feat: 🎸 add again the metrics about cache and queue

See #250 and #279

libs/libcache/src/libcache/cache.py

@@ -120,6 +120,9 @@ def to_split_full_name(self) -> SplitFullName:
     objects = QuerySetManager["DbSplit"]()
 
 
+AnyDb = TypeVar("AnyDb", DbDataset, DbSplit)  # Must be DbDataset or DbSplit
+
+
 class _BaseErrorItem(TypedDict):
     status_code: int
     exception: str
@@ -465,39 +468,31 @@ def get_dataset_names_with_status(status: str) -> List[str]:
     return [d.dataset_name for d in DbDataset.objects(status=status).only("dataset_name")]
 
 
-def get_datasets_count_with_status(status: Status) -> int:
-    # TODO: take the splits statuses into account?
-    return DbDataset.objects(status=status).count()
-
-
-def get_splits_count_with_status(status: Status) -> int:
-    # TODO: take the splits statuses into account?
-    return DbSplit.objects(status=status).count()
-
-
 class CountByStatus(TypedDict):
     empty: int
     error: int
     stalled: int
     valid: int
 
 
-def get_datasets_count_by_status() -> CountByStatus:
+def get_entries_count_by_status(entries: QuerySet[AnyDb]) -> CountByStatus:
+    frequencies: Dict[str, int] = entries.item_frequencies("status", normalize=False)  # type: ignore
+    # ensure that all the statuses are present, even if equal to zero
     return {
-        "empty": get_datasets_count_with_status(Status.EMPTY),
-        "error": get_datasets_count_with_status(Status.ERROR),
-        "stalled": get_datasets_count_with_status(Status.STALLED),
-        "valid": get_datasets_count_with_status(Status.VALID),
+        "empty": frequencies.get(Status.EMPTY.value, 0),
+        "error": frequencies.get(Status.ERROR.value, 0),
+        "stalled": frequencies.get(Status.STALLED.value, 0),
+        "valid": frequencies.get(Status.VALID.value, 0),
     }
 
 
+def get_datasets_count_by_status() -> CountByStatus:
+    # TODO: take the splits statuses into account?
+    return get_entries_count_by_status(DbDataset.objects)
+
+
 def get_splits_count_by_status() -> CountByStatus:
-    return {
-        "empty": get_splits_count_with_status(Status.EMPTY),
-        "error": get_splits_count_with_status(Status.ERROR),
-        "stalled": get_splits_count_with_status(Status.STALLED),
-        "valid": get_splits_count_with_status(Status.VALID),
-    }
+    return get_entries_count_by_status(DbSplit.objects)
 
 
 class DatasetCacheReport(TypedDict):

libs/libcache/tests/test_cache.py

@@ -10,7 +10,9 @@
     clean_database,
     connect_to_cache,
     delete_dataset_cache,
+    get_datasets_count_by_status,
     get_rows_response,
+    get_splits_count_by_status,
     get_valid_or_stalled_dataset_names,
     upsert_dataset,
     upsert_split,
@@ -146,3 +148,28 @@ def test_valid() -> None:
     )
 
     assert get_valid_or_stalled_dataset_names() == ["test_dataset", "test_dataset2"]
+
+
+def test_count_by_status() -> None:
+    assert get_datasets_count_by_status() == {"empty": 0, "error": 0, "stalled": 0, "valid": 0}
+
+    upsert_dataset(
+        "test_dataset", [{"dataset_name": "test_dataset", "config_name": "test_config", "split_name": "test_split"}]
+    )
+
+    assert get_datasets_count_by_status() == {"empty": 0, "error": 0, "stalled": 0, "valid": 1}
+    assert get_splits_count_by_status() == {"empty": 1, "error": 0, "stalled": 0, "valid": 0}
+
+    upsert_split(
+        "test_dataset",
+        "test_config",
+        "test_split",
+        {
+            "split_name": "test_split",
+            "rows_response": {"rows": [], "columns": []},
+            "num_bytes": None,
+            "num_examples": None,
+        },
+    )
+
+    assert get_splits_count_by_status() == {"empty": 0, "error": 0, "stalled": 0, "valid": 1}

