SQL并发事务忽略彼此的锁???死锁 [InnoDB, Python]

Maybe my code will explain it better:

execute("SET TRANSACTION ISOLATION LEVEL SERIALIZABLE")
execute("START TRANSACTION")
execute("SELECT * FROM job WHERE status = %s LIMIT 1 FOR UPDATE", jobStatus.imported)
job_data = cursor.fetchone()
if not job_data:
    connection.rollback()
else:
    execute("UPDATE job SET status = %s WHERE jobID = %s", jobStatus.ingesting, job_data['jobID']) # Update the job data

    if job_data['jobUUID'] == None:
        job_data['jobUUID'] = new_unused_uuid().bytes
        execute("UPDATE job SET jobUUID = %s WHERE jobID = %s LIMIT 1", job_data['jobUUID'], job_data['jobID'])
    if job_data['dateAdded'] == None:
        job_data['dateAdded'] = datetime.datetime.now()
        execute("UPDATE job SET dateAdded = %s WHERE jobID = %s LIMIT 1", job_data['dateAdded'], job_data['jobID'])

    execute("INSERT INTO ingestJob (fk_jobUUID, fk_nodeUUID, status) VALUES (%s, %s, %s) ON DUPLICATE KEY UPDATE fk_nodeUUID = %s, status = %s", job_data['jobUUID'], unique_id.bytes, smallJobStatus.running, unique_id.bytes, smallJobStatus.running)

    connection.commit()

流程如下:

1. 使用 FOR UPDATE 选择一项可能的工作
2. 如果没有作业，则回滚(释放锁)，或者...
3. ...更新该行使其无法重新选择，进行一些不相关的更改
4. 提交

他们都做自己的事，忽略彼此的锁和事务

让我害怕的是它是随机的.它大约每隔一次运行就会发生一次.在隔离环境中尝试相同的查询时，有足够的延迟，我得到了我想要的确切结果.
只要 Alice 调用了 SELECT ... FOR UPDATE，Barry 就无法读取该行，并挂起直到 Alice 提交或回滚.我的现象需要在同一程序的两个实例之间精确同时执行.

while True:
    cursor.execute("SELECT * from job FOR UPDATE")
    print('Selected')
    time.sleep(1)
    connection.rollback()
    print('Released')
    time.sleep(1)