Hi

We are using storm 1.0 and trident

One of our streams uses windowing with an aggregator that counts the number
of events in the window

We find that the counts are incorrect (under certain conditions)
- When we produce test data in small batches to Kafka 100 vs 5000 batch size

Here are the things we tried,
1. increased timeouts
2. removed persistence and just logged our results, to account for replays
due to hbase errors

Both don't yield any benefit , we can consistently replicate incorrect
counts. Kindly advice what we can do

Uday

Reply via email to