Git Product home page Git Product logo

ci-hud's Introduction

Visit https://hud.pytorch.org and see https://github.com/pytorch/pytorch/wiki/Using-hud.pytorch.org for usage details.

Development

This project was bootstrapped with Create React App. To build locally, install dependencies and run in develop mode:

git clone https://github.com/pytorch/pytorch-ci-hud.git ci-hud
cd ci-hud
yarn install
npm start  # start a development server on localhost:3000

The code is routed from App.js to places like:

Submitting a PR

Submitting a PR will trigger a preview build on Netlify so others can preview the changes. Once merged, GitHub Actions will deploy the new site to the gh-pages branch.

Log-ins for Local Development and Deploy Previews

Log-ins don't work on the Netlify deploy previews since the GitHub app has hud.pytorch.org hardcoded as its callback URL. To see changes that require the GitHub API in a preview you can manually copy your OAuth token. In the JS console on hud.pytorch.org, run localStorage.getItem("gh_pat"). Then in the preview's console, run localStorage.setItem("gh_pat", "<the token>").

Data

The data backing the HUD is updated via a series of cron-based GitHub Actions. See .github/ for details.

License

This repository uses the MIT License, as found in the LICENSE file.

ci-hud's People

Contributors

1ntegr8 avatar bigfootjon avatar dependabot[bot] avatar driazati avatar ezyang avatar facebook-github-bot avatar houseroad avatar malfet avatar ochalouhi avatar pjh5 avatar prigoyal avatar samestep avatar seemethere avatar suo avatar walterddr avatar xuzhao9 avatar yns88 avatar zengk95 avatar

Stargazers

 avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar

Watchers

 avatar  avatar  avatar  avatar  avatar

ci-hud's Issues

GitHub Actions jobs are not prefixed by workflow name

Currently, PyTorch has two different GitHub Actions workflows containing a job named test:

Since the HUD uses only the job name, these two workflows both get put into the same column, which in practice seems to mean that only the signal for the latter is shown, while the former gets ignored.

It would be better if GitHub Actions job names were prefixed by the workflow name as they are on GitHub itself, so those columns would instead be called "Test tools / test" and "Linux CI (pytorch-linux-xenial-py3.6-gcc5.4) / test" respectively.

cc @seemethere @malfet

[pr page] Integrate CircleCI

We still rely a bunch on CircleCI and will for at least the coming weeks. It'd be nice if it was added to the list of runs alongside GHA. CircleCI is more liberal with un-auth'ed API requests than GitHub so we may be able to just query it directly without having the user authenticate. If that doesn't work, we could have a AWS lambda that generates the data for each commit, but this would be tricky and it's not clear if this would allow downloading logs.

Add status badges

Feature Request

For each repo we should generate a status badge image that can be used in the HUD itself for a quick view on how the latest commit is faring. We could also embed this in the README

The badge itself would probably have to be generated in GHA and stored in S3 as an image, then the HUD / README can just download it as necessary.

Visually distinguish groups from individual jobs

Right now groups are just a bolded version of whatever their status is which is not always easy to identify in context. We also now have overloaded behavior for clicking, where in some cases it goes to a new webpage and in other it expands the group (which doesn't respond immediately either).

image

We should add something to groups to make them stand out more

[pr page] Break down logs by step

e.g. on https://github.com/pytorch/pytorch/pull/64303/checks?check_run_id=3562967932, the workflow is Lint, the job is mypy, and the step is Run mypy. Right now the logs stop at jobs, but we should let users view at the step level since we often finish each job with some cleanup that creates lots of log spew.

