Scheduled tasks

Two periodic tasks live on the host, not in the containers: a trigger for the worker’s nightly run and the backup. They assume the repository is deployed in /srv/spellcache and that /srv/spellcache/.env is filled in (see Installation).

Install

sudo crontab -e

crontab

SHELL=/bin/bash
PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

# 03:15: nightly import and revaluation.
15 3 * * * cd /srv/spellcache && docker compose -f docker-compose.example.yml exec -T worker node apps/worker/src/index.ts --run-now >> /var/log/spellcache-nightly.log 2>&1

# 04:30: encrypted backup of the user tables, off the server.
30 4 * * * cd /srv/spellcache && set -a && source ./.env && set +a && ./deploy/backup.sh "$BACKUP_DEST" >> /var/log/spellcache-backup.log 2>&1

cron inherits no environment, so the backup line loads .env itself (BACKUP_DEST, BACKUP_PASSPHRASE, BACKUP_RETENTION_DAYS, DATABASE_URL). Without it, backup.sh stops at the first missing variable with an explicit message instead of writing an empty archive.

Database access for backup.sh

backup.sh runs on the host: it needs openssl, gzip, pg_dump and, for an ssh:// destination, the host’s SSH key.

sudo apt install postgresql-client-17

docker-compose.example.yml already publishes Postgres on 127.0.0.1:5432, the host’s loopback and nothing else. The host’s .env therefore holds:

DATABASE_URL=postgres://<POSTGRES_USER>:<POSTGRES_PASSWORD>@127.0.0.1:5432/<POSTGRES_DB>

Why two triggers for the same night

The worker container already has an internal scheduler (apps/worker/src/scheduler.ts, set by WORKER_NIGHTLY_AT and WORKER_TIMEZONE). The cron entry above is a second trigger, useful when the container was restarted outside its window, or to rerun a night by hand.

Both take the same Redis lock (spellcache:worker:lock:nightly) before queueing the job: the second one exits cleanly and logs that it was skipped; it does not start a second import.

[worker] nightly run skipped: another trigger holds the "nightly" lock
[worker] nightly trigger ignored, nothing queued

The lock expires after 6 hours. To force a rerun the same night:

docker compose -f docker-compose.example.yml exec redis valkey-cli DEL spellcache:worker:lock:nightly

Check

# did last night run?
docker compose -f docker-compose.example.yml logs --since 24h worker | grep nightly

# the last import runs (import_runs table, written by the job itself)
docker compose -f docker-compose.example.yml exec postgres \
  psql -U "$POSTGRES_USER" -d "$POSTGRES_DB" \
  -c 'select started_at, finished_at, status from import_runs order by started_at desc limit 5'

# the archives at the destination, and their age
ls -l --time-style=long-iso /path/of/BACKUP_DEST