Mistral's retries ignore join logical state

There is a strange behavior with mistral's retries: if join task fails because of it's child tasks, retries reexecute this join task and ignore it's logical state.

Example:
wf1:
  task-defaults:
    retry:
      count: 1
      delay: 0
  tasks:
    task1:
      action: std.noop
      on-success: join_task
    task2:
      action: std.fail
      on-success: join_task
    join_task:
      action: std.noop
      join: all

Expected result: join_task has ERROR state after retries; wf1 has state info "failed because of task2, join_task"

Actual result: join_task has SUCCESS state; wf1 has state info "failed because of task2"