Overview

• Background: I start a thread [call thread A below]in Spark driver to handle opening files in codis, in which I start six thread [call sub threads below]to handle the files in parallel.
• What happened:
  • I found the codis openingFilePool never update since a given time.
  • So I looked into the Thread Dump tools of spark, and saw that thread A is always been WAITING. As below:
    
  • The normal one is as follow:

    

  • See the difference?
    The thread should be TIMED_WAITING(sleep), not be WAITING(wait for some other threads to awake it)!
  • It seems there is a deadlock?
• Thread A get a jedis instance from JedisResourcePool to get all opening files.
• Then, the sub threads handle these opening files and interact with codis in pipeline. [I use try(Pipeline pipeline = CodisPool.getPool().getResource().pipelined())]

Possible causes

• I am not sure this is because 1) there is no enough instance in JedisPool; or 2) there is a deadlock?

  • try (Pipeline pipeline = CodisPool.getPool().getResource().pipelined())

    • I am not sure whether this code will close jedis and pipeline, or it will just close pipeline?
  • JedisResourcePool is not thread-safe?
    • Some says that we should sync when we getResource() from JedisResourcePool. I am not sure.
    • And also I think should I configure the number of Jedis pool.