[
https://issues.apache.org/jira/browse/FLINK-22643?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17467174#comment-17467174
]
fanrui commented on FLINK-22643:
--------------------------------
[~kevin.cyj] [~pnowojski] Of course, I can do it. So, do I need to run some
benchmarks with different settings (1, 5, Integer.MAX_VALUE) to select the
default value?
My test job is
[https://github.com/1996fanrui/fanrui-learning/commit/fe038478b1250c97724c60bd7bc48a465fb9550a]
For scenarios with fewer tasks, this config may have no effect, the number of
tasks of the test job <3 is meaningless. So my test task looks like this:
Number of Tasks == 10
parallelism = 10
When configuring 1,5, Integer.MAX_VALUE, compare TM's CPU resource usage and
numRecordsInPerSecond metric.
The number of Tasks and parallelism can also set other values, such as
(20,10),(30,10), etc.
Can master provide some suggestions?
> Too many TCP connections among TaskManagers for large scale jobs
> ----------------------------------------------------------------
>
> Key: FLINK-22643
> URL: https://issues.apache.org/jira/browse/FLINK-22643
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Network
> Affects Versions: 1.14.0, 1.13.2
> Reporter: Zhilong Hong
> Priority: Minor
> Labels: auto-deprioritized-major
>
> For the large scale jobs, there will be too many TCP connections among
> TaskManagers. Let's take an example.
> For a streaming job with 20 JobVertices, each JobVertex has 500 parallelism.
> We divide the vertices into 5 slot sharing groups. Each TaskManager has 5
> slots. Thus there will be 400 taskmanagers in this job. Let's assume that job
> runs on a cluster with 20 machines.
> If all the job edges are all-to-all edges, there will be 19 * 20 * 399 * 2 =
> 303,240 TCP connections for each machine. If we run several jobs on this
> cluster, the TCP connections may exceed the maximum limit of linux, which is
> 1,048,576. This will stop the TaskManagers from creating new TCP connections
> and cause task failovers.
> As we run our production jobs on a K8S cluster, the job always failover due
> to exceptions related to network, such as {{Sending the partition request to
> 'null' failed}}, and etc.
> We think that we can decrease the number of connections by letting tasks
> reuse the same connection. We implemented a POC that makes all tasks on the
> same TaskManager reuse one TCP connection. For the example job we mentioned
> above, the number of connections will decrease from 303,240 to 15960. With
> the POC, the frequency of meeting exceptions related to network in our
> production jobs drops significantly.
> The POC is illustrated in:
> https://github.com/wsry/flink/commit/bf1c09e80450f40d018a1d1d4fe3dfd2de777fdc
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
