| FazBrowse GitHub Viewer | Trending | | Home |
| Tools: [Download Repo ZIP] [Original HTTPS Page] |
|
FWIW I was expecting these vars and got all of them but TRAVIS_TEST_RESULT: TRAVIS: true TRAVIS_BUILD_DIR: /home/travis/build/GoogleCloudPlatform/google-cloud-python TRAVIS_LANGUAGE: python TRAVIS_OS_NAME: linux TRAVIS_REPO_SLUG: GoogleCloudPlatform/google-cloud-python TRAVIS_TAG: TRAVIS_PYTHON_VERSION: 2.7 # These ones seem to change with each PR / push TRAVIS_SECURE_ENV_VARS: false TRAVIS_BRANCH: master TRAVIS_EVENT_TYPE: pull_request TRAVIS_PULL_REQUEST: 2309 # These ones seem to change with each build TRAVIS_BUILD_ID: 159449750 TRAVIS_BUILD_NUMBER: 4806 TRAVIS_COMMIT: 07128f1ff0a3930c837728e39d9453984a75efbb TRAVIS_COMMIT_RANGE: 1537d4a74be261c3f30eeed64b37d5dcdd3fba17...d038c32b438c753128945001e2b4e05b8742d312 TRAVIS_JOB_ID: 159449751 TRAVIS_JOB_NUMBER: 4806.1 From the build lifecycle doc, it seems that TRAVIS_TEST_RESULT is only populated in the after_success / after_failure / after_script environments. |
Sorry, something went wrong.
|
TRAVIS_COMMIT (07128f1) seems to be a temporary merge commit of HEAD in this branch merged into HEAD of master, and those are the first and last (respectively) in TRAVIS_COMMIT_RANGE |
Sorry, something went wrong.
|
I rebased on top master (#2305 was merged) and adding another commit on top to see how it changes TRAVIS_COMMIT_RANGE |
Sorry, something went wrong.
|
In the latest build, the only env. var. changes were: TRAVIS_BUILD_ID: 159472570 TRAVIS_BUILD_NUMBER: 4810 TRAVIS_COMMIT: c9e955423f57e99a6be2fda7e4b73ff1a9f5beac TRAVIS_COMMIT_RANGE: 397014a17d80225526b86d5894780e1544a67175...dfbf929a6ca8426c6afeeccd2e587a34d65cb3e7 TRAVIS_JOB_ID: 159472571 TRAVIS_JOB_NUMBER: 4810.1 After the rebase the HEAD commit in master moved to 397014a and the 2 commits in this PR became: As before c9e955423f57e99a6be2fda7e4b73ff1a9f5beac is a hidden commit that merges my branch into HEAD (i.e. dfbf929 into 397014a) |
Sorry, something went wrong.
|
From How-Pull-Requests-are-Tested
|
Sorry, something went wrong.
|
I'm going to push this dont-merge-2277-sleuthing into https://github.com/GoogleCloudPlatform/google-cloud-python/ to see what the differences are there: |
Sorry, something went wrong.
|
ENV VARS: In the pull request build the usual players changed. In the "push" build, there are some differences: # These ones seem to change with each PR / push TRAVIS_SECURE_ENV_VARS: true # vs. false TRAVIS_BRANCH: dont-merge-2277-sleuthing # vs. master TRAVIS_EVENT_TYPE: push # vs. pull_request TRAVIS_PULL_REQUEST: false # vs. 2309 Also TRAVIS_COMMIT is the actual commit at the tip of the branch (i.e. the one that was pushed, c274f02) and TRAVIS_COMMIT_RANGE covers a range from an invalid commit to a valid one. The beginning commit (7147a65) isn't in the branch history: it's from the previous "push" build, but I force-pushed over it. The 2nd (c274f02) is just the final commit in that branch. Also, the commit hashes in TRAVIS_COMMIT_RANGE are 12 chars instead of the full 40 as we've seen in pull request builds (this is also different than the --pretty=%h length, which is 7). As for the state of the git checkout in each, the PR is checked out at FETCH_HEAD which is just the result of running $ git fetch origin +refs/pull/2309/merge: $ git checkout -qf FETCH_HEAD while the "push" build is checked out at a detached HEAD for the lead commit in the branch (c274f02). This is done explicitly $ git checkout -qf c274f02461c82627cc10a71185df46e6091f97a8 In either case, there is only one git remote (I was curious in a pull request if the originating fork could be found anywhere). Finally, a limited checkout is in place, making sure master is the only fetched branch in the PR build and dont-merge-2277-sleuthing is the only branch in the "push" build: $ git clone --depth=50 ... $ git clone --depth=50 --branch=dont-merge-2277-sleuthing ... |
Sorry, something went wrong.
