Github user danny0405 commented on the issue:

    https://github.com/apache/storm/pull/2800
  
    @srdo 
     I think the only difference is that compared to `AtomicReference`, 
`ConcurrentHashMap` can keep thread safe but can not ensure that the value read 
is up to date,  which will cause some inconsistent behavior,cause the 
scheduling thread will read it every 10s, with an out of date heartbeat cache, 
master will kill a fine worker or restart a already started one.


---

Reply via email to