unraid-podman/plugin/sbin/podman-autostart.sh

#!/bin/bash
# =============================================================================
# plugin/sbin/podman-autostart.sh
#
# "Autostart einrichten" — starts containers listed in
# /boot/config/plugins/podman/autostart, in file order, via the running
# Podman API service. Run by `rc.podman start` after the service is up.
# See docs/ARCHITECTURE.md section 12 (Autostart).
#
# Design constraints from the architecture doc, both implemented below:
#   - A single container failing to start must NOT abort the rest of the
#     chain (one broken container shouldn't take down everything else).
#   - After repeated consecutive failures, a container is automatically
#     paused out of the autostart chain ("Safe-Mode pro Container") so a
#     persistently crash-looping container doesn't waste boot time or
#     resources forever — with a GUI notification explaining why.
#
# File formats:
#   autostart          one container name per line; '#' starts a comment;
#                       blank lines ignored. Order = start order.
#   autostart-delay     optional "<container-name>=<seconds>" lines — sleep
#                       that many seconds AFTER starting that container
#                       before moving to the next (for startup dependencies,
#                       e.g. a database before the app that needs it).
#
# Failure tracking: a small per-container counter file under
# $PODMAN_BOOT_DIR/autostart-failures/<name> holds the consecutive-failure
# count. Reset to 0 on any successful start.
# =============================================================================

set -u

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
# shellcheck source=./podman-common.sh
. "$SCRIPT_DIR/podman-common.sh"

podman_load_cfg

# After this many consecutive failed autostart attempts, a container is
# skipped and flagged rather than retried again on the next boot, until a
# human clears it (see clear_failure_flag below).
MAX_CONSECUTIVE_FAILURES=3

FAILURES_DIR="$PODMAN_BOOT_DIR/autostart-failures"
mkdir -p "$FAILURES_DIR"

# -----------------------------------------------------------------------------
# get_delay <container-name>
#
# Looks up an optional post-start delay for a container from
# autostart-delay ("name=seconds" lines). Returns 0 if not configured.
# -----------------------------------------------------------------------------
get_delay() {
  local name="$1"
  if [ -f "$PODMAN_AUTOSTART_DELAY_FILE" ]; then
    awk -F= -v n="$name" '$1 == n { print $2; found=1 } END { if (!found) print 0 }' \
      "$PODMAN_AUTOSTART_DELAY_FILE"
  else
    echo 0
  fi
}

failure_count() {
  local name="$1"
  local f="$FAILURES_DIR/$name"
  [ -f "$f" ] && cat "$f" || echo 0
}

record_failure() {
  local name="$1"
  local count
  count=$(($(failure_count "$name") + 1))
  echo "$count" > "$FAILURES_DIR/$name"
  echo "$count"
}

clear_failure_flag() {
  local name="$1"
  rm -f "$FAILURES_DIR/$name"
}

# -----------------------------------------------------------------------------
# start_one <container-name>
#
# Starts a single container via `podman start`, which talks to the already
# running API service over $PODMAN_SOCKET rather than spawning an unrelated
# podman process tree — see docs/ARCHITECTURE.md section 11 (Container).
# -----------------------------------------------------------------------------
start_one() {
  local name="$1"

  local prior_failures
  prior_failures=$(failure_count "$name")
  if [ "$prior_failures" -ge "$MAX_CONSECUTIVE_FAILURES" ]; then
    podman_log_error "autostart: skipping '$name' — $prior_failures consecutive prior failures (Safe-Mode). Remove $FAILURES_DIR/$name to re-enable."
    return 1
  fi

  podman_log "autostart: starting '$name'"
  if podman --url "unix://$PODMAN_SOCKET" start "$name" > /tmp/podman-autostart-"$name".log 2>&1; then
    clear_failure_flag "$name"
    podman_log "autostart: '$name' started successfully"
    return 0
  fi

  local new_count
  new_count=$(record_failure "$name")
  podman_log_error "autostart: '$name' failed to start (attempt $new_count/$MAX_CONSECUTIVE_FAILURES) — see /tmp/podman-autostart-$name.log"

  if [ "$new_count" -ge "$MAX_CONSECUTIVE_FAILURES" ]; then
    podman_notify "Podman container disabled from autostart" \
      "'$name' failed to start $new_count times in a row and has been paused from autostart. Fix the underlying issue, then remove $FAILURES_DIR/$name to re-enable." \
      "warning"
  fi

  return 1
}

# -----------------------------------------------------------------------------
# main
# -----------------------------------------------------------------------------
if [ ! -f "$PODMAN_AUTOSTART_FILE" ]; then
  podman_log "autostart: no autostart file at $PODMAN_AUTOSTART_FILE, nothing to do"
  exit 0
fi

if [ ! -S "$PODMAN_SOCKET" ]; then
  podman_log_error "autostart: podman API socket ($PODMAN_SOCKET) not present — is the service running?"
  exit 1
fi

started=0
skipped=0
failed=0

while IFS= read -r raw_line || [ -n "$raw_line" ]; do
  # Strip comments and surrounding whitespace; skip blank lines.
  line="${raw_line%%#*}"
  line="$(echo "$line" | xargs || true)"
  [ -z "$line" ] && continue

  if start_one "$line"; then
    started=$((started + 1))
  else
    prior=$(failure_count "$line")
    if [ "$prior" -ge "$MAX_CONSECUTIVE_FAILURES" ]; then
      skipped=$((skipped + 1))
    else
      failed=$((failed + 1))
    fi
    # A single container's failure does not abort the loop — see file
    # header. Continue to the next entry.
    continue
  fi

  delay=$(get_delay "$line")
  if [ "$delay" -gt 0 ] 2> /dev/null; then
    podman_log "autostart: waiting ${delay}s after '$line' (configured dependency delay)"
    sleep "$delay"
  fi
done < "$PODMAN_AUTOSTART_FILE"

podman_log "autostart: complete (started=$started failed=$failed skipped-safe-mode=$skipped)"

# Exit non-zero only if EVERY entry failed outright (as opposed to a mix,
# or entries already in Safe-Mode) — rc.podman treats that as worth
# flagging loudly, whereas a partial failure is already individually
# notified above.
if [ "$started" -eq 0 ] && [ "$failed" -gt 0 ]; then
  exit 1
fi
exit 0