Open albehrens opened 3 years ago
The value of RecordMaxBufferedTime is too high.
When you increase the number of shards - you proportionally increase the number of records - that KPL waits for the buffer to fill before publishing data to Kinesis - most likely the buffer stops filling completely, and KPL always waits for RecordMaxBufferedTime
In most cases, RecordMaxBufferedTime value greater than a second severely reduces throughput.
I achieve most performant throughput when my Kinesis stream has 8 shards. Throughput slows down when using 10 shards and does not deliver any records anymore when using 16+ shards. How can this be?
I set following properties because I do not want to lose any records:
Can this have anything to do with it?