[GitHub] flink issue #6300: [FLINK-9692][Kinesis Connector] Adaptive reads from Kines...
Github user StephanEwen commented on the issue: https://github.com/apache/flink/pull/6300 Nice feature, thanks a lot. Merged this into the 1.6 and 1.7 branches ---
[GitHub] flink issue #6300: [FLINK-9692][Kinesis Connector] Adaptive reads from Kines...
Github user glaksh100 commented on the issue: https://github.com/apache/flink/pull/6300 @StephanEwen I have run this on the following set up: ``` Number of shards on Kinesis stream: 384 Number of task slots: 384 / 192 / 96 Throughput achieved per shard (with adaptive reads) : 1.95 Mb/sec / 1.75 Mb/sec / 1.6 Mb/sec ``` ---
[GitHub] flink issue #6300: [FLINK-9692][Kinesis Connector] Adaptive reads from Kines...
Github user StephanEwen commented on the issue: https://github.com/apache/flink/pull/6300 I like the idea of this - should make it much easier to use. Have you run this code on some heavier data stream to validate that it works well in practice? If yes, I would be +1 to this ---
[GitHub] flink issue #6300: [FLINK-9692][Kinesis Connector] Adaptive reads from Kines...
Github user glaksh100 commented on the issue: https://github.com/apache/flink/pull/6300 The idea here is that `maxNumberOfRecordsPerFetch` should never be a value that gets records that exceeds the read limit (2 Mb/sec) from the math here. ``` 2 Mb/sec / (averageRecordSizeBytes * # reads/sec)) ``` Atleast that's what the intent is - Let me know if that makes sense or if there is something amiss about the approach here. If there is a way in which `maxNumberOfRecordsPerFetch` is set such that the limit is exceeded, then yes, it will still be throttled by Kinesis. ---
[GitHub] flink issue #6300: [FLINK-9692][Kinesis Connector] Adaptive reads from Kines...
Github user tweise commented on the issue: https://github.com/apache/flink/pull/6300 @tzulitai can you please take a look - would be good if we can get this into v1.6.0 ---
[GitHub] flink issue #6300: [FLINK-9692][Kinesis Connector] Adaptive reads from Kines...
Github user glaksh100 commented on the issue: https://github.com/apache/flink/pull/6300 @fhueske @tzulitai @tweise Can you please take a look when you have a chance? ---