Bumping this thread because: 1. Travis CI is becoming increasingly unusable and is blocking valid PRs from merging. 2. Removing Travis and not putting any gate on merging PRs is, to me, an unacceptable medium-to-long term solution. 3. We're very close to having our own hardware to run Jenkins CI runs instead of Travis. 4. It won't fix the problem of PRs being blocked because these test cases are still failing >80% of the time in Jenkins as well as Travis. 5. It'll get worse when we enable cross-platform builds; see my other thread.
Fix the tests, or everyone loses. -Joan "I asked nicely, now I'm telling you" Touzet On 2019-05-02 16:41, Joan Touzet wrote: > Hi everyone, > > Lately, our Jenkins CI runs on master (after merges) have been failing a > lot: > > https://s.apache.org/yuwY > > Just in the last run (#537), we have failures in eunit tests for > couch_mrview, mem3 and ddoc_cache that need active investigation. [1] > > Arguably, the reason no one is actively monitoring this and fixing the > tests is because Jenkins does not (yet) gate commits from landing on master. > > This will change in the not-too-distant future. Travis CI has been > slower and slower as of late, and with ownership/leadership change of > Travis (the company) there's some trepidation in the community at large > about its long-term survivability as well. > > IBM has graciously committed to a targeted hardware donation for build > machines for our CI needs, to help us get runs done faster and in a > controlled environment. I'll be working with them once that machine > arrives to set up the new CI environment and ensure it does what we all > expect. If anyone has any input on what that should look like, do reply > to this email and let me know. > > Fixing Jenkins also will fix our broken snapshot package builds, which > very soon *will include ARM64 support* that the community has been > asking for, for a long time. Until we have regular greens on the board > for ARM64, I'm not willing to approve greenlighting public packages or a > Docker container for this platform. (Same goes for other platforms.) > > > In short: *PLEASE HELP FIX THE FAILING TESTS*. If you want a bug per > failing test, I can do that, let me know. > > -Joan "all green all the time?" Touzet > > > [1]: The failure in 537 on CentOS 7 comes from my recent image rebuild, > and CentOS's/EPEL's very recent decision to drop the python3 alias in > favour of version-specific ones (python3.4, python3.6). I'll add a > workaround for this via a /usr/local symlink in the image today. >
