#!/bin/bash
#
# Weekly restic maintenance for both repos (Sunday, offset from the
# nightly backups — prune takes an EXCLUSIVE lock). Per repo:
#   - unlock: clear locks left by crashed runs
#   - prune: --max-unused 10% (our dead-data rate is ~0.1–0.4 GiB/day
#     vs ~1 T repos, so pruning can skip repacking for months) and
#     --max-repack-size 4G (bounds any single Sunday's rewrite to
#     ~10-15 min even after a mass deletion; the rest defers)
#   - check: structure every week, plus a rotating 1/52 data subset —
#     a full verification of every byte once a year
#
# Usage: restic-maintenance.sh [-r <repo>[,<repo>...]]

set -u

REPOS="s3:s3.sbg.io.cloud.ovh.net/restic-data,s3:s3.sbg.io.cloud.ovh.net/restic-incus"
LOCKFILE=/run/lock/restic-maintenance.lock
ENVFILE=/root/.restic-env
RESTIC=/usr/local/bin/restic   # cron PATH lacks /usr/local/bin

while getopts r: flag; do
    case "${flag}" in
        r) REPOS=${OPTARG};;
        *) echo "Usage: $0 [-r <repo>[,<repo>...]]" >&2; exit 2;;
    esac
done

[ -r "$ENVFILE" ] || { echo "env file $ENVFILE not readable" >&2; exit 2; }
. "$ENVFILE"

exec 9>"$LOCKFILE"
if ! flock -n 9; then
    echo "another restic-maintenance run holds $LOCKFILE, aborting" >&2
    exit 1
fi

log() { echo "[$(date '+%F %T')] $*"; }

# rotate the read-data subset weekly: full coverage once a year
WEEK=$(( ($(date +%s) / 604800) % 52 + 1 ))

rc=0
for repo in $(printf '%s' "$REPOS" | tr ',' ' '); do
    log "maintenance: $repo"
    $RESTIC -r "$repo" unlock || rc=1
    $RESTIC -r "$repo" prune --max-unused 10% --max-repack-size 4G --pack-size 64 || rc=1
    $RESTIC -r "$repo" check || rc=1
    log "check --read-data-subset=$WEEK/52"
    $RESTIC -r "$repo" check --read-data-subset="$WEEK/52" || rc=1
done

log "done (rc=$rc)"
exit $rc