fix(restic/esh-docker-vm): stop dumping paperless's Postgres from this host

Operator decision. paperless-ngx's database lives on esh-vm-db and is backed
up at the source by that host's fail-closed pg_dumpall. esh-docker-vm's
second copy had failed auth every night since 2026-04-24 behind a WARN. Its
`> paperless.pg_dump` redirect left a 0-byte file in every snapshot
(confirmed in snapshot 6ec9f74f), which looked like a dump but held nothing.

The block was the only consumer of /etc/restic/dbcreds.env, so the creds
loader is gone and the template dbcreds.env.example is deleted. The host
file was moved (not deleted) to /var/lib/restic/repair-20260923/.
Paperless's media volumes are still captured under /var/lib/docker/volumes.

Also fixes ownership. elway's sudo upload does scp-as-user then `sudo mv`,
so the hook deployed at 08:04 landed infra-ops:infra-ops even though root
executes it. Both esh-docker-vm playbooks now chown it back to root and
verify root:root 700.

Verified: the live hook hash matches canonical (e0d3ddcef1bddf43), and
the manual backup saved snapshot decfae71 with 3 staged dumps and no WARN
lines.
This commit is contained in:
vh
2026-09-23 09:18:03 -07:00
parent 25e41d2ab5
commit 6e8da46a28
5 changed files with 100 additions and 71 deletions
+20 -19
View File
@@ -29,25 +29,34 @@ Cross-site writes to `rest-server-ana` at `10.250.50.70:8000/esh-docker-vm/`.
## Pre-backup hooks
Unique to this host: most containers don't bundle sqlite3, so
`pre-backup.sh` runs sqlite3 and pg_dump **from the host** against the
volume bind-mount paths. Simpler than building custom images for HA
`pre-backup.sh` runs sqlite3 **from the host** against the volume
bind-mount paths. Simpler than building custom images for HA
and pgadmin. (uptime-kuma moved to ana-docker on 2026-09-22 and its block
was removed; see the header of `pre-backup.sh` for why a stale block took
the whole backup down.)
**paperless-ngx's database is NOT dumped here** (removed 2026-09-23, operator
decision). It lives on esh-vm-db (`10.0.50.60`) and is backed up at the source
by that host's fail-closed `pg_dumpall` — see `configs/restic/esh-vm-db/`.
The host-side `pg_dump` this hook used to run had failed auth every night
since 2026-04-24 behind a WARN, and its `> paperless.pg_dump` redirect left a
**0-byte file in every snapshot** that looked like a dump. It was the only consumer of
`/etc/restic/dbcreds.env`, so the creds file and its template went too.
Paperless's *media* volumes are still captured under `/var/lib/docker/volumes`.
| Service | DB | Approach |
|---|---|---|
| paperless-ngx | external Postgres `10.0.50.60` / `paperless-ng` | host pg_dump |
| home-assistant | `/var/lib/docker/.../homeassistant_v2.db` | host sqlite3 |
| pgadmin | `/var/lib/docker/.../pgadmin4.db` | host sqlite3 |
| calibre-web-automated | `/config/app.db` inside container | in-container sqlite3 (it has the binary) |
## Host prerequisites
Needs `sqlite3` and `postgresql-client` installed on esh-docker-vm:
Needs `sqlite3` installed on esh-docker-vm (`postgresql-client` was only for
the removed paperless dump; it is still installed, and harmless):
```bash
ssh -t esh-docker-vm 'sudo apt update && sudo apt install -y sqlite3 postgresql-client restic'
ssh -t esh-docker-vm 'sudo apt update && sudo apt install -y sqlite3 restic'
```
(restic too — not installed on this host yet.)
@@ -57,7 +66,7 @@ ssh -t esh-docker-vm 'sudo apt update && sudo apt install -y sqlite3 postgresql-
### 1. Install restic + db-client tooling
```bash
ssh -t esh-docker-vm 'sudo apt update && sudo apt install -y restic sqlite3 postgresql-client'
ssh -t esh-docker-vm 'sudo apt update && sudo apt install -y restic sqlite3'
```
### 2. Add `esh-docker-vm` entry on ana-docker rest-server
@@ -84,16 +93,6 @@ ssh -t esh-docker-vm 'sudo bash -c "cat > /etc/restic/restic.env && chmod 600 /e
ssh -t esh-docker-vm 'sudo bash -c "cat > /etc/restic/password && chmod 600 /etc/restic/password"'
# paste: <REPO-PASSPHRASE>
# Enter, Ctrl-D
# Install DB creds (from dbcreds.env.example — edit in a temp file first)
cp configs/restic/esh-docker-vm/dbcreds.env.example /tmp/dbcreds.env
${EDITOR:-vi} /tmp/dbcreds.env # set PAPERLESS_PGPASS to real value
scp /tmp/dbcreds.env esh-docker-vm:/tmp/
ssh -t esh-docker-vm '
sudo install -o root -g root -m 0600 /tmp/dbcreds.env /etc/restic/dbcreds.env &&
rm /tmp/dbcreds.env
'
shred -u /tmp/dbcreds.env 2>/dev/null || rm -f /tmp/dbcreds.env
```
### 4. Init the repo
@@ -126,7 +125,6 @@ ssh -t esh-docker-vm 'sudo ls -lh /var/lib/restic/stage/'
```
Expected files:
- `paperless.pg_dump` — should be 1–50 MB depending on doc count
- `home-assistant.sqlite3` — ~50 MB (matches live DB size)
- `calibre-web-automated.app.db` — ~250 KB
- `pgadmin4.db` — ~200 KB
@@ -182,8 +180,11 @@ ssh -t esh-docker-vm '
## Follow-ups after restic is proven (separate session)
- **Rotate paperless-ngx Postgres password.** Currently `paperless-ng` —
trivially weak. Update the DB, the compose, and `dbcreds.env`.
