Deploy on managed hosting

Owned by ngit-graspVerified 3.0.1 @ 3f8157693fdeimported guide

The supplied container can run on Railway, Render, and Fly.io when each service has a custom domain, one persistent /data volume, and exactly one running instance. These platforms terminate TLS and proxy WebSockets to ngit-grasp.

Managed hosting is best for a small or moderate relay whose operator accepts brief upgrade downtime. For a large existing relay, use NixOS, systemd, or Compose so startup migrations and storage snapshots remain under direct operator control.

Read the deployment contract first. In particular, provider replicas do not make ngit-grasp highly available: their local volumes do not replicate application state.

Common requirements

For every provider:

• set NGIT_DOMAIN to the final custom hostname, not the provider hostname;
• mount durable storage at /data before the first successful start;
• keep the instance count at one and disable scale-to-zero;
• keep NGIT_BASE_PATH=/ unless path routing has been designed explicitly;
• allow at least five minutes between SIGTERM and forced termination; and
• take an external backup of /data, including .relay-owner.nsec.

The image maps a platform-provided PORT to 0.0.0.0:${PORT}. Do not set NGIT_BIND_ADDRESS unless the provider template below does so explicitly.

Railway

railway.toml selects the Dockerfile, uses / as the health check, disables old/new deployment overlap, and gives shutdown five minutes. Railway storage is configured separately from config-as-code. See Railway's config-as-code reference and volume guide for the provider-side details.

From the repository root with the Railway CLI installed:

railway login
railway init
railway add --service ngit-grasp
railway service ngit-grasp
railway variable set NGIT_DOMAIN=ngit.example.com NGIT_BASE_PATH=/
railway volume add --mount-path /data
railway up
railway domain ngit.example.com

Add the DNS records returned by railway domain, then inspect status and verify the public relay:

railway domain status ngit.example.com
railway deployment list --json
scripts/verify-deployment.sh https://ngit.example.com

Do not add replicas or multi-region configuration. Railway mounts a new volume as root; the image entrypoint prepares it and drops privileges before starting ngit-grasp.

Railway volume backups are useful recovery points, but also export or snapshot state outside the platform. During a large storage migration, increase the health-check timeout or move the upgrade to a directly operated host.

Render

render.yaml describes a paid starter web service with a 10 GB disk mounted at /data, a root health check, a five-minute shutdown delay, and automatic deploys disabled. Render prompts for NGIT_DOMAIN when the Blueprint is created. The fields follow Render's Blueprint specification; its persistent disk guide describes the storage and scaling constraints.

Render Blueprints require a repository connected through one of Render's supported Git providers. Until ngit-grasp has an authorized mirror or published OCI image, use Render's Public Git Repository flow with the canonical URL:

https://gitnostr.com/npub15qydau2hjma6ngxkl2cyar74wzyjshvl65za5k5rl69264ar2exs5cyejr/ngit-grasp.git

Choose Docker and apply the values from render.yaml in the service form:

• environment variable NGIT_DOMAIN=ngit.example.com;
• persistent disk mounted at /data;
• health check path /;
• maximum shutdown delay 300 seconds; and
• one instance with automatic deploys disabled.

Add and verify the custom domain before publishing it in repository announcements. Render services backed by a persistent disk cannot run multiple instances and have brief downtime during deploys, which matches the relay's single-writer requirement.

After a manual deploy, run:

scripts/verify-deployment.sh https://ngit.example.com

Use Render disk snapshots plus an independent backup. Only /data persists; all other container filesystem changes are ephemeral.

Fly.io

Fly volumes are local to one Machine and are not automatically replicated. The template therefore disables autostop, uses an immediate replacement strategy, and must be deployed with high-availability seeding disabled. See Fly's configuration reference and volume overview for the underlying platform behavior.

Copy and edit the template:

cp fly.toml.example fly.toml
# Set a unique app name, primary region, and NGIT_DOMAIN in fly.toml.
fly config validate --strict --config fly.toml

Create the app and one volume in the same region, then deploy one Machine:

fly apps create replace-with-a-unique-ngit-grasp-name
fly volumes create ngit_grasp_data --region ord --size 10
fly deploy --ha=false
fly scale count 1
fly certs add ngit.example.com

Use the app name and region selected in fly.toml, add the DNS records shown by fly certs, then verify:

fly checks list
fly scale show
scripts/verify-deployment.sh https://ngit.example.com

Never scale above one: Fly creates a separate empty volume for each additional Machine and does not copy ngit-grasp state. A single Machine and volume can be unavailable after host failure, so maintain an external backup rather than relying only on Fly snapshots.

Upgrades

Managed volume deployments cannot provide a safe zero-downtime writer handoff. Expect brief downtime:

1. disable automatic deploys;
2. take a complete provider snapshot and an external backup;
3. deploy the pinned revision with one instance;
4. inspect startup and integrity logs; and
5. run the public verifier.

For a storage-changing release, restore both the prior executable and its matching pre-upgrade /data snapshot when rolling back.