Added more knobs

birb_scheduler/config.py

@@ -35,3 +35,12 @@ class Config:
 
     REDIS_URL = os.environ["REDIS_URL"]
     COLLECTOR_URL = os.environ["COLLECTOR_URL"]  # http://example.com/report/
+
+    # how many times an incraising queue must be observed for an action to be taken
+    RESCHEDULE_TIRGGER_LEVEL = int(os.environ.get("RESCHEDULE_TIRGGER_LEVEL", 5))
+
+    # The counter clears itself after some time
+    RESCHEDULE_TRIGGER_COUNTER_TTL = int(os.environ.get("RESCHEDULE_TRIGGER_COUNTER_TTL", 60))
+
+    # how long a mark should live on a site which had troubled recently
+    RECENT_TROUBLE_TTL = int(os.environ.get("RECENT_TROUBLE_TTL", 120))

birb_scheduler/run_scheduler.py

@@ -8,10 +8,6 @@ from k8s_buzerator import ensure_running_pod_on_site
 
 from urllib.parse import urljoin
 
-RESCHEDULE_TIRGGER_LEVEL = 5  # how many times an incraising queue must be observed for an action to be taken
-RESCHEDULE_TRIGGER_COUNTER_TTL = 60  # The counter clears itself after some time
-RECENT_TROUBLE_TTL = 120  # how long a mark should live on a site which had trouble recently
-
 
 def run(redis_client: Redis, site_url_map: Dict[str, str]):
     run_count = redis_client.incr("RUNCOUNT")
@@ -31,17 +27,16 @@ def run(redis_client: Redis, site_url_map: Dict[str, str]):
             key = f"INCRAISINGQUEUE:{site_name}"
             incraising_queue_detected_times = redis_client.incr(key)
 
-            if incraising_queue_detected_times > RESCHEDULE_TIRGGER_LEVEL:
+            if incraising_queue_detected_times > Config.RESCHEDULE_TIRGGER_LEVEL:
                 logging.debug(f"Tirgger level reached at {site_name}")
                 redis_client.delete(key)
                 incraising_queue_at.append(site_name)
             else:
-                logging.debug(
-                    f"Suspicious queue size change at {site_name} ({incraising_queue_detected_times}/{RESCHEDULE_TIRGGER_LEVEL})")
-                redis_client.expire(key, RESCHEDULE_TRIGGER_COUNTER_TTL)  # Probably extend lifetime
+                logging.debug(f"Suspicious queue size change at {site_name} ({incraising_queue_detected_times}/{Config.RESCHEDULE_TIRGGER_LEVEL})")
+                redis_client.expire(key, Config.RESCHEDULE_TRIGGER_COUNTER_TTL)  # Probably extend lifetime
 
     # decide on default for the first time
-    if run_count > RESCHEDULE_TIRGGER_LEVEL * 2:
+    if run_count > Config.RESCHEDULE_TIRGGER_LEVEL * 2:
         default_site = redis_client.get("DEFAULT:SCHEDULED")
         if not default_site:
             logging.debug("Default site is not set. Selecting one...")
@@ -89,7 +84,7 @@ def run(redis_client: Redis, site_url_map: Dict[str, str]):
 
     # If attention required, schedule a single workload to one tier lower
     for site_seeking_attention in incraising_queue_at:
-        redis_client.set(f"RECENTTROUBLE:{site_seeking_attention}", b"\x01", ex=RECENT_TROUBLE_TTL)
+        redis_client.set(f"RECENTTROUBLE:{site_seeking_attention}", b"\x01", ex=Config.RECENT_TROUBLE_TTL)
         if current_scheduling_table_counters[site_seeking_attention] == 0:
             logging.warning("Wtf? Site reporting trouble, but there are no workload scheduled to it... nothing to do")
             continue

k8s/scheduler.yml

@@ -36,6 +36,7 @@ data:
   DEBUG: "yes"
   REDIS_URL: "redis://birb-scheduler-redis:6379/0"
   COLLECTOR_URL: "http://birb-latency-collector/report/"
+  RECENT_TROUBLE_TTL: "240"
 ---
 kind: ClusterRole
 apiVersion: rbac.authorization.k8s.io/v1