WIP

detector/detector/detector.py

@@ -357,6 +357,7 @@ def rebuild_status(click, start_date, end_date, services) -> None:
 #  If there are changes we then extract the whole history for each changed
 #  TCAI in rebuild_feeds.
 
+
 def run_detection_sql(start_date, end_date, services) -> None:
     if not conf.reprocess:
         log.info(f"Running detection from {start_date} to {end_date}")
@@ -633,24 +634,25 @@ def process_data(blocking_status, new):
         + m.accessible_perc_BS * nu
     )
 
-    # Stability should grow slowly when accessible_perc is constant over time
-    # but drop quickly if accessible_perc changes a lot
+    # Stability moves slowly towards 1 when accessible_perc is constant over
+    # time but drop quickly towards 0 when accessible_perc changes a lot.
+    # It is later on used to decide when we are confident enough to make
+    # statements on BLOCKED/OK status. It is also immediately set to 0 when we
+    # detect a blocking change event to mitigate flapping.
     s_def = 0.7  # default stability
-    stability = np.cos(3.14 / 2 * delta)
-    m["stab_insta"] = stability
-
-    tau = 0.99
-    good = stability * (1 - tau) + m.stability.fillna(s_def) * tau
+    stability_thr = 0.8  # threshold to consider a TCAI stable
+    gtau = 0.99  # moving average tau for
+    btau = 0.7
 
-    tau = 0.7
-    bad = stability * (1 - tau) + m.stability.fillna(s_def) * tau
+    stability = np.cos(3.14 / 2 * delta)
+    m["stab_insta"] = stability  # used for charting
+    good = stability * (1 - gtau) + m.stability.fillna(s_def) * gtau
+    gstable = stability >= stability_thr
+    m.loc[gstable, "stability"] = good[gstable]
 
-    thr = 0.8
-    s = stability >= thr
-    # m[s]["stability"] = good[s]
-    m.loc[s, "stability"] = good[s]
-    s = stability < thr
-    m.loc[s, "stability"] = bad[s]
+    bad = stability * (1 - btau) + m.stability.fillna(s_def) * btau
+    bstable = stability < stability_thr
+    m.loc[bstable, "stability"] = bad[bstable]
 
     m.status = m.status.fillna("UNKNOWN")
     m.old_status = m.status.fillna("UNKNOWN")
@@ -688,23 +690,27 @@ def process_data(blocking_status, new):
     # moving average on cnt
     m["cnt"] = mavg_cnt
 
-    assert sorted(m.columns) == [
-        "accessible_perc",
-        "change",
-        "cnt",
-        "confirmed_perc",
-        "input",
-        "input_ap",
-        "input_cnt",
-        "old_status",
-        "probe_asn",
-        "probe_cc",
-        "pure_anomaly_perc",
-        "stab_insta",
-        "stability",
-        "status",
-        "test_name",
-    ] or True, sorted(m.columns)
+    assert (
+        sorted(m.columns)
+        == [
+            "accessible_perc",
+            "change",
+            "cnt",
+            "confirmed_perc",
+            "input",
+            "input_ap",
+            "input_cnt",
+            "old_status",
+            "probe_asn",
+            "probe_cc",
+            "pure_anomaly_perc",
+            "stab_insta",
+            "stability",
+            "status",
+            "test_name",
+        ]
+        or True
+    ), sorted(m.columns)
     return m, events
 
 
@@ -780,8 +786,7 @@ def load_blocking_status_df(click):
 
 
 def reprocess_data_from_df(idf, debug=False):
-    """Reprocess data using Pandas
-    Used for testing/tuning
+    """Reprocess data using Pandas. Used primarily for testing.
     """
     status = pd.DataFrame(
         columns=[
@@ -832,6 +837,7 @@ def reprocess_data_from_df(idf, debug=False):
         events = pd.concat(events_tmp)
     else:
         events = pd.DataFrame()
+    print("DONE")
     return events, status, status_history
 
 
@@ -844,7 +850,7 @@ def main():
     click = Clickhouse.from_url(conf.db_uri)  # settings={"use_numpy":True})
 
     # create_tables()
-    # FIXME: configure services
+    # TODO: configure services
     services = {
         "Facebook": ["https://www.facebook.com/"],
         "Twitter": ["https://twitter.com/"],