|
It looks like some new vars like TRAVIS_PULL_REQUEST_BRANCH and TRAVIS_PULL_REQUEST_SHA have been added so I triggered a new build. (Oops) From that build we see the new variables as expected # NEW! TRAVIS_PULL_REQUEST_BRANCH: dont-merge-2277-sleuthing TRAVIS_PULL_REQUEST_SHA: d4bc9d575e95146071b9892edf0b467a91f49f4c d4bc9d5 is the HEAD commit in the branch, which agrees with the end commit in TRAVIS_COMMIT_RANGE and distinguishes from the target GitHub merge commit in TRAVIS_COMMIT (bc0a24f). |
Sorry, something went wrong.
|
I just pushed a tag to trigger a tag build. Have already deleted tag, but the build remains. The non-trivial changes we see are as follows NEW (TAG) | OLD (PR) --------------------------------|------------------------------------------------------------------ TRAVIS_TAG: dont-tag-sleuth | TRAVIS_TAG: TRAVIS_BRANCH: dont-tag-sleuth | TRAVIS_BRANCH: master TRAVIS_EVENT_TYPE: push | TRAVIS_EVENT_TYPE: pull_request TRAVIS_PULL_REQUEST: false | TRAVIS_PULL_REQUEST: 2309 TRAVIS_PULL_REQUEST_BRANCH: | TRAVIS_PULL_REQUEST_BRANCH: dont-merge-2277-sleuthing TRAVIS_PULL_REQUEST_SHA: | TRAVIS_PULL_REQUEST_SHA: d4bc9d575e95146071b9892edf0b467a91f49f4c
|
Sorry, something went wrong.
This is sniffing around towards googleapis#2277, to see how much info we can use to restrict the tests we run.
* update pyproject.toml to follow PEP 639 * Update pyproject.toml PEP 639 Thanks for the feedback, I've removed the version number completely as requested. * Update pyproject.toml --------- Co-authored-by: Chalmer Lowe <chalmerlowe@google.com>
This PR addresses a flaky ZMQError: Address already in use that occasionally occurred during parallel notebook test execution. **Problem:** The bigframes.display module eagerly imported anywidget and traitlets at module load time (`bigframes/display/__init__.py`). This meant that when multiple Jupyter kernels were spun up simultaneously by nox for parallel testing, they would all try to initialize traitlets.HasTraits objects with sync=True properties. This led to race conditions and ZMQ port conflicts, causing notebook tests (including those that did not directly use anywidget like `streaming_dataframe.ipynb`) to fail. Log is [here](https://fusion2.corp.google.com/invocations/72088900-0196-4441-944b-ad68e491a8f8/targets/bigframes%2Fpresubmit%2Fnotebook/log). **Solution:** The TableWidget class import in `bigframes/display/__init__.py` has been refactored to use Python's `__getattr__` for lazy loading. This ensures that anywidget and traitlets are only imported and their associated kernel communication channels are initialized when display.TableWidget is actually accessed by the code. This prevents premature initialization and eliminates the port collision race condition during parallel test startup. Fixes #<465768150> 🦕
| Back | FazBrowse Home | New Git URL |
This is sniffing around towards #2277, to see how much info we can use to restrict the tests we run.
Context: I tried to build a similar feature in late 2014 / early 2015 and the env. vars. weren't very useful. Hoping that has changed in the last two years.