Merge pull request #208 from notoraptor/cw-556-cluster-state-per-day

[cw-556] add a script to generate cluster status

clockwork_web/core/users_helper.py

@@ -581,7 +581,12 @@ def render_template_with_user_settings(template_name_or_list, **context):
     context["web_settings_json_str"] = json.dumps(context["web_settings"])
 
     # Send the clusters infos to the template
-    context["clusters"] = get_all_clusters()
+    # NB: get_all_clusters() seems to return the clusters dict itself
+    # from config, not a copy. So, any modification on clusters dict
+    # returned by this function will be propagated into config.
+    # As we don't want this behaviour here, we will make a copy
+    # of each cluster dict.
+    context["clusters"] = {k: v.copy() for k, v in get_all_clusters().items()}
     # List clusters available for connected user,
     # or set an empty list for anon user.
     context["user_clusters"] = (
@@ -591,40 +596,28 @@ def render_template_with_user_settings(template_name_or_list, **context):
     )
 
     # Get cluster status (if jobs are old and cluster has error).
-    """
     for cluster_name in context["clusters"]:
-        # Cluster error cannot yet be checked, so
-        # cluster_has_error is always False for now.
-        cluster_has_error = False
-        context["clusters"][cluster_name]["status"] = {
-            "jobs_are_old": _jobs_are_old(cluster_name),
-            "cluster_has_error": cluster_has_error,
-        }
-    """
+        context["clusters"][cluster_name]["status"] = _get_cluster_status(cluster_name)
 
     return render_template(template_name_or_list, **context)
 
 
-def _jobs_are_old(cluster_name):
-    """Return True if last slurm update in given cluster is older than 2 days."""
-    jobs_are_old = False
+def _get_cluster_status(cluster_name):
+    """
+    Get cluster status from DB collection `cluster_status`.
 
-    mongodb_filter = {"slurm.cluster_name": cluster_name}
+    Collection should be updated from an independent script
+    (`scripts/update_clusters_status.py`) regularly.
+    """
     mc = get_db()
-    job_with_max_cw_last_slurm_update = list(
-        mc["jobs"].find(mongodb_filter).sort([("cw.last_slurm_update", -1)]).limit(1)
-    )
-
-    if job_with_max_cw_last_slurm_update:
-        (job,) = job_with_max_cw_last_slurm_update
-        if "last_slurm_update" in job["cw"]:
-            most_recent_job_edition = job["cw"]["last_slurm_update"]
-            current_timestamp = datetime.now().timestamp()
-            elapsed_time = timedelta(
-                seconds=current_timestamp - most_recent_job_edition
-            )
-            # Let's say the latest jobs edition must not be older than max_delay.
-            max_delay = timedelta(days=2)
-            jobs_are_old = elapsed_time > max_delay
-
-    return jobs_are_old
+    statuses = list(mc["cluster_status"].find({"cluster_name": cluster_name}))
+    if statuses:
+        # Status found
+        (status,) = statuses
+        return status
+    else:
+        # No status found, return default values
+        return {
+            "jobs_are_old": False,
+            "cluster_has_error": False,
+        }

clockwork_web/templates/base.html

@@ -232,17 +232,17 @@ <h1><a data-bs-toggle="collapse" data-bs-target=".formCollapse" aria-expanded="f
 														<i class="fa-solid fa-file-lines" data-bs-toggle="tooltip" data-bs-placement="right" title="{{ gettext('Grafana cluster link') }}"></i>
 													</a>
 													<!-- cluster status -->
-													<!--
-													{# if D_cluster['status']['jobs_are_old'] #}
+													{% if D_cluster['status']['jobs_are_old'] %}
 														<span class="cluster-info cluster-warning">
 															<i class="fa-solid fa-triangle-exclamation" data-bs-toggle="tooltip" data-bs-placement="right" title="{{ gettext('Possible stale jobs. Most recent update was more than 30 days ago') }}"></i>
 														</span>
-														
-													{# else #}
+
+													{% else %}
 														<span class="cluster-info cluster-good">
 															<i class="fa-solid fa-ballot-check" data-bs-toggle="tooltip" data-bs-placement="right" title="{{ gettext('Latest update to jobs is relatively recent (at most 30 days ago)') }}"></i>
 														</span>
-													{# endif #}
+													{% endif %}
+													<!-- `cluster_has_error` not yet updated, thus still not displayed
 													{# if D_cluster['status']['cluster_has_error'] #}
 														<span class="cluster-info cluster-error">
 															<i class="fa-solid fa-octagon-exclamation" data-bs-toggle="tooltip" data-bs-placement="right" title="{{ gettext('A cluster error occurred recently') }}"></i>

scripts/update_clusters_status.py

@@ -0,0 +1,84 @@
+from datetime import datetime, timedelta
+
+from clockwork_web.config import register_config
+from clockwork_web.core.clusters_helper import get_all_clusters
+from slurm_state.mongo_client import get_mongo_client
+from slurm_state.config import get_config
+
+
+def main():
+    # Register the elements to access the database
+    register_config("mongo.connection_string", "")
+    register_config("mongo.database_name", "clockwork")
+
+    # Get database and collection objects
+    client = get_mongo_client()
+    db_insertion_point = client[get_config("mongo.database_name")]
+    collection_name = "cluster_status"
+    collection = db_insertion_point[collection_name]
+
+    # Get clusters
+    clusters = get_all_clusters()
+
+    # Generate clusters statuses
+    cluster_to_status = []
+    for cluster_name in clusters:
+        # Cluster error cannot yet be checked, so
+        # cluster_has_error is always False for now.
+        cluster_has_error = False
+        cluster_to_status.append(
+            {
+                "cluster_name": cluster_name,
+                "jobs_are_old": _jobs_are_old(db_insertion_point, cluster_name),
+                "cluster_has_error": cluster_has_error,
+            }
+        )
+
+    # Create collection index if necessary
+    if not list(collection.list_indexes()):
+        print("Create index for collection:", collection_name)
+        collection.create_index(
+            [
+                ("cluster_name", 1),
+                ("jobs_are_old", 1),
+                ("cluster_has_error", 1),
+            ],
+            name="cluster_status_index",
+        )
+    # Save clusters statuses in database
+    for cluster_status in cluster_to_status:
+        collection.update_one(
+            {"cluster_name": cluster_status["cluster_name"]},
+            {"$set": cluster_status},
+            upsert=True,
+        )
+
+    print("Updated.")
+
+
+def _jobs_are_old(mc, cluster_name):
+    """Return True if last slurm update in given cluster is older than 2 days."""
+    jobs_are_old = False
+
+    mongodb_filter = {"slurm.cluster_name": cluster_name}
+    job_with_max_cw_last_slurm_update = list(
+        mc["jobs"].find(mongodb_filter).sort([("cw.last_slurm_update", -1)]).limit(1)
+    )
+
+    if job_with_max_cw_last_slurm_update:
+        (job,) = job_with_max_cw_last_slurm_update
+        if "last_slurm_update" in job["cw"]:
+            most_recent_job_edition = job["cw"]["last_slurm_update"]
+            current_timestamp = datetime.now().timestamp()
+            elapsed_time = timedelta(
+                seconds=current_timestamp - most_recent_job_edition
+            )
+            # Let's say the latest jobs edition must not be older than max_delay.
+            max_delay = timedelta(days=2)
+            jobs_are_old = elapsed_time > max_delay
+
+    return jobs_are_old
+
+
+if __name__ == "__main__":
+    main()