From 28aaded575c858a4960768ee62118a66ab9559ce Mon Sep 17 00:00:00 2001
From: namikmesic <namikmesic@gmail.com>
Date: Wed, 5 Oct 2022 11:48:13 +0200
Subject: [PATCH] aliasAdd exporter healthcheck metric (#28)

* Add exporter healthcheck metric

* Yapf
---
 src/collectors/evm.py |  4 +---
 src/collectors/ws.py  |  1 +
 src/exporter.py       | 13 ++++++++++++-
 src/settings.py       |  1 +
 4 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/src/collectors/evm.py b/src/collectors/evm.py
index 945a806..d524119 100644
--- a/src/collectors/evm.py
+++ b/src/collectors/evm.py
@@ -47,9 +47,7 @@ async def _probe(self) -> results:
                 results.record_head_count(self.url, self.sub.head_counter)
                 results.record_disconnects(self.url, self.sub.disconnects)
         except asyncio.exceptions.TimeoutError:
-            logger.error(
-                f"Timed out while trying to establish websocket connection. Current response_timeout value in config: {cfg.response_timeout}.",
-                url=self.stripped_url)
+            logger.error(f"Timed out while trying to establish websocket connection.", url=self.stripped_url)
             results.record_health(self.url, False)
         except Exception as exc:
             results.record_health(self.url, False)
diff --git a/src/collectors/ws.py b/src/collectors/ws.py
index 9fb192f..7e30761 100644
--- a/src/collectors/ws.py
+++ b/src/collectors/ws.py
@@ -34,6 +34,7 @@ async def _subscribe(self):
             try:
                 # When we establish connection, we arm the first_time boolean, so we can record disconnect if it occurs.
                 self.first_disconnect = True
+                logger.info("Subscription connection established.", url=self.stripped_url)
                 await websocket.send(json.dumps(self.payload))
                 await self._message_counter(websocket)
             except websockets.exceptions.ConnectionClosed:
diff --git a/src/exporter.py b/src/exporter.py
index 6c8925e..e4075bd 100644
--- a/src/exporter.py
+++ b/src/exporter.py
@@ -11,7 +11,7 @@
 from collectors.starkware import starkware_collector
 from settings import logger, cfg
 from metrics_processor import results
-from settings import logger
+from settings import logger, cfg
 
 
 class prom_registry(object):
@@ -84,6 +84,9 @@ def _instantiate_starkware(self):
             self.collectors.append(starkware_collector(item))
             self.labels = self.collectors[0].labels
 
+    def _report_exporter_health(self, health_metric):
+        health_metric.add_metric([cfg.blockchain], True)
+
     def collect(self):
         metrics = {
             "brpc_health":
@@ -143,6 +146,14 @@ def collect_metrics(prom_collector):
             # Only yield metric if samples were provided by the probe
             if len(metric.samples) > 0:
                 yield metric
+        # The last step is to report exporter health
+        # This metric will be used to monitor if exporter is alive and forwarding metrics to prometheus endpoints.
+        exporter_health_metric = GaugeMetricFamily(
+            'brpc_exporter_health',
+            'Returns 1 if exporter was able to finalise scraping loop without exceptions.',
+            labels=['blockchain'])
+        self._report_exporter_health(exporter_health_metric)
+        yield exporter_health_metric
 
 
 def dummy_report(environ, start_fn):
diff --git a/src/settings.py b/src/settings.py
index 8e7686e..c625c82 100644
--- a/src/settings.py
+++ b/src/settings.py
@@ -16,6 +16,7 @@ def __init__(self, config_file_path: str, validation_file_path: str):
         self.configuration = self._load_configuration_file(config_file_path)
         self._populate_endpoints_metadata()
         self._populate_chain_id_metadata()
+        self.blockchain = self.configuration['blockchain']
         self.endpoints = self.configuration['endpoints']
 
         try: