Github user srdo commented on a diff in the pull request:
https://github.com/apache/storm/pull/2911#discussion_r234901127
--- Diff:
external/storm-kafka-client/src/main/java/org/apache/storm/kafka/spout/KafkaSpoutConfig.java
---
@@ -137,7 +142,11 @@ public KafkaSpoutConfig(Builder<K, V> builder) {
/**
* The kafka spout polls records from the last committed offset,
if any. If no offset has been committed it behaves as LATEST
*/
- UNCOMMITTED_LATEST;
+ UNCOMMITTED_LATEST,
+ /**
+ * Start at the earliest offset whose timestamp is greater than or
equal to the given startTimestamp
+ */
+ TIMESTAMP;
--- End diff --
Not sure I follow. I mean that I think we should have both TIMESTAMP, which
unconditionally seeks to a timestamp, and UNCOMMITTED_TIMESTAMP which only
seeks to a timestamp if no offset has been committed yet. That way you can use
UNCOMMITTED_TIMESTAMP to start at a timestamp the first time you deploy the
topology, but the topology doesn't reset when you redeploy.
---
