Hi Alex,

Thanks for the guidance!

Is there any formula to determine what should be  the ideal Max Spout Pending 
based on the number of tuples per second and the latency time? I read some 
blogs where the same problem occurred because the max spout pending was too 
high.

The worker log doesn’t show any messages on why the KafkaSpout has stopped 
picking the messages from the queue . Is there any other place when we can get 
more information on this.

Thanks & Regards,
Nitin Gupta

From: Alexander T [mailto:[email protected]]
Sent: Thursday, April 07, 2016 1:13 PM
To: [email protected]
Subject: RE: Storm Topology Hanging


Hi Nitin,

You might get problems with max spout pending if you delay acking of one 
message until another arrives. In this case having a low max spout pending can 
cause the buffers to fill before the awaited message arrives. But this type of 
deadlock should cause a message timeout. If you have disabled message timeout, 
however, you could get a deadlock without symptoms.

Another possible problem is auth failure. I don't know how the Kafka spout 
works, but the Kinesis spout which we use is completely silent if it fails to 
connect. Very annoying.

Best regards,
Alexander
On Apr 7, 2016 7:58 AM, "Nitin Gupta" 
<[email protected]<mailto:[email protected]>> wrote:
I checked the worker log files and found there is no specific exception. It 
looks the KafkaSpout just stops picking the records from kafka queue. Any 
pointers?

From: Nitin Gupta [mailto:[email protected]<mailto:[email protected]>]
Sent: Wednesday, April 06, 2016 1:30 PM
To: [email protected]<mailto:[email protected]>
Subject: RE: Storm Topology Hanging

Thanks for the response!


•         The Storm version is 0.9.4

•         I have set a message timeout to 300 seconds.

•         Yes I also think the latency is too high. The Bolt is doing too much 
of work there. It is basically requesting documents for a User from multiple 
sources(Google Drive, Box,DropBox etc). I can split the work into 3 topologies. 
But wanted to know is latency the root cause for the hang.

I will check if the worker stacktrace can be shared.

Thanks !

Nitin Gupta

From: Jungtaek Lim [mailto:[email protected]]
Sent: Wednesday, April 06, 2016 1:15 PM
To: [email protected]<mailto:[email protected]>
Subject: Re: Storm Topology Hanging

Hi Nitin,

Though it decreases throughput when max spout pending value is small, it should 
not hang.
Which version of Storm do you use, and could you share stack trace of workers 
if you don't mind?

Btw, your spout latency is a bit high (10s). Which value you set message 
timeout secs?

Thanks,
Jungtaek Lim (HeartSaVioR)


2016년 4월 6일 (수) 오후 4:34, Nitin Gupta 
<[email protected]<mailto:[email protected]>>님이 작성:
Dear All,

I am facing an issue where in one of my Topology hangs after 14-16 hours.

From the Storm UI I can see the below statics:

•         Number of Records Emitted: 4000

•         Number of Records Acked : 3180

•         Number of Records Failed : 1140


•         The Spout Latency is 10913 milliseconds

•         I have configured the Max Spout Config to the Number of  Executors : 
40 in my case

•         I have only one Spout

From various blogs on this topic , I understand if the value of Max Spout 
Config is not set properly it can result in such issues.
Can someone please guide me what should the correct value in my case.

Thanks & Regards,
Nitin Gupta


Reply via email to