Open yichao-figma opened 2 weeks ago
Looks like the SQS send is done on a per-record basis: https://github.com/Nordstrom/kafka-connect-sqs/blob/master/src/main/java/com/nordstrom/kafka/connect/sqs/SqsSinkConnectorTask.java#L111
Given SQS.send usually takes 10ms+, this wouldn't scale for partition that has > 100 RPS.
SQS.send
There could be two options for optimization:
put()
There is an outstanding MR for add an ASYNC mode; does this meet your expectations for optimization?
https://github.com/Nordstrom/kafka-connect-sqs/pull/53
Looks like the SQS send is done on a per-record basis: https://github.com/Nordstrom/kafka-connect-sqs/blob/master/src/main/java/com/nordstrom/kafka/connect/sqs/SqsSinkConnectorTask.java#L111
Given
SQS.send
usually takes 10ms+, this wouldn't scale for partition that has > 100 RPS.There could be two options for optimization:
put()