We might need to fetch the entire log archive for the workflow which may be prohibitively large, or we could manually parse out the ##[group] and ##[endgroup] sigils that GitHub adds to logs (this is probably best if JS can handle it)

/status page URL is broken

it redirects straight to /build2/pytorch-master when dialed directly, but when clicking on "status" from that page it loads correctly.

https://hud.pytorch.org/status

semi-related: the JS for the page isn't updating for me on non-incognito windows (i.e. I don't see any of the charts I do when I visit the same page in incognito on Chrome)

Sort PR jobs

We should put the failing jobs first, followed by successful ones, then skipped/grouped jobs

iOS Safari is broken

The status page doesn’t render and just shows a white screen, we’re probably missing some guards somewhere for browser APIs that aren’t on mobile (like Notification)

Using iOS + macbook safari to debug should be pretty simple, maybe we could also mock out the lack of APIs in our tests so too

One click rerun with SSH

We should add a button for GHA workflows on PRs that:

  1. Adds the with-ssh label to the associated PR
  2. Triggers a re-run of the workflow
  3. Monitors workflow annotations and reports the SSH command when its available

Builds page is too wide

Right now it goes way off my laptop's screen and has come up a few times (see #68), a couple of solutions we came up with internally:

  • freeze PR number column so you don't have to scroll around to get it
  • add option to hide columns where all builds are both present + passed

Mark known erroneous failures so people don't worry about them

Infra failures, for example, could be marked as something other than red X's, to make it easier for people to focus their attention on new/unknown/genuine failures.

Another possibility (as opposed to identifying these automatically) would be to just let people update the HUD UX to mark failures as erroneous or reported/mitigated to reduce redundant work.

Consolidate test shards

Test shards are not a logical distinction in CI jobs, they're an artifact of us increasing parallelization in our workflows. We could de-duplicate these on the HUD to compress some info:

  • aggregate status
  • merge test reports

Cannot access to PyTorch build artifacts from hud.pytorch.org/pr

Access to PyTorch build artifacts from hud.pytorch.org/pr/${PR_NUM} is not working after GitHub login and authentication. Page displayed is the same after refreshing except for the authorize_github code. This occurs on incognito/anonymous browser modes on different browsers such as Chrome, Edge and Firefox.

login-attempt-1

Show properly when multiple commits are in a bundle in build2

The build1 page nicely shows all the commit corresponding to a CI run. But build2 doesn't.
It would be helpful to at least get a visual marker on build2 indicating that this is a bundle of commits (so people should check build1 for other commits) or even better have the same multi-line visualization to know what are the other commits.

Windows nightly builds aggregate status is always error even all subitems are green

🐛 Bug

For Group jobs, ci/circle: binary_windows in hud.pytorch.org.
The aggregate status are always red, but all sub jobs are green.

To Reproduce

From the hud.pytorch.org, select and set the Name filter as Windows

image

Expected behavior

If all sub jobs are green, the aggregate status should be green.

Additional details / screenshot

It looks the some circle jobs are still included. For example, https://circleci.com/gh/pytorch/pytorch/17013182?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link, their status are canceled though they're not shown in the UI.
Since all nightly jobs are transferred to Github actions, the backend should not return circleci jobs.

image

Identify existing failures

We could leverage Dr. CI's ruleset / identification to locate existing failures. We can then highlight these on the HUD, use them in PRs to auto-scroll logs to the relevant line, and mark certain failures as already broken / flaky. We'd need some way to extract that information from Dr. CI which we can probably grab from the GitHub comment (and if we're reporting it on the HUD we can hide it behind a <details> tag in the comment.

Integrate CIFlow

Currently probot adds a comment (example to all (non opted-out) PyTorch PRs with information about CIFlow. We should integrate this into the PR HUD and use the authorization token that we already have to control the labels on a PR. In order, we can:

  • implement the same view as the comment but with click-to-add-a-label functionality as its own React component
  • compress the view with some interactive features
  • answer questions like: what labels do I need to add X jobs

[pr page] Show previous runs

We could at the very least query PR commits -> runs on those refs, but this misses job re-runs on a specific commit. These would be useful to revisit historic failures (especially if we do a better job encoding things in the URL these can be easily shared / referenced)

Encode more state in URL

All user intractable elements should encode their state in the URL so pages can be shared, for example:

PR Page

  • current workflow
  • expanded test results
  • expanded log scroll position

HUD

  • name filter
  • group jobs checkbox

[feature]: Dark mode

Feature Request

Add a toggle-able dark mode. It should respect the OS preference first but be toggle-able and saved in localStorage. We should start using SCSS too to make this easier. For color scheme-ing we can probably just copy all the colors from GitHub's dark/light mode

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.