#!/usr/bin/env bash
# infra/bin/fly-loop-reset — the milestone step of fly-loop-recover's ladder, as root.
#
# fly-loop-recover runs as User=fly and reaches this through one NOPASSWD sudoers line
# (config/fly-sudoers). It does what infra/docs/runbook.md "Restart the run from a rung"
# does by hand for a release whose adapter already wrote the archive: stop flysim, promote
# milestone-<N>.checkpoint with fly-reset-to-milestone, start flysim. flysim is started
# again whether or not the reset worked, so the stream never stays down on a failure;
# fly-reset-to-milestone copies both stores aside before it rewrites anything.
#
# Usage: fly-loop-reset <rung>
set -euo pipefail

: "${FLY_STATE_DIR:=/srv/fly/state}"
: "${FLY_SERVICE:=flysim.service}"
: "${FLY_RESET_BIN:=/opt/fly/bin/fly-reset-to-milestone}"

log() { echo "fly-loop-reset: $*" >&2; }

RANK="${1:-}"
if [ "$#" -ne 1 ] || ! [[ "$RANK" =~ ^[0-9]{1,2}$ ]]; then
    log "usage: fly-loop-reset <rung>"
    exit 2
fi
if [ ! -f "${FLY_STATE_DIR}/milestone-${RANK}.checkpoint" ]; then
    log "no milestone archive for rung ${RANK}; nothing touched"
    exit 1
fi

log "stopping ${FLY_SERVICE} to reset to rung ${RANK}"
systemctl stop "$FLY_SERVICE"
status=0
"$FLY_RESET_BIN" "$RANK" || status=$?
[ "$status" -eq 0 ] || log "fly-reset-to-milestone ${RANK} failed (exit ${status}); starting ${FLY_SERVICE} on the state it left"
systemctl start "$FLY_SERVICE"
exit "$status"