[GitHub] storm issue #2338: [STORM-2744] add in restart timeout for backpressure

[Ethanlm]

Github user Ethanlm commented on the issue:

    https://github.com/apache/storm/pull/2338
  
    I did some simulation test.
    
    The command I ran is:
    ```
    storm jar /tmp/storm-loadgen-2.0.0-SNAPSHOT.jar 
org.apache.storm.loadgen.ThroughputVsLatency --rate 20000 --counters  12 
--splitters 6 --spouts 3 -t 60 -c topology.workers=3
    ```
    
    **The behavior before the code change:**
    
    At about timestamp 241, I created a znode under backpressure/topology-id 
manually. Quickly the rate (tuple/s) dropped and became 0. At about timestamp 
391, I deleted the znode manually. The rate increased and the spouts were 
catching up. This is the expected behavior of original storm.
    
    
![image](https://user-images.githubusercontent.com/14900612/30712889-234b3bbe-9ed3-11e7-97ab-01c5dffc1f2a.png)
    
    
    
    **The behavior after the code change:**
    
    At about timestamp 211, I created a znode (with timestamp set to 
`System.currentTimeMillis()`) under backpressure/topology-id. The rate dropped 
quickly. Then after 30s, the spouts were catching up. Even though the znode I 
created still exists in zookeeper, it's no long valid because its timestamp is 
out of date.
    
    
![image](https://user-images.githubusercontent.com/14900612/30713002-8dcabbe0-9ed3-11e7-89c7-c88c5dc055c5.png)
    
    
    
    
    
    
    
    



---