krishvishal opened a new pull request, #4702: URL: https://github.com/apache/iggy/pull/4702
## Which issue does this PR address? Relates to #4515 ## Rationale No test ran real Kafka clients through more than one gateway on an Iggy cluster. The in-process cluster tests send hand-built group requests, and the real-client tests run on one Iggy node. A failure of one gateway or node was never tested for lost or duplicate records. The docs also marked the takeover time after a crash as not tested. ## What changed? Three tests restart each gateway with SIGTERM, stop each node, and stop each gateway with its node. Three more crash the coordinator with SIGKILL: alone, alone with a short session timeout, and with its node. Each test checks that every acknowledged record is stored once, at its acknowledged offset, and read once. It also checks that the group does not rebalance, and that no two gateways answer as coordinator at once. The node test stops the metadata primary first, then the node that holds the primary of the group's coordinator partition. After each restart, it waits until the balancer of #4619 moves a primary back to the node, under load. The crash tests measure the takeover. It takes about 39 s with the default 36 s session timeout. With an 8 s `heartbeat.interval` and a 1 s `consumer_group.heartbeat_interval`, it takes 10 to 11 s. The README and `CONSUMER_GROUPS.md` now give these times in place of "not tested". The tests run in a new `kafka_cluster` nextest group, one at a time, alone on the machine, with up to 10 minutes each. CI pre-pulls their client images, which are the ones the real-client suite uses. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
