hyperhive/nix/script-tests/refresh-consumer.sh

# Cases for ./refresh-consumer.nix. Each case sets the credential file's mtime
# and the consumer's last start, runs the helper with a stub `systemctl` that
# records every call it is asked to act on, and compares that record and what
# the helper wrote to stderr.
set -uo pipefail

failures=0
count=0

fail() {
  echo "FAILED: $case: $*" >&2
  failures=$((failures + 1))
}

dir=$(mktemp -d)
mkdir "$dir/bin"
calls=$dir/calls
path=$dir/token
: >"$path"

# `show` prints $STARTED as systemd's `--timestamp=us+utc` does; `is-active`
# and `is-failed` answer from $ACTIVE and $FAILED; every other verb is
# appended to $calls.
cat >"$dir/bin/systemctl" <<STUB
#!$BASH
args=" \$* "
case "\$args" in
  *" show "*) printf '%s\n' "\$STARTED" ;;
  *" is-active "*)
    case "\$args" in
      *" container@"*) [ "\$ACTIVE" = 1 ] ;;
      *) [ "\$UNIT_ACTIVE" = 1 ] ;;
    esac ;;
  *" is-failed "*) [ "\$FAILED" = 1 ] ;;
  *) printf '%s\n' "\$*" >>"$calls" ;;
esac
STUB
chmod +x "$dir/bin/systemctl"
export PATH="$dir/bin:$PATH" ACTIVE=1 UNIT_ACTIVE=1 FAILED=0 STARTED=

export LC_ALL=de_DE.UTF-8

# shellcheck source=/dev/null
. "$HELPER"

# The locale the cases run under has to print the comma the helper must not
# depend on, or none of them tests it.
case="setup"
touch -d @1791232888.193159 "$path"
got=$(stat -c %.6Y "$path")
if [ "$got" != "1791232888,193159" ]; then fail "stat under $LC_ALL printed '$got', wanted 1791232888,193159"; fi

# systemd's `--timestamp=us+utc` rendering of <epoch seconds>.<micros>.
systemd_ts() {
  LC_ALL=C date -u -d "@$1" '+%a %Y-%m-%d %H:%M:%S.%6N UTC'
}

# check <description> <want systemctl calls, one per line, or ""> <want stderr, or "">
#       <helper call…>
check() {
  case=$1
  local want_calls=$2 want_err=$3 got err
  shift 3
  count=$((count + 1))
  : >"$calls"
  # `set -e` as the callers' NixOS `script`s run: a failed read must not end the
  # caller. Not under `||` or `if`, which would switch `set -e` off.
  local rc
  (
    set -e
    "$@" >/dev/null 2>"$dir/err"
  )
  rc=$?
  if [ "$rc" != 0 ]; then fail "exited $rc"; fi
  err=$(cat "$dir/err")
  got=$(cat "$calls")
  if [ "$got" != "$want_calls" ]; then
    fail "systemctl was asked:"$'\n'"$got"$'\n'"wanted:"$'\n'"$want_calls"
  fi
  # A failed read also leaves `date`'s own message ahead of the helper's line.
  if [ -n "$want_err" ]; then err=$(printf '%s\n' "$err" | tail -n 1); fi
  if [ "$err" != "$want_err" ]; then
    fail "stderr:"$'\n'"$err"$'\n'"wanted:"$'\n'"$want_err"
  fi
}

touch -d @1791232888.193159 "$path"

STARTED=$(systemd_ts 1791232900.000001)
check "a file older than the start restarts nothing" "" "" \
  refresh_unit gitea-runner-swarm.service "$path"
check "nor through a container" "" "" \
  refresh_consumer swarm-ci-runner gitea-runner-swarm.service "$path"

STARTED=$(systemd_ts 1791232888.193159)
check "a file written in the start's microsecond restarts nothing" "" "" \
  refresh_unit gitea-runner-swarm.service "$path"

STARTED=$(systemd_ts 1791232888.193158)
check "a file one microsecond newer than the start restarts the unit" \
  "try-restart --no-block gitea-runner-swarm.service" "" \
  refresh_unit gitea-runner-swarm.service "$path"
check "through a container, with --machine" \
  "--machine=swarm-ci-runner try-restart --no-block gitea-runner-swarm.service" "" \
  refresh_consumer swarm-ci-runner gitea-runner-swarm.service "$path"

FAILED=1
check "a failed unit is reset and started" \
  "reset-failed gitea-runner-swarm.service"$'\n'"start --no-block gitea-runner-swarm.service" "" \
  refresh_unit gitea-runner-swarm.service "$path"
FAILED=0

STARTED=$(systemd_ts 1791232888.193158)
got=$(refresh_unit gitea-runner-swarm.service "$path" 2>&1)
case="control: a running unit is said to be restarted"
count=$((count + 1))
if [ "$got" != "$path changed after gitea-runner-swarm.service last started — restarting it" ]; then fail "logged: $got"; fi

STARTED=
UNIT_ACTIVE=0
check "a unit never active is not asked to restart" "" "" \
  refresh_unit gitea-runner-swarm.service "$path"
got=$(refresh_unit gitea-runner-swarm.service "$path" 2>&1)
case="a unit never active is not said to be restarted"
count=$((count + 1))
if [ -n "$got" ]; then fail "logged: $got"; fi
UNIT_ACTIVE=1

ACTIVE=0
check "nothing while the container is down" "" "" \
  refresh_consumer swarm-ci-runner gitea-runner-swarm.service "$path"
ACTIVE=1

STARTED="not a time"
check "an unreadable start restarts nothing and says so at err" "" \
  "<3>cannot compare the mtime of $path ('1791232888193159') with the last start of gitea-runner-swarm.service ('not a time') — leaving gitea-runner-swarm.service as it is" \
  refresh_unit gitea-runner-swarm.service "$path"

STARTED=$(systemd_ts 1791232900.000001)
check "a missing file restarts nothing and says so at err" "" \
  "<3>cannot compare the mtime of $dir/missing ('') with the last start of gitea-runner-swarm.service ('$STARTED') — leaving gitea-runner-swarm.service as it is" \
  refresh_unit gitea-runner-swarm.service "$dir/missing"

# compare <description> <want status> <path>. `written_after_start` as
# `swarm-bao-agent-pki` calls it, which acts only on 0.
compare() {
  case=$1
  count=$((count + 1))
  local rc
  (
    set -e
    written_after_start openbao.service "$3" --machine=swarm-bao 2>/dev/null
  )
  rc=$?
  if [ "$rc" != "$2" ]; then fail "returned $rc, wanted $2"; fi
}

STARTED=$(systemd_ts 1791232900.000001)
compare "the compare is 1 for a file older than the start" 1 "$path"
STARTED=$(systemd_ts 1791232888.193158)
compare "the compare is 0 for a file newer than the start" 0 "$path"
STARTED="not a time"
compare "the compare is 2 for an unreadable start" 2 "$path"
STARTED=$(systemd_ts 1791232900.000001)
compare "the compare is 2 for a missing file" 2 "$dir/missing"

echo "$count cases, $failures failed"
[ "$failures" = 0 ]