Skip to content

Latest commit

 

History

History
405 lines (253 loc) · 11 KB

README.md

File metadata and controls

405 lines (253 loc) · 11 KB
Raw

Hands on lab : Prometheus and Grafana

Slides here

0 - Introduction

Full setup (with workshop solutions)

Locally with Docker

Locally without Docker

Download Prometheus and official exporters: https://prometheus.io/download/

Download Grafana: https://grafana.com/grafana/download

1 - Metrics types

Take a look on Prometheus metric types (counter, gauges, histogram, summary) => https://prometheus.io/docs/concepts/metric_types/

2 - Start Prometheus

# Starts Prometheus
docker-compose up -d prometheus

# Starts system metrics exporter
docker-compose up -d node-exporter

3 - Let's grab some system metrics (memory, CPU, disk...)

Update prometheus.yml config file, to scrape node-exporter metrics every 10 seconds. 🚀

💡 Solution 
#
# /etc/prometheus/prometheus.yml
#

global:
scrape_interval: 30s

scrape_configs:
- job_name: 'node-exporter'
  scrape_interval: 10s
  static_configs:
    - targets: ['node-exporter:9100']

Then reload Prometheus with docker-compose exec prometheus kill -HUP 1 and see what happens here: http://localhost:9090/targets.

4 - Execute your first PromQL query

PromQL documentation:

4.0 - Memory usage

Go to http://localhost:9090/graph and write a query displaying a graph of free memory on your OS.

Metric name is node_memory_MemFree_bytes.

💡 Solution

Query: node_memory_MemTotal_bytes{}

4.1 - Human readable

Same metric but in GigaBytes ?

💡 Solution

Query: node_memory_MemTotal_bytes{} / 1024 / 1024 / 1024

4.2 - Relative to total memory

Same metric, but in percent of total available memory ?

Tips: node-exporter metrics are prefixed by node_.

💡 Solution

Query: (node_memory_MemTotal_bytes{} - node_memory_MemFree_bytes{}) / node_memory_MemTotal_bytes{} * 100

5 - Setup Grafana

Uncomment grafana in docker-compose.yml and launch it:

docker-compose up -d grafana

Open http://localhost:3000 (user: grep / pass: demo).

Add a new datasource to Grafana.

6 - Hand-made dashboard

Add a new dashboard to Grafana.

6.0 - Simple graph

Create a graph showing current memory usage.

💡 Solution

Query: (node_memory_MemTotal_bytes{} - node_memory_MemFree_bytes{}) / node_memory_MemTotal_bytes{} * 100

6.1 - Some formatting

Grafana should be displaying graph in %, such as:

💡 Solution

6.2 - CPU load

In the same dashboard, add a new graph for CPU load (1min, 5min, 15min).

Tips: you will need a new metric prefixed by node_.

💡 Solution

6.3 - Disk usage

In the same dashboard, add a new graph for sda disk usage (ko written per second).

You will need rate() PromQL function: https://prometheus.io/docs/prometheus/latest/querying/functions/#rate

💡 Solution

Query: rate(node_disk_written_bytes_total{device="sda"}[30s])

7 - Dashboards from community

Let's import a dashboard from Grafana website.

Those dashboards are only compatible with Prometheus data-source and node-exporter.

8 - Monitor services: nginx, postgresql...

8.1 - Export Nginx and PostgreSQL metrics

Uncomment postgres, postgresql-exporter and nginx-exporter services in docker-compose.yml, and launch containers.

docker-compose up -d nginx-exporter
docker-compose up -d postgres postgresql-exporter

Update Prometheus configuration to scrape Nginx and PostgreSQL exporters.

💡 Solution 
scrape_config:

[...]

