From 734bf3a42ac2dab7b39ed4023608cc1d8b52b642 Mon Sep 17 00:00:00 2001
From: Flavien David <flavien.david74@gmail.com>
Date: Thu, 23 May 2024 18:52:47 +0200
Subject: [PATCH] Qdrant cluster monitor (#5252)

* Monitor more metrics on Qdrant cluster

* :sparkles:
---
 alerting/temporal/src/qdrant/index.ts | 49 ++++++++++++++++++++++++---
 1 file changed, 45 insertions(+), 4 deletions(-)

diff --git a/alerting/temporal/src/qdrant/index.ts b/alerting/temporal/src/qdrant/index.ts
index 784e1ae9f08a..899224d1b21e 100644
--- a/alerting/temporal/src/qdrant/index.ts
+++ b/alerting/temporal/src/qdrant/index.ts
@@ -2,12 +2,35 @@ import axios from "axios";
 
 import { client, v2 } from "@datadog/datadog-api-client";
 import assert from "assert";
+import {
+  COUNT,
+  GAUGE,
+} from "@datadog/datadog-api-client/dist/packages/datadog-api-client-v2/models/MetricIntakeType";
 
 const { QDRANT_CLUSTERS, QDRANT_MONITORING_API_KEY } = process.env;
 
 assert(QDRANT_CLUSTERS, "QDRANT_CLUSTERS is not set.");
 assert(QDRANT_MONITORING_API_KEY, "QDRANT_MONITORING_API_KEY is not set.");
 
+const QDRANT_METRICS_TO_WATCH: Record<
+  "count_metrics" | "gauge_metrics",
+  ReadonlyArray<String>
+> = {
+  count_metrics: ["app_info"],
+  gauge_metrics: [
+    "cluster_peers_total",
+    "collections_total",
+    "collections_vector_total",
+    "grpc_responses_avg_duration_seconds",
+    "grpc_responses_fail_total",
+    "grpc_responses_max_duration_seconds",
+    "grpc_responses_min_duration_seconds",
+    "rest_responses_avg_duration_seconds",
+    "rest_responses_max_duration_seconds",
+    "rest_responses_min_duration_seconds",
+  ],
+};
+
 // This automatically pulls API keys from env vars DD_API_KEY.
 const configuration = client.createConfiguration();
 
@@ -18,6 +41,10 @@ configuration.setServerVariables({
 const datadogMetricsApi = new v2.MetricsApi(configuration);
 const qdrantClusters = QDRANT_CLUSTERS.split(",");
 
+function formatMetricName(rawMetricName: string) {
+  return `qdrant.${rawMetricName.replace("_", ".")}`;
+}
+
 async function fetchPrometheusMetrics(
   clusterName: string
 ): Promise<v2.MetricSeries[]> {
@@ -35,17 +62,31 @@ async function fetchPrometheusMetrics(
     metricLines.forEach((line) => {
       const [metricName, metricValue] = line.split(" ");
 
-      if (metricName === "collections_total") {
+      const timestamp = Math.floor(Date.now() / 1000);
+
+      if (QDRANT_METRICS_TO_WATCH.gauge_metrics.includes(metricName)) {
         metrics.push({
-          metric: `qdrant.${metricName.replace("_", ".")}`,
+          metric: formatMetricName(metricName),
           points: [
             {
-              timestamp: Math.floor(Date.now() / 1000),
+              timestamp,
               value: parseFloat(metricValue),
             },
           ],
           tags: ["resource:qdrant", `cluster:${clusterName}`],
-          type: 3,
+          type: GAUGE,
+        });
+      } else if (QDRANT_METRICS_TO_WATCH.count_metrics.includes(metricName)) {
+        metrics.push({
+          metric: formatMetricName(metricName),
+          points: [
+            {
+              timestamp,
+              value: parseInt(metricValue),
+            },
+          ],
+          tags: ["resource:qdrant", `cluster:${clusterName}`],
+          type: COUNT,
         });
       }
     });