V2
Ask or search…
K

Setting up a monitoring stack for Boost

This tutorial goes through the steps required to run our Docker monitoring setup to collect and visualize metrics for various Boost processes

Background

The monitoring stack we will use includes:
  • Prometheus - collects metrics and powers dashboards in Grafana
  • Tempo - collects traces and powers traces search in Grafana with Jaeger
  • Grafana - provides visualization tools and dashboards for all metrics and traces
Lotus and Boost are already instrumented to produce traces and stats for Prometheus to collect.
The Boost team also packages a set of Grafana dashboards that are automatically provisioned as part of this setup.
The monitoring setup should be done on the same node where boostd and other Boost services are running. If you are running multiple Boost services (like boostd-data/booster-http) across multiple machines then any one of those machine can be used.

Prerequisites

This setup has been tested on macOS and on Linux. We haven’t tested it on Windows, so YMMV.
All the monitoring stack containers run in Docker.

Steps

  1. 1.
    Install Docker
We have tested this setup with Docker 20.10.23 on macOS and Ubuntu.
  1. 2.
    DNS resolution for Prometheus
Update extra_hosts in docker-compose.yaml for prometheus, so that the Prometheus container can reach all its targets - boostd, lotus-miner, booster-bitswap, booster-http, etc. https://github.com/filecoin-project/boost/blob/main/docker/monitoring/docker-compose.yaml#L54-L58
Depending on where your Filecoin processes (boostd, lotus, lotus-miner, booster-bitswap, etc.) are running, you need to confirm that they are reachable from Prometheus so that it can scrape their metrics.
By default the setup expects to find them within the same Docker network, so if you are running them elsewhere (i.e. on the `host` network), make sure to update the docker-compose file for the same.
  1. 3.
    Loki plugin for Docker
The loki plugin for docker is required to allow collecting logs from the services running on docker itself
  1. 4.
    Storage location
The prometheus and tempo services requires access to local storage to ensure all historical data is safe across the docker restarts. This path can be defined in the env file.
The default value is $HOME/.boost-monitoring
Please ensure that the storage directory exists and has permission 0775 or 0777 depending on your user.
mkdir ~/.boost-monitoring
chmod 0777 ~/.boost-monitoring
  1. 5.
    Update metrics endpoints for Prometheus
In case, you are running all Boost services and monitoring stack on the same node, you can skip this step.
If any of your Boost services like boostd-data or booster-http are running on a different host then we must modify the prometheus configuration to update the endpoint to be scraped. Please edit https://github.com/filecoin-project/boost/blob/main/docker/monitoring/prometheus.yaml file and update the IP addresses and port to scrap the metrics. Example:
- job_name: 'booster-bitswap'
static_configs:
- targets: [ '<BOOSTER_BITSWAP IP>:<METRICS PORT>' ]
labels:
host: 'local'
  1. 6.
    Start the monitoring stack using docker-compose
cd docker/monitoring
docker-compose up -d

Verify Prometheus Targets

Confirm that Prometheus targets are scraped at http://localhost:9190 Targets
You can also access the Grafana on the localhost interface on a remote server, you can open an SSH tunnel from your local machine:
ssh -L 9190:localhost:9190 myserver
If you are running software firewall like `ufw`, you might need to modify your iptables and allow access from the Prometheus container / network to the Filecoin stack network, for example:
sudo docker network inspect monitoring # note the Subnet for the network sudo ufw allow from 172.18.0.0/16

Grafana dashboards

Go to Grafana at http://localhost:3333 and inspect the dashboards.
You can also access the Grafana on the localhost interface on a remote server, you can open an SSH tunnel from your local machine:
ssh -L 3333:localhost:3333 myserver
The default username and password for Grafana dashboard is "admin" and "admin" respectively.