-
Notifications
You must be signed in to change notification settings - Fork 10.3k
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
[json-loader] Don't emit new file node until previous is finished processing #4785
Conversation
This is an experiment to use [xstate](http://davidkpiano.github.io/xstate/docs/#/) to setup state machines to better handle complex state changes as we sometimes have. Ideally this happens in core and then gatsby-source-filesystem just has a simple queue and emits a new file node every time the system returns to idle. In a future refactor we'll do that plus refactor other parts of core that should be handled in a state machine e.g. pages-query-runner.js This PR also reinforced the need for us to implement [tracing](https://github.com/jaegertracing/jaeger) in core / plugins as that'd make it far far easier to understand what's happening and when.
Nice article about tracing and why it's so much nicer than logs if you haven't read about it before https://medium.com/opentracing/take-opentracing-for-a-hotrod-ride-f6e3141f7941 |
This looks really nice! Reading this on my phone, are all three CHOKIDAR_ states used? |
Nope! Good point — I think I intended to use something with the other states but didn't. I'll just make them NOT_READY and READY. Nice to see that they were readily understandable! |
This looks really nice - very cool that state transitions "definition" is kept in one place to get overview on possible states and general flow. Do You want to merge it as-is or just use it as proof of concept? I guess filesystem is only real source plugin that will update mid develop (refresh hook doesn't seem to be actively used), so this could definitely go in. Only thing for me apart from extra states pointed out by @m-allanson would be some explanation what problem this solves for people who will try to figure it out later on. |
Yeah — I was planning on doing a general purpose solution but realized that gatsby-source-filesystem is the only way right now to get Gatsby in this sort of inconsistent state. So build a specific fix now that we can generalize in the future once we understand the problem better and possible solutions better. I'll add some comments on the states + remove the extra Chokidar states. |
…ion (aka Ludicrous Mode) (#4555) * Create placeholder JSON store * Rename * Websocket placeholder * Push query results JSON over websockets * More descriptive variable name * Fix queries being overwritten * Remove eslint-disable flag * Remove junk * test require error fix for windows * dont require json data in sync-require * dont add layout data to json array multiple times * initial async loading * revert saving json directly to public for now * updated production-app to sync with prop name change in ComponentRenderer * we load json data via json-loader component in develop and not handling it with webpack import/require * hashes for json files * fix preloading, use xhr instead of fetch - for some reason can't force fetch to not create additional request, with any `cache` or `mode` configuration * dont use full paths in dataPath - remove static/d/ path and .json ext - results in smaller app bundle especially with large ammount of pages * Enable cached query results to be loaded * Don't dump all query results out to the client Instead only push results out if the data is for a path that's currently being viewed in a client. * fix preload link to json data * remove not used function * remove more not used code * Update to latest webpack/mini-css-extract-plugin * don't write new (a)sync-requires.js if components didn't change (#4759) * create just one websocket client (#4763) * Filter out duplicate query jobs and create secondary queue for jobs if path already has query in flight * [json-loader] Don't emit new file node until previous is finished processing (#4785) * Don't emit new file node until previous is finished processing This is an experiment to use [xstate](http://davidkpiano.github.io/xstate/docs/#/) to setup state machines to better handle complex state changes as we sometimes have. Ideally this happens in core and then gatsby-source-filesystem just has a simple queue and emits a new file node every time the system returns to idle. In a future refactor we'll do that plus refactor other parts of core that should be handled in a state machine e.g. pages-query-runner.js This PR also reinforced the need for us to implement [tracing](https://github.com/jaegertracing/jaeger) in core / plugins as that'd make it far far easier to understand what's happening and when. * Document state machine and remove extraneous Chokidar states * Remove console.log * [json-loader] Only log file events if we're past bootstrap (#4826) * Don't emit new file node until previous is finished processing This is an experiment to use [xstate](http://davidkpiano.github.io/xstate/docs/#/) to setup state machines to better handle complex state changes as we sometimes have. Ideally this happens in core and then gatsby-source-filesystem just has a simple queue and emits a new file node every time the system returns to idle. In a future refactor we'll do that plus refactor other parts of core that should be handled in a state machine e.g. pages-query-runner.js This PR also reinforced the need for us to implement [tracing](https://github.com/jaegertracing/jaeger) in core / plugins as that'd make it far far easier to understand what's happening and when. * Document state machine and remove extraneous Chokidar states * Remove console.log * Only log file events if we're past bootstrap * [json-loader] dont recompile on data change - part 2 (#4837) * prevent adding duplicate redirects * don't write new `redirects.json` if redirects didn't change prevents webpack recompilation on data change * [json-loader] develop - reading results from file improvments (#4850) * dont emit results for layouts * [develop] store query results in memory, read json data from file only if we don't have it stored yet (we didn't run this query, but results are cached) * Add query prioritization based on what page(s) user(s) are on Query running is sadly not very ludicrous right now on gatsbyjs.org — not sure why — each markdown file change causes ~20 queries to run but even with prioritizing the active page's query, it's still ~2 seconds before the page updates. This sort of thing will be much easier to debug with tracing support. * Add initial forward slash * Actually this is how we add back the initial forward slash
