Github user aljoscha commented on the pull request:
https://github.com/apache/flink/pull/906#issuecomment-121168918
The job I ran was source -> repartition -> sink, where the source is
generating tuples in an infinite loop.
The gist of the data is this (I'm rerunning some of them with longer
runtimes to remove outliers):
- Introducing timestamp for each record that is set at source incurs about
a 15 % performance hit in these tests
- Watermarks have no additional overhead once the interval is higher than
100 sec (I didn't test with lower intervals, such as 50 sec, but I will do
these today.)
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
