Skip to content

Free, Open Source, Self-Hosted WhatsApp API Gateway

fix(scripts): keep an in-container backup's staging off the tmpfs

Yudhi Armyndharisd8811fec012e
diff --git a/CHANGELOG.md b/CHANGELOG.md
index b1fc0882a..9963c5f0b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -119,7 +119,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
- Ingress manifests with an ack header value Node cannot send, a non-final ack status, a non-numeric `toleranceSec`, or a route that cannot travel as one URL path segment are refused at load, and minted ingress URLs percent-encode the route.
- `GET` and `DELETE /mcp` answer `405` instead of `404`, so Streamable HTTP clients stop reporting an SSE error on connect.
- A mixed-case or padded `POSTGRES_SCHEMA` is refused at boot, by the init script, by the migration CLI and on the Infrastructure page, instead of splitting the tables across two schemas.
-- `backup.sh` fails before staging anything when `BACKUP_DIR` is not writable, such as the default on the container's read-only root.
+- `backup.sh` fails before staging anything when `BACKUP_DIR` is not writable, such as the default on the container's read-only root, and names the `BACKUP_DIR` to use inside the container, plus the `TMPDIR` under docker compose, whose `/tmp` is memory-backed.
- Dashboard: the multi-group send refuses an empty message or a missing media source and stops on a `409` instead of repeating the failure for every group.
- Dashboard: the Message Tester follows a selected session that drops out of the ready list instead of sending to it.
- Dashboard: a bulk send no longer starts polling its progress after the page is left.
diff --git a/docs/11-operational-runbooks.md b/docs/11-operational-runbooks.md
index 02aa2c329..089117a2e 100644
--- a/docs/11-operational-runbooks.md
+++ b/docs/11-operational-runbooks.md
@@ -596,12 +596,19 @@ OPENWA_DATA_DIR=/srv/openwa/data \
> volume as the data does not survive losing that volume:
>
> ```bash
-> docker exec -e BACKUP_DIR=/app/data/backups openwa-api ./scripts/backup.sh
+> docker exec -e BACKUP_DIR=/app/data/backups -e TMPDIR=/app/data/backups openwa-api ./scripts/backup.sh
> docker cp openwa-api:/app/data/backups/. ./backups/
> # Helm: kubectl exec <pod> -- env BACKUP_DIR=/app/data/backups ./scripts/backup.sh
> # kubectl cp <pod>:/app/data/backups ./backups
> ```
>
+> The script stages a full copy of the data in `TMPDIR` before archiving it. The compose file mounts
+> `/tmp` as a tmpfs charged to the container's memory limit, so the compose line points `TMPDIR` at
+> the data volume, which then needs free space for about the size of the data plus the archive;
+> staging in the tmpfs gets the running gateway OOM-killed. The Helm chart's `/tmp` is an `emptyDir`
+> on node disk, so the Helm lines leave it alone. A run killed outright, such as by a container
+> restart mid-backup, leaves its `tmp.*` staging directory behind in `TMPDIR`; delete it.
+>
> The scripts resolve every other path the way the application does: an explicit environment value
> first, then `./.env`, then `<data dir>/.env.generated`. Settings made through Dashboard >
> Infrastructure therefore apply without being restated on the command line. Two caveats when
diff --git a/scripts/backup.sh b/scripts/backup.sh
index 06cb621ae..5c836b047 100755
--- a/scripts/backup.sh
+++ b/scripts/backup.sh
@@ -79,12 +79,15 @@ log() { echo "[backup] $*"; }
# Check the destination before staging anything. The shipped container's root filesystem is
# read-only, so the default ./backups (/app/backups) cannot be created there; failing only at the
-# end would first copy every database, session and media file into the memory-backed /tmp.
+# end would first copy every database, session and media file into the staging directory.
if ! mkdir -p "$BACKUP_DIR" 2>/dev/null || [ ! -w "$BACKUP_DIR" ]; then
- log "ERROR: BACKUP_DIR=$BACKUP_DIR is not writable; point BACKUP_DIR at a writable, persistent directory (inside the read-only container use BACKUP_DIR=/app/data/backups)"
+ log "ERROR: BACKUP_DIR=$BACKUP_DIR is not writable; point BACKUP_DIR at a writable, persistent directory (inside the container use BACKUP_DIR=/app/data/backups, and under docker compose also TMPDIR=/app/data/backups: its /tmp is a tmpfs charged to the container's memory)"
exit 1
fi
+# The staging copy goes to TMPDIR. Under docker compose that is a tmpfs charged to the container's
+# memory limit, so an in-container run there points TMPDIR at the data volume (docs/11); staging the
+# data in the tmpfs gets the running gateway OOM-killed once the data outgrows its headroom.
STAGE="$(mktemp -d)"
trap 'rm -rf "$STAGE"' EXIT