From 381104cbc1d1ef5afcc9fb70158b11a2bdb5e51c Mon Sep 17 00:00:00 2001
From: Antonis Kouzoupis <kouzoupis.ant@gmail.com>
Date: Fri, 14 Mar 2025 08:56:38 +0100
Subject: [PATCH] Update instructions for exporting Prometheus metrics (#446)

---
 .../admin/monitoring/export-metrics.md        | 122 ++++++++++++------
 1 file changed, 83 insertions(+), 39 deletions(-)

diff --git a/docs/setup_installation/admin/monitoring/export-metrics.md b/docs/setup_installation/admin/monitoring/export-metrics.md
index 26d10e794..dd94d72fc 100644
--- a/docs/setup_installation/admin/monitoring/export-metrics.md
+++ b/docs/setup_installation/admin/monitoring/export-metrics.md
@@ -2,78 +2,122 @@
 
 ## Introduction
 Hopsworks services produce metrics which are centrally gathered by [Prometheus](https://prometheus.io/) and visualized in [Grafana](../grafana).
-Although the system is self-contained, it is possible to export these metrics to third-party services or another Prometheus instance.
+Although the system is self-contained, it is possible for another *federated* Prometheus instance to scrape these metrics or directly push them to another system.
 This is useful if you have a centralized monitoring system with already configured alerts.
 
 ## Prerequisites
-In order to configure Prometheus to export metrics you need `root` SSH access to either Hopsworks or to the target server depending on the export method you choose below.
+In order to configure Prometheus to export metrics you need to have the right to change the remote Prometheus configuration.
 
 ## Exporting metrics
 Prometheus can be configured to export metrics to another Prometheus instance (cross-service federation) or to a custom service which knows how to handle them.
 
 ### Prometheus federation
-Prometheus servers can be federated to scale better or to just clone all metrics (cross-service federation). Prometheus federation is well [documented](https://prometheus.io/docs/prometheus/latest/federation/#cross-service-federation)
-but there are some specificities to Hopsworks.
+Prometheus servers can be federated to scale better or to just clone all metrics (cross-service federation).
 
 In the guide below we assume **Prometheus A** is the service running in Hopsworks and **Prometheus B** is the server you want to clone metrics to.
 
 #### Step 1
-**Prometheus B** needs to be able to connect to TCP port `9089` of **Prometheus B** to scrape metrics. If you have any firewall (or Security Group) in place, allow ingress for that port.
+**Prometheus B** needs to be able to connect to TCP port `9090` of **Prometheus A** to scrape metrics. If you have any firewall (or Security Group) in place, allow ingress for that port.
 
 #### Step 2
-SSH into **Prometheus B** server, edit Prometheus configuration file and add the following under the `scrape_configs`
+The next step is to expose **Prometheus A** running inside Hopsworks Kubernetes cluster. If **Prometheus B** has direct access to **Prometheus A** then you can skip this step.
+
+We will create a Kubernetes *Service* of type *LoadBalancer* to expose port `9090`
+
+!!!Warning
+    If you need to apply custom **annotations**, then modify the Manifest below
+    The example below assumes Hopsworks is **installed** at Namespace *hopsworks*
+
+```bash
+kubectl apply -f - <<EOF
+apiVersion: v1
+kind: Service
+metadata:
+  name: prometheus-external
+  namespace: hopsworks
+  labels:
+    app: prometheus
+spec:
+  type: LoadBalancer
+  selector:
+    app.kubernetes.io/name: prometheus
+    app.kubernetes.io/component: server
+  ports:
+    - protocol: TCP
+      port: 9090
+      targetPort: 9090
+EOF
+```
+
+Then we need to find the External IP address of the newly created Service
+
+```bash
+export NAMESPACE=hopsworks
+kubectl -n $NAMESPACE get svc prometheus-external -ojsonpath='{.status.loadBalancer.ingress[0].ip}'
+```
+
+!!!Warning
+    It will take a few seconds until an IP address is assigned to the Service
+
+We will use this IP address in Step 2
+
+#### Step 2
+Edit the configuration file of **Prometheus B** server and append the following Job under `scrape_configs`
 
 !!! note
-    Replace IP_ADDRESS with the actual address of Hopsworks server
+    Replace IP_ADDRESS with the IP address from Step 1 or the IP address of Prometheus service if it is directly accessible.
+    The snippet below assumes Hopsworks services runs at Namespace **hopsworks**
 
 ```yaml
 - job_name: 'federate'
-    scrape_interval: 15s
+  scrape_interval: 15s
 
-    honor_labels: true
-    metrics_path: '/federate'
+  honor_labels: true
+  metrics_path: '/federate'
 
-    params:
-      'match[]':
-        - '{job="airflow"}'
-        - '{job="pushgateway"}'
-        - '{job="hadoop"}'
-        - '{job="hopsworks"}'
+  params:
+    'match[]':
+      - '{namespace="hopsworks"}'
 
-    static_configs:
-      - targets:
-        - 'IP_ADDRESS:9089'
+  static_configs:
+    - targets:
+      - 'IP_ADDRESS:9090'
 ```
 
-These are the basic labels gathered by Hopsworks.
+The configuration above will scrape for services metrics under the *hopsworks* Namespace. If you want to additionally
+scrape *user application* metrics then append `'{job="pushgateway"}'` to the matchers, for example:
 
-* If your Hopsworks cluster runs **without** Kubernetes append `'{job="cadvisor"}'` to `match[]` list
-
-* If your Hopsworks cluster runs **with** Kubernetes append the following labels to `match[]`
-    * `'{job=~"knative.+"}'`
-    * `'{job="kubernetes-cadvisor"}'`
-    * `'{job="istio-envoy"}'`
-    * `'{job="kube-state-metrics"}'`
-    * `'{job="cadvisor"}'`
-    * `'{job="cadvisor"}'`
-    * `'{job="cadvisor"}'`
+```yaml
+  params:
+    'match[]':
+      - '{namespace="hopsworks"}'
+      - '{job="pushgateway"}'
+```
 
-#### Step 3
-Finally restart Prometheus service with `sudo systemctl restart prometheus`
+Depending on the Prometheus setup you might need to restart **Prometheus B** service to pick up the new configuration.
+For more details on federation visit Prometheus [documentation](https://prometheus.io/docs/prometheus/latest/federation/#cross-service-federation)
 
 ### Custom service
 Prometheus can push metrics to another custom resource via HTTP. The custom service is responsible for handling the received metrics.
 To push metrics with this method we use the `remote_write` configuration.
 
-
 We will only give a sample configuration as `remote_write` is extensively documented in Prometheus [documentation](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write)
 In the example below we push metrics to a custom service listening on port 9096 which transforms the metrics and forwards them.
 
+In order to configure Prometheus to push metrics to a remote HTTP service we need to customize our Helm chart values file with the following snippet after changing the *url* accordingly. You can also tweak other configuration parameters to your needs.
+
 ```yaml
-remote_write:
-  - url: "http://localhost:9096"
-    queue_config:
-      capacity: 10000
-      max_samples_per_send: 5000
-      batch_send_deadline: 60s
+prometheus:
+  prometheus:
+    server:
+      remoteWrite:
+      - url: "http://localhost:9096"
+        queue_config:
+          capacity: 10000
+          max_samples_per_send: 5000
+          batch_send_deadline: 60s
 ```
+
+If the section already exists, then append the `remoteWrite` section.
+
+Run `helm install` or `helm upgrade` if it's the first time you install Hopsworks or you want to apply the change to an existing cluster respectively.
\ No newline at end of file