This looks good to me! 👍

Hey folks! re: complexity I sort of agree - I could move this to a separate file / plugin type of thing, but it's definitely less straightforward than I'd hoped. We could switch to just storing line numbers that are folded instead of YAML paths, but I think being able to move between tabs and have your folding preserved is kind of nice, and with just indexes we might apply the folding incorrectly and not know. (eg: "fold line 12" might result in something being incorrectly folded if the YAML had changed.)

• Fix out of sync Cancel.ts

Rebase

• Add cursor stack for back-pagination, rename selector to PipelineRunsFilter

Looks great!

• Client-side pagination and server-side search, addition of pipelineRunTags root query

Looks good to me! I was seeing an exception on kwargs['mode'] so this should fix it 👍

Rebase

Rebase

• Add missing type

Move the reloader interface to dagster-graphql, fix tests again

• Make the reloader optional in create_app as well

Define a Reloader ABC in dagster core and move the implementation to Dagit, fix tests by making it optional

Replaces the window.reload() at the beginning of the refresh, which only worked when the React app was running on a separate port in dev mode, with a "soft reload" and a toast to let you know when the reload is complete.

Rebase on master to fix build(?)

Abstract custom subscription into DirectGraphQLSubscription, implement retries

Hey @alangenfeld this should be ready to go now. I did some research and it looks like watchmedo only allows you to watch directories, not single files, which is why the watch dir is a separate temp directory with one file in it. I was afraid that if we watched the same folder that it uses as a fallback for DAGSTER_HOME, it'd be possible for it to hit the "too many files to watch" scenario again? Let me know what you think

• Move reloading into a separate class given to the GraphQL layer

Rename _open => _data

• Address diff feedback

Sounds good! I think handling reconnects is a good idea - will see if we can do that and pull the websocket impl out into a class for reuse.

No - renaming modes will be a separate fix, working on that one now!
Will put a better name on _open

Add file I forgot to commit

Sync pipeline run status changes back to Apollo
Rename and move types a bit

Remove extra graphql types, use debounce because smaller updates are smoother than infrequent large ones.

Ran this locally and it looks good to me 👍

• Basic diff feedback
• Remove the loading spinner
• Add to changelog

Hey folks! I pushed up a few small changes to address the inline comments.

Fix tests, black formatting, snapshots

Ugh apparently I have to rebase and not merge master

• Merge branch 'master' into bengotow/-run-stats

• Remove cancel

• Avoid Date.now, add better explanation of sort logic
• Show number of steps out of total
• Run check in PipelineRunStats constructor

Update the text labels after discussion with team

Rebuild snapshot tests

Good catch—I switched this over to using onFocus and onBlur on the input to manage the open / close state rather than inheriting behavior from the popover and that seems to work much better!

Rebase and address feedback

Rebasing

Make runId searchable, support /runs?q=<term> so nobody is copy-pasting runIds

Sure! Don't see why not. While I'm at it I'll make it so this search field can be populated from the query string - If there's some other place people are copy-pasting runIds from, it'd be cool to switch to a dagit link.

Hmm, I'll see if we can pull the total number of steps - being able to show a total count rather than accumulating them as we go would be nice. I think we may be pulling the execution plan into the UI already so it's probably not a big leap.

Sounds good! I think this snapshot testing isn't working out very well because the mock data for Dagit is very hard to generate / very large compared to your typical react application, and is still changing often enough that we need to be able to re-generate it or it'll fall out of sync with the actual runtime values pretty quickly. I'm going to abandon this and I'll see if we can start writing more component-level / panel-level tests that might require less sophisticated / more piece-meal mock data.

• Fix sort of runs with no start time

Switch to True/False

Sounds good! The reason this exists vs. doing each dialog inline is just because the dialog doesn't animate open / closed unless it's container is in the DOM when it's open AND closed (set via the <Dialog isOpen={}> in this case). So if you're not careful, you can end up with scenarios where every single row of a table contains an invisible not-yet-opened dialog DOM tree (or you just lose the animation, which would be better).

• Turn on React hooks eslint rules

Ahh good call re: React hooks eslint rules. I turned them on and verified that they work by breaking hooks rules in a few places. Very slick. Will wait for this build to pass and get this merged!

Just a few inline comments—nothing blocking!

Rebase

• Update autogenerated types

• Update app snapshot tests to reflect new page URLs

• Update snapshot tests
• Remove complexity from CustomAlertProvider in favor of passing a ReactNode
• Change paths to /explore/<pipeline> instead of <pipeline>/explore
• Remove occurrences of if \([^\)]*\)$ in the TS (two line no-bracket ifs)
• Consolidat time and stepKey formatting to the same utils
• Tick up the elapsed time when runs are in-flight
• Rebase
• Switch <StorageProvider> to a hook to clean up the redirect
• Move the hack to a separate page to keep things really nice and clean
• Move the hack out of the main pipeline execution page
• Revert changes to apollo version
• Fix item missing in package cache
• Rebase on master