Repository navigation
Fixes
- Late backlog after a start or restore in
kafka-safeandupsert-kafka-safe, with idle advancement on. (#452)- Before: a source subtask could be reported idle with a backlog left, before its partitions were assigned or while one had no row yet; a downstream temporal join then followed its other input to wall-clock time, and every backlog row arrived late
- Now: a subtask is reported idle only while none of its partitions has produced a row and each is read to the end of its log; active again on a first row or a newly assigned partition
- Start and restore: every subtask holds back downstream watermarks until the enumerator has assigned the partitions of its first discovery
- Limit: at an idle subtask, records written to a partition before the connector discovers it can still arrive late; discovery runs every
scan.topic-partition-discovery.interval, 5 minutes by default
- Releases missing from Maven repositories when Maven Central rejects them, for builds that depend on the runner's modules. (#453)
- Before: releases went only to Maven Central; it rejected 0.12.5 (the
com.datasqrlnamespace is over Sonatype's monthly cap), so 0.12.5 is in no Maven repository - Now: each release also goes to GitHub Packages, so it resolves even when Maven Central rejects it
- Setup: add the repository
https://maven.pkg.github.com/DataSQRL/flink-sql-runner, with a GitHub token that hasread:packages - GitHub Release: attaches only the uber jar and
stdlib-utils, which goes in/opt/flink/lib; the module jars that 0.12.5 also attached come from Maven
- Before: releases went only to Maven Central; it rejected 0.12.5 (the
Full Changelog: 0.12.5...0.12.6