#!/usr/bin/env bash
#
# Pull the latest production Postgres dump off the DB server and restore
# it into the local `prop_dev` database, replacing whatever is there.
#
# The production server runs scripts/backup_prop_db.sh on a daily cron,
# which writes pg_dump custom-format dumps to /backups/prop/ as
# `prop_YYYYMMDD_HHMMSS.dump`. We SSH in, pick the newest one, scp it
# locally, then drop/recreate prop_dev and restore into it.
#
# Usage:
#
#   scripts/restore_prod_to_local.sh
#   REMOTE_HOST=my-db-host REMOTE_USER=graham scripts/restore_prod_to_local.sh
#
# Before running: stop `mix phx.server` so Ecto doesn't hold connections.

set -euo pipefail

REMOTE_HOST="${REMOTE_HOST:-db-prop}"
REMOTE_USER="${REMOTE_USER:-$USER}"
REMOTE_BACKUP_DIR="${REMOTE_BACKUP_DIR:-/backups/prop}"
LOCAL_DB="${LOCAL_DB:-prop_dev}"
LOCAL_USER="${LOCAL_USER:-postgres}"
LOCAL_HOST="${LOCAL_HOST:-localhost}"
LOCAL_TMP="${LOCAL_TMP:-/tmp}"

ssh_target="${REMOTE_USER}@${REMOTE_HOST}"

log() {
    printf '%s restore_prod_to_local: %s\n' "$(date -u +'%Y-%m-%dT%H:%M:%SZ')" "$*"
}

die() {
    log "ERROR: $*"
    exit 1
}

command -v ssh >/dev/null || die "ssh not on PATH"
command -v scp >/dev/null || die "scp not on PATH"
command -v pg_restore >/dev/null || die "pg_restore not on PATH (install postgresql client tools)"
command -v psql >/dev/null || die "psql not on PATH"

log "looking up latest dump on ${ssh_target}:${REMOTE_BACKUP_DIR}"
remote_file=$(ssh "$ssh_target" "ls -1t ${REMOTE_BACKUP_DIR}/prop_*.dump 2>/dev/null | head -n1")
[[ -n "$remote_file" ]] || die "no prop_*.dump files found in ${REMOTE_BACKUP_DIR} on ${REMOTE_HOST}"

filename=$(basename "$remote_file")
local_file="${LOCAL_TMP}/${filename}"
log "latest dump: ${remote_file}"

# Always remove the downloaded dump, even on mid-run failure. Saves disk
# on laptops where dumps are multi-GB.
cleanup() {
    [[ -n "${local_file:-}" && -f "$local_file" ]] && rm -f "$local_file"
}
trap cleanup EXIT

log "downloading to ${local_file}"
scp "${ssh_target}:${remote_file}" "$local_file"

# Kick any existing connections off the DB before dropping it — otherwise
# `DROP DATABASE` fails with "is being accessed by other users". This is
# aggressive, but it's a dev box and the user was told to stop the server.
log "terminating existing connections to ${LOCAL_DB}"
psql -h "$LOCAL_HOST" -U "$LOCAL_USER" -d postgres -v ON_ERROR_STOP=1 <<SQL >/dev/null
SELECT pg_terminate_backend(pid)
FROM pg_stat_activity
WHERE datname = '${LOCAL_DB}' AND pid <> pg_backend_pid();
SQL

log "dropping and recreating ${LOCAL_DB}"
psql -h "$LOCAL_HOST" -U "$LOCAL_USER" -d postgres -v ON_ERROR_STOP=1 <<SQL >/dev/null
DROP DATABASE IF EXISTS ${LOCAL_DB};
CREATE DATABASE ${LOCAL_DB} OWNER ${LOCAL_USER};
SQL

# --no-owner / --no-acl so ownership in the prod dump doesn't cause role
# errors locally. -j 4 parallelises table data restore.
log "restoring dump into ${LOCAL_DB}"
pg_restore \
    -h "$LOCAL_HOST" \
    -U "$LOCAL_USER" \
    -d "$LOCAL_DB" \
    --no-owner \
    --no-acl \
    -j 4 \
    "$local_file"

log "done — ${LOCAL_DB} is a copy of prod as of $(basename "$filename" .dump)"