libs/libqueue/src/libqueue/queue.py

@@ -2,7 +2,7 @@
 import logging
 import types
 from datetime import datetime
-from typing import Generic, List, Optional, Tuple, Type, TypedDict, TypeVar
+from typing import Dict, Generic, List, Optional, Tuple, Type, TypedDict, TypeVar
 
 from mongoengine import Document, DoesNotExist, connect
 from mongoengine.fields import DateTimeField, EnumField, StringField
@@ -304,12 +304,14 @@ def cancel_started_split_jobs() -> None:
 
 
 def get_jobs_count_by_status(jobs: QuerySet[AnyJob]) -> CountByStatus:
+    frequencies: Dict[str, int] = jobs.item_frequencies("status", normalize=False)  # type: ignore
+    # ensure that all the statuses are present, even if equal to zero
     return {
-        "waiting": get_jobs_with_status(jobs, Status.WAITING).count(),
-        "started": get_jobs_with_status(jobs, Status.STARTED).count(),
-        "success": get_jobs_with_status(jobs, Status.SUCCESS).count(),
-        "error": get_jobs_with_status(jobs, Status.ERROR).count(),
-        "cancelled": get_jobs_with_status(jobs, Status.CANCELLED).count(),
+        "waiting": frequencies.get(Status.WAITING.value, 0),
+        "started": frequencies.get(Status.STARTED.value, 0),
+        "success": frequencies.get(Status.SUCCESS.value, 0),
+        "error": frequencies.get(Status.ERROR.value, 0),
+        "cancelled": frequencies.get(Status.CANCELLED.value, 0),
     }
 
 

libs/libqueue/tests/test_queue.py

@@ -8,7 +8,9 @@
     connect_to_queue,
     finish_dataset_job,
     get_dataset_job,
+    get_dataset_jobs_count_by_status,
     get_split_job,
+    get_split_jobs_count_by_status,
     is_dataset_in_queue,
     is_split_in_queue,
 )
@@ -87,3 +89,29 @@ def test_is_split_in_queue() -> None:
     add_split_job(dataset_name, config_name, split_name)
     assert is_split_in_queue(dataset_name, config_name, split_name) is True
     assert is_split_in_queue(dataset_name_2, config_name, split_name) is False
+
+
+def test_count_by_status() -> None:
+    assert get_dataset_jobs_count_by_status() == {
+        "waiting": 0,
+        "started": 0,
+        "success": 0,
+        "error": 0,
+        "cancelled": 0,
+    }
+
+    add_dataset_job("test_dataset")
+
+    assert get_dataset_jobs_count_by_status() == {"waiting": 1, "started": 0, "success": 0, "error": 0, "cancelled": 0}
+
+    assert get_split_jobs_count_by_status() == {
+        "waiting": 0,
+        "started": 0,
+        "success": 0,
+        "error": 0,
+        "cancelled": 0,
+    }
+
+    add_split_job("test_dataset", "test_config", "test_split")
+
+    assert get_split_jobs_count_by_status() == {"waiting": 1, "started": 0, "success": 0, "error": 0, "cancelled": 0}

services/api/src/api/prometheus.py

@@ -55,7 +55,6 @@ def updateMetrics(self):
             self.metrics["cache_entries_total"].labels(cache="splits", status=status).set(total)
 
     def endpoint(self, request: Request) -> Response:
-        # disabled for now to fix https://github.com/huggingface/datasets-server/issues/279
-        # self.updateMetrics()
+        self.updateMetrics()
 
         return Response(generate_latest(self.getRegistry()), headers={"Content-Type": CONTENT_TYPE_LATEST})

services/api/tests/test_app.py

@@ -343,6 +343,5 @@ def test_metrics(client: TestClient) -> None:
     assert name in metrics
     assert metrics[name] > 0
     name = "process_start_time_seconds"
-    # Disabled for now
-    # assert 'queue_jobs_total{queue="datasets",status="waiting"}' in metrics
-    # assert 'cache_entries_total{cache="datasets",status="empty"}' in metrics
+    assert 'queue_jobs_total{queue="datasets",status="waiting"}' in metrics
+    assert 'cache_entries_total{cache="datasets",status="empty"}' in metrics