So the only method to improve kafkaspout performance is to increase the partition count of specific topic to increase the parallelism of kafkaspout?

 

发送自 Windows 10 邮件应用

 

发件人: Jungtaek Lim
发送时间: 20161019 9:48
收件人: Chen Junfeng; Chien Le; user@storm.apache.org
主题: Re: Bolt slow down the emit speed of spout

 

You need to make it clear that you're using multiple 'topics' or multiple 'partitions'.
There's no way to read partition in parallel, and it's a design of Kafka. If you mean topic, you can increase parallelism of Spout to same as partition count.

2016년 10월 19일 (수) 오전 10:42, Chen Junfeng <k-2feng@hotmail.com>님이 작성:

Thanks for your reply.

 

Maybe I am close to the reason. I found the cpu usage of spout worker is really very high, about 200% to 400% via ‘top’ command. But KafkaSpout with Storm seems not support using multi spouts instance to read one topic.

 

发送自 Windows 10 邮件应用

 

发件人: Jungtaek Lim
发送时间: 20161019 9:18
收件人: Chien Le; user@storm.apache.org
主题: Re: Bolt slow down the emit speed of spout

 

Could you change your bolt to no-op but ack and run again? It makes your benchmark more clearer.

And if you pick your number from the first 10 minutes, could you pick your number from another 10 minutes? It gets rid of worker / executor deploy / startup time, and warm-up of JVM.

- Jungtaek Lim (HeartSaVioR)
On Wed, 19 Oct 2016 at 10:00 AM Chen Junfeng <k-2feng@hotmail.com> wrote:

I used ‘dstat -a’ which measures cpu, memory, disk and network.  Disk should not be the bottleneck as my topology doesn’t have much disk operation. Log.info print tuple to disk every 1000 tuples.  

 

发送自 Windows 10 邮件应用

 

发件人: Chien Le
发送时间: 20161019 5:39
收件人: user@storm.apache.org
主题: RE: Bolt slow down the emit speed of spout

 

dstat -d is measuring your disk throughput, yes? Did you mean dstat -n?

 

Otherwise, dstat -d showing 300MB/s doesn't eliminate disk as your bottleneck. I would try using iostat -x 1 and see what your %util and svc times are to truly eliminate disk as the bottleneck.

 

-Chien

 

From: Chen Junfeng [mailto:k-2feng@hotmail.com]
Sent: Monday, October 17, 2016 11:30 PM
To: user@storm.apache.org
Subject: Bolt slow down the emit speed of spout

 

 

I found a tough but interesting issue about Storm 1.0.2 to share with you.

Firstly I introduce my system structure. My topology has 471 workers, 6 kafka topic and some processor bolt. Backpressure has been disabled in topology by

conf.put(Config.TOPOLOGY_BACKPRESSURE_ENABLE, false);

 

If I submit KafkaSpout only, the statistic data from Storm UI is as below:

Topology stats

Window

Emitted

Transferred

Complete latency (ms)

Acked

Failed

10m 0s

875161429

0

0

875141691

3h 0m 0s

927117120

0

0

927117320

1d 0h 0m 0s

927117120

0

0

927117320

All time

927117120

0

0

927117320

Spouts (10m 0s)

Search:

Id

Executors

Tasks

Emitted

Transferred

Complete latency (ms)

Acked

Failed

Spout_JSCTGW_144

2

2

128701261

0

0

128702858

0

Spout_JSCTGW_145

2

2

162347454

0

0

162347639

0

Spout_JSCTGW_146

2

2

135598494

0

0

135598608

0

Spout_JSCTGW_147

2

2

128307822

0

0

128306102

0

Spout_JSCTGW_148

2

2

160369513

0

0

160360423

0

Spout_JSCTGW_149

2

2

159836885

0

0

159826061

0

 

 

In 10 minutes time, 6 kafkaSpout read more than 800 million tuple from kafka cluster.

Then I submit the topology with a simple PrintBolt, the 10min stat data is as following:

Spouts (10m 0s)

Search:

Id

Executors

Tasks

Emitted

Transferred

Complete latency (ms)

Acked

Failed

Error Host

Error Port

Spout_JSCTGW_144

2

2

113599789

113599789

0

113513630

0

Spout_JSCTGW_145

2

2

122501522

122501575

0

122659659

0

Spout_JSCTGW_146

2

2

91308915

91308915

0

91308652

0

Spout_JSCTGW_147

2

2

105029568

105029568

0

104968275

0

Spout_JSCTGW_148

2

2

115889172

115889178

0

115890165

0

Spout_JSCTGW_149

2

2

115185591

115185591

0

115185638

0

Showing 1 to 6 of 6 entries

663526019

Bolts (10m 0s)

Search:

Id

Executors

Tasks

Emitted

Transferred

Capacity (last 10m)

Execute latency (ms)

Executed

Process latency (ms)

Acked

PrintBolt

240

240

0

0

0.241

0.041

665063824

0

665,036,902

 

The printbolt contains nothing but a simple log.info method, but the total number of emitted tuple of spout decrease to about 600 million. All of machines have 32 cores but the system average load never exceed 15. Also I have checked the network load. My network adapter has 10Gb bandwidth and the dstat -d command shows the max rec or send speed is 300MB/s at most , which doesnt reach the limit of network adapter. The total number of executors is less than that of total workers, so performance should not be problem.

 

Then I do some more test,

Only one kafkaspout reading on topic

 

Spouts (10m 0s)

Search:

Id

Executors

Tasks

Emitted

Transferred

Complete latency (ms)

Acked

Failed

Spout_JSCTGW_145

2

2

148918096

0

0

148917805

0

 

One kafkaspout and one data processing bolt

 

Spouts (10m 0s)

Search:

Id

Executors

Tasks

Emitted

Transferred

Complete latency (ms)

Acked

Failed

Spout_JSCTGW_145

2

2

106481751

106481751

0

106367801

0

Showing 1 to 1 of 1 entries

Bolts (10m 0s)

Search:

Id

Executors

Tasks

Emitted

Transferred

Capacity (last 10m)