Mainnet Setup and Tooling
All of the hard work of the community has paid off, and now it's time to take the network live. Preparing your validator for mainnet involves a few extra considerations. They are detailed below, but a sensible checklist is:
- How will you handle chain upgrades?
- consider: Cosmovisor
- How will you know your node is up?
- consider: Monitoring and alerts
- How will you mitigate DDOS attacks?
- consider: Sentry Nodes
- How much storage will you need?
Answering these questions can be daunting, so there is some advice below.
In order to streamline chain upgrades and minimise downtime, you may want to set up cosmovisor to manage your node. A guide for this is provided in the Rebus docs.
Backups of chain state are possible using the commands specified here. If you are using a recent version of Cosmovisor, then the default configuration is that a state backup will be created before upgrades are applied. This can be turned off using environment flags.
Alerting and monitoring
Alerting and monitoring is desirable as well - you are encouraged to explore solutions and find one that works for your setup. Prometheus is available out-of-the box, and there are a variety of open-source tools. Recommended reading:
- Tenderduty: https://github.com/blockpane/tenderduty
- PANIC: https://github.com/SimplyVC/panic_cosmos
Simple setup using Grafana Cloud
Using only the raw metrics endpoint provided by
rebusd you can get a working dashboard and alerting setup using Grafana Cloud. This means you don't have to run Grafana on the instance.
- First, in
config.tomlenable Prometheus. The default metrics port will be
- Download Prometheus - this is needed to ship logs to Grafana Cloud.
- Create a
prometheus.ymlfile with your Grafana Cloud credentials in the Prometheus folder. You can get these via the Grafana UI. Click 'details' on the Prometheus card:
global: scrape_interval: 15s scrape_configs: - job_name: cosmops static_configs: - targets: ['localhost:26660'] labels: group: 'cosmops' remote_write: - url: https://your-grafana-cloud-endpoint/api/prom/push basic_auth: username: ID_HERE password: "API KEY HERE"
3. Set up a service file, with
sudo nano /etc/systemd/system/prometheus.service, replacing
<prometheus-folder> with the location of Prometheus. This sets the Prometheus port to
[Unit] Description=prometheus After=network-online.target [Service] User=<your-user> ExecStart=/home/<your-user>/<prometheus-folder>/prometheus --config.file=/home/<your-user>/<prometheus-folder>/prometheus.yml --web.listen-address=:6666 --storage.tsdb.path=/home/<your-user>/<prometheus-folder>/data Restart=always RestartSec=3 LimitNOFILE=4096 [Install] WantedBy=multi-user.target /etc/systemd/system/prometheus.service
4. Enable and start the service.
sudo -S systemctl daemon-reload sudo -S systemctl enable prometheus sudo systemctl start prometheus
5. Import a dashboard to your Grafana. Search for 'Cosmos Validator' to find several options. You should see logs arriving in the dashboard after a couple of minutes.
For more info:
Avoiding DDOS attacks
If you are comfortable with server ops, you might want to build out a Sentry Node Architecture validator to protect against DDOS attacks.
The current best practice for running mainnet nodes is a Sentry Node Architecture. There are various approaches, as detailed here. Some validators advocate co-locating all three nodes in virtual partitions on a single box, using Docker or other virtualisation tools. However, if in doubt, just run each node on a different server.
Bear in mind that Sentries can have pruning turned on, as outlined here. It is desirable, but not essential, to have pruning disabled on the validator node itself.
If you are using a cloud services provider, you may want to mount
$HOME on an externally mountable storage volume, as you may need to shuffle the data onto a larger storage device later. You can specify the
home directory in most commands, or just use symlinks.
Disk space is likely to fill up, so having a plan for managing storage is key.
If you are running sentry nodes:
- 1TB storage for the full node will give you a lot of runway
- 200GB each for the sentries with pruning should be sufficient
Managing backups is outside the scope of this documentation, but several validators keep public snapshots and backups.
To give you an idea of cost, on AWS EBS (other cloud providers are available, or you can run your own hardware), with two backups a day, this runs to roughly:
- $150 for 1TB
- $35 for 200GB
- Total cost: $220
What approach you take for this will depend on whether you are running on physical hardware co-located with you, running in a data centre, or running on virtualised hardware.