deploy.sh reference

RequirementsCopy Anchor Link

curl -fsSL -o deploy.sh https://raw.githubusercontent.com/marvinli001/edgeweir/master/deploy.sh
sudo bash deploy.sh install

Save the script to a file before running it: through a pipe (curl … | bash) or process substitution (bash <(curl …)) it exits with 1. After install, a copy lives at <install directory>/deploy.sh; run the other commands there as ./deploy.sh <command>. Full flow on BaoTa / aaPanel: baota.en.md.

CommandsCopy Anchor Link

Deployment directoryCopy Anchor Link

Commands other than install, template, and help look for the deployment directory in this order and use the first match:

1. EDGEWEIR_DIR
2. The script's directory
3. The working directory
4. /www/dk_project/edgeweir
5. /opt/edgeweir

installCopy Anchor Link

PromptsCopy Anchor Link

Variables supply the defaults; in unattended mode they are the answers. A set EDGEWEIR_DB or DATABASE_URL skips the matching prompt. Loopback means 127.*, localhost, or ::1. The prompts are in Chinese.

ValidationCopy Anchor Link

ResultCopy Anchor Link

1. Resolve and pull the image version, see Version resolution.
2. Create the install directory (700), write .env (600), compose.yml (600), .compose.cksum, and deploy.sh (700).
3. Pull the compose images, start, and wait for health checks.
4. Print the running version, next steps, and the setup token.

Unattended installCopy Anchor Link

When EDGEWEIR_YES is non-empty, or /dev/tty cannot be opened, the script reads no input: each prompt takes its default or the variable below.

EDGEWEIR_YES=1 EDGEWEIR_DB=bundled \
EDGEWEIR_PUBLIC_URL=https://cdn-admin.example.com \
bash deploy.sh install

EDGEWEIR_NO_PULL=1 needs the target tag (or latest) of the console image and postgres:18.6-alpine (host mode checks and backups, the bundled database) on the host.

Database checkCopy Anchor Link

In host mode the database is checked before any file is written. The check runs psql from postgres:18.6-alpine (pinned by digest, as in compose.baota.yml) on the host network; the password travels in the environment; the connect timeout is 8 seconds.

The check and backups connect according to sslmode in DATABASE_URL:

updateCopy Anchor Link

./deploy.sh update [tag] [--no-backup] runs these steps:

1. Resolve the target version, see Version resolution. When the target equals EDGEWEIR_VERSION in .env and the container already runs it, print 已经是 <version>。 (already at) and exit 0.

2. A target other than latest that is older than the current version is a rollback: warn and ask whether to continue, default no. The dates in the tags are compared first; two tags of the same day compare the commit times of the two images (label org.opencontainers.image.created). When the order cannot be told (a tag is not <YYYYMMDD>-<commit>, or one image of the same day is not local), only a note is printed.

3. Back up to backups/<time>-before-<target version>/; --no-backup skips this. A failed backup aborts with the deployment unchanged.

4. When the compose file differs from the built-in template:

   • Same as the script last wrote it (per .compose.cksum): replaced with the new template.
   • Edited since: the difference is shown and the script asks whether to replace it, default no.
   • No .compose.cksum (older installs): the difference is shown and the script asks, default yes interactively, kept in unattended mode.

   The old file is in the step 3 backup; local changes belong in the override file; ./deploy.sh template <mode> prints the template.

5. Write EDGEWEIR_VERSION and recreate the containers with the start flow. Database migrations run when the console starts.

6. Print the rollback command ./deploy.sh update <previous version> (valid only when no migration was added between the two versions).

7. When /app/deploy.sh in the new image differs from this script, ask whether to replace this script, default yes.

Version policy and rollback constraints: upgrade.en.md.

Version resolutionCopy Anchor Link

install resolves EDGEWEIR_VERSION (default latest) with the same rules.

configCopy Anchor Link

Interactive only; in unattended mode or without a terminal it aborts; edit .env directly and run ./deploy.sh start instead.

1. Ask for the console URL and node channel URL; Enter keeps the current value. Validation as in install.
2. When the port in the node channel URL changes: an EDGEWEIR_NODE_API_PORT that is unset or equal to the old URL's port follows it, with a warning to open the new port and re-enroll nodes with the new URL; a value set apart (e.g. 127.0.0.1:18443 behind an nginx stream) is kept, with a note to adjust it yourself.
3. When the node channel host name changes, warn: enrolled nodes re-enroll, or add the old host name to EDGEWEIR_NODE_API_HOSTNAMES.
4. After confirmation, write EDGEWEIR_PUBLIC_URL, EDGEWEIR_NODE_API_URL, and EDGEWEIR_NODE_API_PORT when step 2 changes it, then recreate the containers with the start flow.

config does not change EDGEWEIR_HTTP_PORT and does not check whether the new port is in use. A URL saved in the console under System settings → Node channel wins over EDGEWEIR_NODE_API_URL: once one is saved, the node channel URL config writes does not reach install commands; to change only the node channel URL, change it in System settings, which needs no restart; see Node channel URL and certificate.

Start, stop, and trusted proxy syncCopy Anchor Link

install, update, and config use the start flow.

Meaning of trusted proxies: networking.en.md.

After a host mode start, the script checks where the node channel listens, in this order, and stops at the first warning:

self-updateCopy Anchor Link

Identical content is left alone. A replacement is written to a new file (700) and renamed over this script; the running process keeps reading the old file.

Generated filesCopy Anchor Link

Values are unquoted. Changing a key keeps the other lines and mode 600. Other variables: Environment variables.

BackupsCopy Anchor Link

<dir>/backups/
└── 20260929-153000-before-20260930-b2c3d4e/
    ├── edgeweir.dump
    ├── env
    ├── compose.yml
    └── compose.override.yml   # when present

Restore with restore; manual restores and restore acceptance: backup.en.md.

restoreCopy Anchor Link

./deploy.sh restore <backup> [--no-backup] replaces the deployment's database with a backup. <backup> is a backup directory (its edgeweir.dump is used) or a dump file, looked up as given, then relative to the deployment directory, then relative to backups/, e.g. ./deploy.sh restore 20261001-080000.

1. Check the backup: bundled starts postgres first; pg_restore --list must read it, and it must hold table data and drizzle.__drizzle_migrations (a console database), or the command aborts.
2. host: the DATABASE_URL user must own the database (or be a superuser) and have CREATEDB, or the command aborts; restore into a new database by hand instead.
3. Confirm, default no; unattended: see confirmations.
4. Back up the current database to backups/<time>-before-restore/ without removing older backups; --no-backup skips this. A failed backup aborts with the deployment unchanged.
5. Stop console.
6. Drop and recreate the database: DROP DATABASE … WITH (FORCE) and CREATE DATABASE. bundled runs them in the postgres container as edgeweir; host runs them with postgres:18.6-alpine against the postgres maintenance database, and the new database belongs to the DATABASE_URL user.
7. Import with pg_restore --exit-on-error --single-transaction --no-owner --no-privileges.
8. Start and wait for the health checks as start does.

Exit and abort behaviorCopy Anchor Link

template and setup-token write their result to standard output; progress, prompts, and errors go to standard error, in color when standard error is a terminal.