+1 (binding) (rabreu) Rui Abreu Checks done on the Apache Storm 3.2.0 release candidate (ATR revision 00002, tag v3.2.0, commit 98ce70603d9f7fa28725c5cf41af92ebea9a38b1).
Downloaded from the public ATR URLs of the candidate (https://releases.apache.org/download/urls/storm/3.2.0), without logging in. 1. Signatures and checksums - Verified the OpenPGP signature (.asc) and the SHA-512 checksum of all 7 files: apache-storm-3.2.0.tar.gz / .zip, apache-storm-3.2.0-lite.tar.gz / .zip, apache-storm-3.2.0-src.tar.gz / .zip and RELEASE_NOTES.html. All good. - Signing key: 72B436558AA9CDCA2C4CBAC340D4B35E2C1452E5 (Rui Abreu, [email protected]), which is in https://downloads.apache.org/storm/KEYS. 2. Full and lite binary distributions on a local cluster Docker Compose: ZooKeeper 3.9.3, Kafka 3.7.1, Nimbus, UI and 5 single-slot supervisors, each distribution installed from the ATR tarball (signature checked again at install). The UI reports version 3.2.0 with 5/5 supervisors and 5 slots, for both editions. Example topologies (storm-starter and storm-kafka-client-examples 3.2.0), one at a time, on both editions. All were ACTIVE and emitted tuples with 0 failed: - WordCount: 3 workers on 3 distinct hosts (about 40,000 tuples emitted) - Lambda: 2 workers on 2 distinct hosts (about 1.1 million tuples emitted) - MultiThreadWordCount: 1 worker (about 8,200 tuples emitted) - Kafka spout example (3 producers + 1 consumer, 4 workers): the consumer topology reported 0 failed (full: 120 emitted, 160 acked; lite: 160 emitted, 180 acked) The storm-starter spouts emit unanchored tuples, so their acked count stays 0 by design. 3. Lite distribution, optional jars Ran bin/storm-kafka-monitor-fetch against the Nexus staging repository https://repository.apache.org/content/repositories/orgapachestorm-1131. The Kafka spout lag then showed up in the Storm UI. Note: on the first run of the full edition, MultiThreadWordCount showed no stats in the UI within my test script's 60 second wait. The topology was ACTIVE and emitting (8,280 tuples, 0 failed, no errors). It passed on a re-run with a 180 second wait, and I raised the script's default to match. I did not see a failure in Storm itself. Not covered by these checks: building from the source release and running its unit tests, a secure (Kerberos/TLS) cluster, the UI actions (jstack, heap dump, rebalance, log search), and platforms other than Linux with Docker. -- Rui Abreu (rabreu) This email was sent by [email protected] on the Apache Trusted Releases platform
