Hi Nikolay, The idea makes perfect sense for me, and we should definitely take the best practices from other big Apache projects.
But it could be a lot of work to handle mentions (probably from the email account and inbox). I would like to suggest the following algorithm: Patch available ticket, which was never checked by the bot will be processed in the following steps: 1. check existing run all (by PR or by branch name), if found go to the step 3 2. run-all to be triggered by PR 3. results should be analyzed for the presence of possible blockers. If there is no blockers go to step 5. 4. re-run of particular suites containing possible blockers should be applied to try to get success for very rare flaky failures (<1%). Go to 3 (this go to should be done only once). 5. comment should be added to JIRA ticket containing information about results. If a ticket was processed by bot early (probably author added some fixes) but still in PA state, the bot will check comments list and find possible new mentions (made after the previous build complete date). If it finds such comments it goes to step 1 (trying to find only new builds available). What do you think? Sincerely, Dmitriy Pavlov пн, 24 сент. 2018 г. в 21:43, Nikolay Izhikov <nizhi...@apache.org>: > Hello, Igniters. > > I propose to implement following behaviour: > > 1. Execute "Run all" suite for specific PR when the author of PR makes a > comment > "@mtcga.bot Run Tests!" in GitHub comments. > > 2. Send a comment with "Run All" results both: to a Jira ticket and GitHub > comment. > > 3. Label PR based on "Run All" results like it done in Apache Kafka [1] > > I've create ticket for this proposal [2] > > Thoughts? > > [1] https://github.com/apache/kafka/pulls > [2] https://issues.apache.org/jira/browse/IGNITE-9678