failure should be counted only on failure

ooniapi/services/ooniprobe/src/ooniprobe/metrics.py

@@ -91,6 +91,12 @@ class Metrics:
         labelnames=["status", "instance"],
     )
 
+    FASTPATH_INSTANCE_FAILURE = Counter(
+        "fastpath_instance_failure",
+        "How many times a measurement submission per fastpath instance failed",
+        labelnames = ["instance"]
+    )
+
     SEND_S3_CNT = Counter(
         "measurement_s3_upload_count",
         "How many times ooniprobe sent a measurement to s3 disaggregated by status"

ooniapi/services/ooniprobe/src/ooniprobe/routers/reports.py

@@ -194,13 +194,14 @@ async def receive_measurement(
                 break
 
             except Exception as e:
+                Metrics.FASTPATH_INSTANCE_FAILURE.labels(
+                    instance=fastpath_url
+                    ).inc()
                 log.exception(
                     f"[{i + 1} / {len(fastpath_urls)}] Unable to send measurement to fastpath "
                     f"({fastpath_url}): {e}"
                 )
 
-    Metrics.SEND_FASTPATH_CNT.labels(status="fail", instance="NA").inc()
-
     if success:
         # Geoip anomaly detection runs only when the measurement was successfully
         # submitted to the fastpath, so retries don't cause duplicate anomaly entries.
@@ -222,6 +223,8 @@ async def receive_measurement(
 
         return ReceiveMeasurementResponse(measurement_uid=msmt_uid)
 
+    Metrics.SEND_FASTPATH_CNT.labels(status="fail", instance="NA").inc()
+
     # wasn't possible to send msmnt to fastpath, try to send it to s3
     ts_prefix = now.strftime("%Y%m%d%H")
     tn = test_name.replace("_", "")

ooniapi/services/ooniprobe/src/ooniprobe/routers/v1/probe_services.py

@@ -998,7 +998,7 @@ async def submit_measurement(
                     f"({fastpath_url}): {e}"
                 )
 
-    Metrics.SEND_FASTPATH_CNT.labels(status="fail", instance="NA").inc()
+        Metrics.SEND_FASTPATH_CNT.labels(status="fail", instance="NA").inc()
 
     if success:
         # Geoip anomaly detection runs only when the measurement was successfully