- job_name: 'postgresql-exporter'
  static_configs:
    - targets: ['postgresql-exporter:9187']

- job_name: 'nginx-exporter'
  static_configs:
    - targets: ['nginx-exporter:9101']

Then docker-compose exec prometheus kill -HUP 1

Check everything is working well here: http://localhost:9090/targets

Take a look on /metrics routes of exporters: http://localhost:9187/metrics + http://localhost:9101/metrics

8.2 - Generate some metrics

Send tens of requests to Nginx on localhost:8080 (200, 404...) and fill PostgreSQL database:

# 2xx
./infinite-200-req.sh

# 4xx
./infinite-404-req.sh
# inserts data into pg
./infinite-pg-insert.sh

8.3 - Import PG dashboards to Grafana

Go on https://grafana.com/dashboards and find a dashboard for PostgreSQL, compatible with Prometheus and wrouesnel/postgres_exporter.

💡 Solution

Those exporters looks nice: https://grafana.com/dashboards/6742, https://grafana.com/dashboards/6995.

8.4 - Create Nginx dashboards

Display 2 graphs:

  • number of 2xx http requests per second

  • number of 4xx http requests per second

Tips: you should use sum by(<label>) (<metric>) and irate(<metric>) (cf PromQL doc).

💡 Solution

Query graph 1: sum by (status) (irate(nginx_http_requests_total{status=~"2.."}[1m]))

Legend graph 1: Status: {{ status }}

Query graph 2: sum by (status) (irate(nginx_http_requests_total{status=~"4.."}[1m]))

Legend graph 2: Status: {{ status }}

9 - Export some business metrics

Let's display in real time:

  • number of users
  • number of posts per user

9.0 - Export data

Grab custom metrics with postgresql-exporter by adding queries to custom-queries.yml:

  • Metric user_count of type counter => SELECT COUNT(*) FROM users;
  • Metric post_per_user_count of type gauge with user_id and email in labels => SELECT u.id, u.email, COUNT(*) FROM posts p JOIN users u ON u.id = p.user_id GROUP BY u.id;

Example and syntax here.

http://localhost:9187/metrics should output:

[...]

# HELP user_count_count Number of users
# TYPE user_count_count counter
user_count_count 2

# HELP post_per_user_count_count Number of posts per user
# TYPE post_per_user_count_count gauge
post_per_user_count_count{email="[email protected]",id="e1c10ca1-60c8-405c-a9f3-3ff41456ca9f"} 1
post_per_user_count_count{email="[email protected]",id="fde08ee6-5fb9-4c4f-9b40-dc2ad69bb855"} 2

[...]
💡 Solution

Append to custom-queries.yml:

user:
  query: "SELECT COUNT(*) FROM users;"
  metrics:
    - count:
        usage: "COUNTER"
        description: "Number of users"

post_per_user:
  query: "SELECT u.id, u.email, COUNT(*) FROM posts p JOIN users u ON u.id = p.user_id GROUP BY u.id;"
  metrics:
    - count:
        usage: "GAUGE"
        description: "Number of posts per user"
    - id:
        usage: "LABEL"
        description: "User id"
    - email:
        usage: "LABEL"
        description: "User email"

9.1 - Graph time!

With user_count{} and post_per_user_count{id,email} metrics, build following graphs:

Simple graph of users signup (rate(<metric>)):

imgs/grafana-user-signups.png

Heatmap of signups (increase(<metric>)):

docker-compose exec grafana grafana-cli plugins install petrslavotinek-carpetplot-panel
docker-compose restart grafana

Table of top 10 users per post count (topk(), sum by(<label>) (<metric>)):

💡 Solution

Query 1: rate(user_count{}[1m])

Query 2: increase(user_count{}[$__interval]) > 0

Query 3: topk(10, sum by (id, email) (post_per_user_count{}) > 0)

9.2 - Expose /metrics from a micro-service

You can play with this sample in NodeJS: microservice-demo/README.md.

Don't forget to update Prometheus configuration in prometheus.yml !

42 - More

  • Monitor a Redis server, a RabbitMQ cluster, Mysql...
  • Increase data retention (15d by default).
  • Setup alerting with AlertManager and basic rules
  • Setup Prometheus service discovery (consul, etc, dns...) to import configuration automatically
  • Limits: multitenancy - partitionning/sharding - scaling - cron tasks