@davidkpiano hey David! Thanks for stopping by and for the great library! Had a great time watching your talk and reading through your docs and using it to implement this PR! We'll be using xstate a bunch more in Gatsby in the future 💪 Cool visualization! How did you make it? |
It was automatically generated - based on this visualizer, but I'll be releasing an upgraded version at React Finland: https://codepen.io/davidkpiano/pen/ayWKJO |
Nice! Looking forward to watching your recent JSConf talk https://www.youtube.com/watch?time_continue=1&v=RqTxtOXcv8Y |
…ion (aka Ludicrous Mode) (gatsbyjs#4555) * Create placeholder JSON store * Rename * Websocket placeholder * Push query results JSON over websockets * More descriptive variable name * Fix queries being overwritten * Remove eslint-disable flag * Remove junk * test require error fix for windows * dont require json data in sync-require * dont add layout data to json array multiple times * initial async loading * revert saving json directly to public for now * updated production-app to sync with prop name change in ComponentRenderer * we load json data via json-loader component in develop and not handling it with webpack import/require * hashes for json files * fix preloading, use xhr instead of fetch - for some reason can't force fetch to not create additional request, with any `cache` or `mode` configuration * dont use full paths in dataPath - remove static/d/ path and .json ext - results in smaller app bundle especially with large ammount of pages * Enable cached query results to be loaded * Don't dump all query results out to the client Instead only push results out if the data is for a path that's currently being viewed in a client. * fix preload link to json data * remove not used function * remove more not used code * Update to latest webpack/mini-css-extract-plugin * don't write new (a)sync-requires.js if components didn't change (gatsbyjs#4759) * create just one websocket client (gatsbyjs#4763) * Filter out duplicate query jobs and create secondary queue for jobs if path already has query in flight * [json-loader] Don't emit new file node until previous is finished processing (gatsbyjs#4785) * Don't emit new file node until previous is finished processing This is an experiment to use [xstate](http://davidkpiano.github.io/xstate/docs/#/) to setup state machines to better handle complex state changes as we sometimes have. Ideally this happens in core and then gatsby-source-filesystem just has a simple queue and emits a new file node every time the system returns to idle. In a future refactor we'll do that plus refactor other parts of core that should be handled in a state machine e.g. pages-query-runner.js This PR also reinforced the need for us to implement [tracing](https://github.com/jaegertracing/jaeger) in core / plugins as that'd make it far far easier to understand what's happening and when. * Document state machine and remove extraneous Chokidar states * Remove console.log * [json-loader] Only log file events if we're past bootstrap (gatsbyjs#4826) * Don't emit new file node until previous is finished processing This is an experiment to use [xstate](http://davidkpiano.github.io/xstate/docs/#/) to setup state machines to better handle complex state changes as we sometimes have. Ideally this happens in core and then gatsby-source-filesystem just has a simple queue and emits a new file node every time the system returns to idle. In a future refactor we'll do that plus refactor other parts of core that should be handled in a state machine e.g. pages-query-runner.js This PR also reinforced the need for us to implement [tracing](https://github.com/jaegertracing/jaeger) in core / plugins as that'd make it far far easier to understand what's happening and when. * Document state machine and remove extraneous Chokidar states * Remove console.log * Only log file events if we're past bootstrap * [json-loader] dont recompile on data change - part 2 (gatsbyjs#4837) * prevent adding duplicate redirects * don't write new `redirects.json` if redirects didn't change prevents webpack recompilation on data change * [json-loader] develop - reading results from file improvments (gatsbyjs#4850) * dont emit results for layouts * [develop] store query results in memory, read json data from file only if we don't have it stored yet (we didn't run this query, but results are cached) * Add query prioritization based on what page(s) user(s) are on Query running is sadly not very ludicrous right now on gatsbyjs.org — not sure why — each markdown file change causes ~20 queries to run but even with prioritizing the active page's query, it's still ~2 seconds before the page updates. This sort of thing will be much easier to debug with tracing support. * Add initial forward slash * Actually this is how we add back the initial forward slash
This is an experiment to use
xstate to setup state
machines to better handle complex state changes as we sometimes have.
Ideally this happens in core and then gatsby-source-filesystem just has a
simple queue and emits a new file node every time the system returns to
idle.
In a future refactor we'll do that plus refactor other parts of core that
should be handled in a state machine e.g. pages-query-runner.js
This PR also reinforced the need for us to implement
tracing in core / plugins as
that'd make it far far easier to understand what's happening and when.