docs

Backups ​

backup_stream.py runs pg_dump inside the Postgres container and pipes its stdout straight into an S3 multipart upload — no intermediate dump file on disk.

python
proc = subprocess.Popen(
    ["docker", "exec", CONTAINER, "pg_dump", "-U", DB_USER, "-d", DB_NAME, "-F", "c"],
    stdout=subprocess.PIPE,
)
s3.upload_fileobj(proc.stdout, BUCKET, key)

Object key: backups/<db>-<UTC timestamp>.dump. Credentials come from Infisical, not env vars checked into the repo.

restore.py reverses the flow: pull the object back from S3 and replay it with pg_restore.