- **Rotate paperless-ngx Postgres password.** Was `paperless-ng` —
trivially weak. Update the DB and the compose (no copy lives on this
backup path any more). The removed backup copy failed auth from
2026-04-24 on, so it may already have changed at the DB side; confirm
before assuming.
- **Retire offen/docker-volume-backup sidecars** on paperless-ngx + pgadmin
stacks after ~1 week of clean restic runs. Delete the compose's
sidecar services + the tarballs under `/mnt/backup/docker/esh-vm-docker/`.
@@ -1,22 +0,0 @@
# /etc/restic/dbcreds.env on esh-docker-vm — consumed by pre-backup.sh.
#
# Deploy to the host as:
# sudo install -o root -g root -m 0600 dbcreds.env /etc/restic/dbcreds.env
#
# Never commit the real file — it carries production DB passwords. This
# template is the only thing tracked in the repo.
# --- Paperless-ngx → external Postgres on PFI-Postgres VM (10.0.50.60) ------
# NOTE (2026-04-21): the current password in the paperless-ngx compose is
# literally "paperless-ng" — trivially weak. Rotate at the DB side before
# this backup is considered secure.
PAPERLESS_PGHOST=10.0.50.60
PAPERLESS_PGPORT=5432
PAPERLESS_PGUSER=paperless-ng
PAPERLESS_PGDB=paperless-ng
PAPERLESS_PGPASS=replace-with-postgres-password
# --- All other services on this host use local SQLite ------------------------
# (home-assistant, pgadmin, uptime-kuma, calibre-web-automated)
# No external creds required — sqlite3 runs as root either on the host or
# inside the container depending on image.
+9 -30
View File
@@ -6,12 +6,10 @@
# whose raw volume files risk inconsistency during live restic capture.
#
# Unique approach for this host: most containers don't bundle sqlite3,
# so we run sqlite3 and pg_dump from the HOST against the volume
# bind-mount paths. Requires sqlite3 + postgresql-client installed
# on esh-docker-vm (apt install sqlite3 postgresql-client).
# so we run sqlite3 from the HOST against the volume bind-mount paths.
# Requires sqlite3 installed on esh-docker-vm (apt install sqlite3).
#
# Services handled:
# - paperless-ngx (external Postgres on 10.0.50.60 — pg_dump from host)
# - home-assistant (local SQLite in volume — sqlite3 .backup from host)
# - calibre-web-automated (local SQLite — sqlite3 .backup inside container, has sqlite3)
# - pgadmin (local SQLite in volume — sqlite3 .backup from host)
@@ -21,8 +19,13 @@
# `docker ps | grep` lookup exited 1 and set -e aborted this script, and
# resticprofile then skipped the WHOLE host backup (stale 09-22 → 09-23).
#
# External DB credentials live in /etc/restic/dbcreds.env (root:600).
# Template: configs/restic/esh-docker-vm/dbcreds.env.example.
# paperless-ngx's Postgres (on esh-vm-db, 10.0.50.60) was also dumped from
# here until 2026-09-23, when the block was removed by operator decision. It
# is backed up at the source by esh-vm-db's own fail-closed pg_dumpall, and
# this copy had failed auth every night since 2026-04-24 behind a WARN
# nobody read, leaving a 0-byte paperless.pg_dump in every snapshot. That
# block was the only consumer of /etc/restic/dbcreds.env, so the creds
# loader went with it.
#
# Errors in individual blocks log a WARN; whole script doesn't abort.
# That only holds if every lookup for an optional service sits inside an
@@ -41,11 +44,6 @@ warn() { log "WARN: $*" >&2; }
# Purge previous stage so stale dumps don't pile up in the snapshot.
find "$STAGE" -mindepth 1 -maxdepth 1 -exec rm -rf {} +
# Load external-DB creds
if [ -r /etc/restic/dbcreds.env ]; then
set -a; . /etc/restic/dbcreds.env; set +a
fi
# Helper: host-side sqlite .backup against a volume-bind path.
# $1 = source .db path (host absolute, typically under /var/lib/docker/volumes/.../_data/)
# $2 = stage filename (just the leaf name)
@@ -68,25 +66,6 @@ host_sqlite_backup() {
fi
}
