podman-container-tools/podman · warning

pause.pid file refers to PID %ld which is not a pause proces

Error message

pause.pid file refers to PID %ld which is not a pause process, the process may have exited and the PID been recycled. Removing %s

What it means

Rootless podman keeps a 'pause' process holding the user and mount namespaces alive between invocations; its PID is cached in $XDG_RUNTIME_DIR/libpod/tmp/pause.pid. check_stale_pause_pid() validates the cached PID by reading /proc/<pid>/environ for _PODMAN_PAUSE=1 (is_pause_process); on mismatch it prints this warning and unlinks the stale file (pkg/rootless/rootless_linux.c:684-691). This is self-healing recovery, not a fatal error - podman then falls back to creating a fresh pause process / re-exec path.

Source

Thrown at pkg/rootless/rootless_linux.c:689

  /* environ entries are separated by '\0'.  Search for "_PODMAN_PAUSE=1".  */
  for (char *p = buf; p < buf + n; )
    {
      if (strcmp (p, "_PODMAN_PAUSE=1") == 0)
        return 1;
      p += strlen (p) + 1;
    }
  return 0;
}

/* If the process referred to by pause.pid is not actually a pause process,
   it means the PID was recycled.  Warn the user and remove the stale file.  */
static void
check_stale_pause_pid (long pid, const char *path)
{
  if (!is_pause_process (pid))
    {
      fprintf (stderr, "pause.pid file refers to PID %ld which is not a pause process, the process may have exited and the PID been recycled. Removing %s\n", pid, path);
      unlink (path);
    }
}

static int
open_namespace (int pid_to_join, const char *ns_file)
{
  char ns_path[PATH_MAX];
  int ret;

  ret = snprintf (ns_path, PATH_MAX, "/proc/%d/ns/%s", pid_to_join, ns_file);
  if (ret == PATH_MAX)
    {
      fprintf (stderr, "internal error: namespace path too long\n");
      return -1;
    }

  return open (ns_path, O_CLOEXEC | O_RDONLY);

View on GitHub (pinned to a2409076ef)

Solutions

1. Treat it as benign - the stale file is removed automatically and the next podman command recreates the pause process
2. If it recurs, find why the pause process dies: 'journalctl --user' and 'dmesg | grep -i oom'
3. Reset rootless state cleanly: 'podman system reset' or remove $XDG_RUNTIME_DIR/libpod/tmp manually after logging out
4. Ensure XDG_RUNTIME_DIR points at a tmpfs (/run/user/$UID) that systemd recreates per login

# Optionally detect the stale state before podman warns about it
p="$XDG_RUNTIME_DIR/libpod/tmp/pause.pid"
if [ -r "$p" ]; then
  pid=$(cat "$p")
  if [ -r "/proc/$pid/environ" ] && ! tr '\0' '\n' < "/proc/$pid/environ" | grep -qx '_PODMAN_PAUSE=1'; then
    rm -f "$p"   # proactively remove the stale file podman would remove anyway
  fi
fi
podman "$@"

When it happens

Trigger: The pause process died (OOM-killed, manually killed via pkill, killed by 'podman system migrate/reset') and the kernel recycled its PID for an unrelated process; /run/user/$UID contents survived while the pause process did not (crash, forced logout); PID collision after reboot when XDG_RUNTIME_DIR is on persistent storage.

Common situations: Right after an OOM kill of 'catatonit'/'podman pause'; after the user or a cleanup script killed the pause process; in environments where /run/user is not cleaned at logout; on fast-PID-cycling busy hosts.

Related errors

• unable to print to string
• error creating temporary file: %m
• cannot write to file descriptor: %m
• cannot write to pipe: %m
• cannot save namespace handles: %m

AI-assisted analysis of podman-container-tools/podman@a2409076ef (2026-08-15). Data as JSON: /api/errors/22aae93d63e4680d. Report an issue: GitHub.