...
Current state: Under Discussion
Discussion thread: here here
JIRA: here
Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).
...
When processing billions of records a day, it is likely that some of these records may be corrupt (e.g., because of a random data flip) or that the serialization or processing logic may be incorrect and buggy or it cannot handle all record types. We call such records “poison pills” in this KIP. In this cases, users might want some options with how to handle poison pills. Currently either the entire streams pipeline is closed upon such a record, or the user has to do lots of extra work with operations like flatMap, branch etc to circumvent the exceptions.
...