# ---------- paperless-ngx (external Postgres on 10.0.50.60) -------------------
if docker inspect paperless-ngx-webserver-1 >/dev/null 2>&1; then
if [ -z "${PAPERLESS_PGPASS:-}" ]; then
warn "paperless-ngx: PAPERLESS_PGPASS unset in /etc/restic/dbcreds.env — skipping"
elif ! command -v pg_dump >/dev/null 2>&1; then
warn "paperless-ngx: pg_dump not installed — apt install postgresql-client"
else
log "dumping paperless postgres (${PAPERLESS_PGHOST}:${PAPERLESS_PGPORT:-5432})"
PGPASSWORD="$PAPERLESS_PGPASS" pg_dump \
-h "$PAPERLESS_PGHOST" -p "${PAPERLESS_PGPORT:-5432}" \
-U "$PAPERLESS_PGUSER" -d "$PAPERLESS_PGDB" \
-Fc --clean --if-exists \
> "$STAGE/paperless.pg_dump" \
|| warn "paperless pg_dump failed"
fi
else
log "skip paperless: container not present"
fi
# ---------- home-assistant (SQLite in named volume, host-side .backup) --------
# HA's DB is ~50MB and actively written. SQLite .backup is the proper way
# to grab a consistent snapshot while HA is running.
@@ -24,6 +24,12 @@ steps:
dest: /etc/restic/pre-backup.sh
mode: '0700'
# elway's sudo upload is scp-as-user then `sudo mv`, so the file lands owned
# by the SSH user. This hook is executed by root: it must be root-owned.
- name: Make the hook root-owned
sudo: true
shell: chown root:root /etc/restic/pre-backup.sh
verify:
- name: Hook parses under bash
sudo: true
@@ -36,3 +42,7 @@ verify:
- name: Hook runs to completion (stage summary line reached)
sudo: true
shell: /etc/restic/pre-backup.sh 2>&1 | grep -q 'stage ready:'
- name: Hook is root:root 0700
sudo: true
shell: test "$(stat -c '%U:%G %a' /etc/restic/pre-backup.sh)" = "root:root 700"
@@ -0,0 +1,61 @@
# esh-docker-vm: install the pre-backup hook without the paperless pg_dump
# block (operator decision 2026-09-23).
#
# paperless-ngx's Postgres is backed up at the source by esh-vm-db's
# fail-closed pg_dumpall. This host's second copy had been failing auth on a
# stale password for months behind a WARN. That block was the only consumer of
# /etc/restic/dbcreds.env, so the creds file is MOVED (not deleted) into the
# repair dir, out of the live config path.
#
# Rerunnable: the preserve/move steps are `creates:`-guarded; the chown step
# re-applies root ownership after every upload.
steps:
- name: Preserve the pre-change hook
sudo: true
shell: |
set -eu
install -d -m 0700 /var/lib/restic/repair-20260923
cp -p /etc/restic/pre-backup.sh /var/lib/restic/repair-20260923/pre-backup.sh.pre-paperless
creates: /var/lib/restic/repair-20260923/pre-backup.sh.pre-paperless
- name: Move the now-unused dbcreds.env out of /etc/restic
sudo: true
shell: mv /etc/restic/dbcreds.env /var/lib/restic/repair-20260923/dbcreds.env
creates: /var/lib/restic/repair-20260923/dbcreds.env
- name: Install the hook without the paperless block
sudo: true
upload:
src: configs/restic/esh-docker-vm/pre-backup.sh
dest: /etc/restic/pre-backup.sh
mode: '0700'
# elway's sudo upload is scp-as-user then `sudo mv`, so the file lands owned
# by the SSH user. This hook is executed by root: it must be root-owned.
- name: Make the hook root-owned
sudo: true
shell: chown root:root /etc/restic/pre-backup.sh /var/lib/restic/repair-20260923/pre-backup.sh.pre-paperless
verify:
- name: Hook parses under bash
sudo: true
shell: bash -n /etc/restic/pre-backup.sh
- name: No paperless dump or creds loader left in the live hook
sudo: true
shell: "! grep -q -E 'PAPERLESS_PG|dbcreds.env\"?$|pg_dump \\\\' /etc/restic/pre-backup.sh"
- name: dbcreds.env is gone from /etc/restic
sudo: true
shell: test ! -e /etc/restic/dbcreds.env
- name: Hook runs to completion with no WARN lines
sudo: true
shell: |
out=$(/etc/restic/pre-backup.sh 2>&1)
printf '%s\n' "$out" | grep -q 'stage ready:' && ! printf '%s\n' "$out" | grep -q 'WARN'
- name: Hook is root:root 0700
sudo: true
shell: test "$(stat -c '%U:%G %a' /etc/restic/pre-backup.sh)" = "root:root 700"