ def list(self, all: bool = False, limit: Optional[int] = None) -> List[Run]:
             repo_id=None,
             only_active=only_active,
             limit=limit or 100,
+            job_submissions_limit=1,  # no need to return more than 1 submission per job
         )
         if only_active and len(runs) == 0:
             runs = self._api_client.runs.list(

src/dstack/api/server/_runs.py

@@ -33,12 +33,14 @@ def list(
         prev_run_id: Optional[UUID] = None,
         limit: int = 100,
         ascending: bool = False,
+        job_submissions_limit: Optional[int] = None,
     ) -> List[Run]:
         body = ListRunsRequest(
             project_name=project_name,
             repo_id=repo_id,
             username=username,
             only_active=only_active,
+            job_submissions_limit=job_submissions_limit,
             prev_submitted_at=prev_submitted_at,
             prev_run_id=prev_run_id,
             limit=limit,