-
-
Notifications
You must be signed in to change notification settings - Fork 1.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[API Concept] - Infinite Query API #4393
base: master
Are you sure you want to change the base?
Conversation
Review or Edit in CodeSandboxOpen the branch in Web Editor • VS Code • Insiders |
✅ Deploy Preview for redux-starter-kit-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
This pull request is automatically built and testable in CodeSandbox. To see build info of the built libraries, click here or the icon next to each commit SHA. Latest deployment of this branch, based on commit c55de7f:
|
Hi! I finally found some free time today to start doing some research on infinite query APIs in other libs like React Query, and am starting to wrap my brain around this space a bit. I haven't tried to dig through this PR yet, but I definitely endorse the idea of I am trying to understand the space enough to have actual feedback, but as a very first starting point, one goal would be that this ought to work as part of the UI-agnostic RTKQ core, so that any UI layer can use it (React, Angular, etc). Looking at React Query's impl, there's a core I haven't even looked at your code yet, so I'm not sure what the specific implementation approach is, but that's the kind of ideal approach I'd like to end up with. Let me see if I can rebase this PR so it's at least up to date and builds. |
…uerying will happen
…of args and forward/backwards direction
…Page functions + Thunk
This reverts commit 4089f11
21d642f
to
bfd8de6
Compare
Okay, I just rebased this. I probably broke things somewhere in the process - the rebase was complicated due to us having drastically rearranged a lot of the internal typedefs a few weeks ago. The one test added to the PR appears to run and pass. There's a dozen TS errors, which look to be related to places where we expect values like FWIW, @phryneas separately commented that he happens to like SWR's infinite query API a bit better:
I don't know enough about this space yet to have an opinion here :) I think it's worth pushing this branch forward and seeing if we can reach feature and test parity with React Query's behavior, but maybe also putting together another branch that takes more of an SWR-style approach and see how they compare. Also, I'm going to be seeing Dominik ( @TkDodo ) at React Advanced and we'll try to chat about this topic as well. |
Okay, I think this at least passes enough to let the CodeSandbox CI build succeed, so we should have a viable PR package preview to play with now! Off the top of my head, the big open questions are:
|
Hm, in my email notification there is a quote from @phryneas but it seems edited out now
Of course I defer to his experience and expertise, but it's kind of funny because it's something that seemed desirable for us... We have a direct message UI that uses a paginated API request, which we further enhanced with WebSockets for update/delete of existing messages and the addition of new ones. Currently we're using the If there are separate cache entries for each page, now we have to iterate through them all in order to find the entity to update/delete, and we're left with a dilemma of where to add new messages that came from the WebSockets instead of the paginated REST requests...now the page size will be out of sync if the user scrolls around and triggers new queries. (I guess with our current I'm super curious to understand what formed the opinion that "merging multiple pages into one cache object is not a good choice" :) and it would be great to have this use-case considered, as I imagine it could very well be a common one. |
Both ways are not fool-proof, but at this point I believe that having to maintain one big cache object can be a burden to the developer, and I would like to avoid going that route in a new implementation. Some thoughts: In cursor-based pagination, additions and deletions can work easier with multiple pages that get stitched.
With offset-based or page-based pagination, you kinda end up with the opposite problem, where the "big blob" approach seems simpler, but the SWR api helps here by passing in the So... cursor-based is in my eyes easier, offset-based not necessarily more easy or difficult, just different.
That could be done without iterating, with
That's why I like the SWR approach, where a page could grow or shrink without causing a lot of trouble. |
@phryneas let’s discuss this in person in London if you’re there, just some quick, high level thoughts:
I don’t think we have problems with pages shrinking / growing when you have one cache entry, as each page is still stored separately. It’s fine to have one request return 10 pages, and then the next one just returns 9. You can also just delete one item from a page, that doesn’t mean an item from the next page must be moved manually - it can just stay the way it is. One thing that’s a conscious decision for us to have one big cache entry is that it commits or errors (or retries) as one entity. That means the page only renders and receives an update after all refetches are done. Assume someone added an entry on the first page, which will “shift” all follow-up pages. If they are separate cache entries - wouldn’t you see the UI temporarily show duplicate entries? And wouldn’t it stay that way if fetching the 2nd or 3rd page fails?
This seems great - the cursor is an input to the other cache entries, making them refresh automatically. But it also means that updates to pages before my page aren’t reflected. If someone renames an entry on page 1 of 5, and I rename something on page 3 of 5, only 3,4,5 will refetch and I won’t know about the change on page 1. Also, refetches staring from a page in the middle can be weird with paginated queries. Suppose I change an entry on page 3 (pageSize=3) and someone deletes 3 entries before that:
Let’s rename H to X, while at the same time, C, D and E were deleted by others. If we only start to refetch with
so the entry we updated isn’t even visible anymore, while the database actually has:
so I think the only safe thing to do is to refetch all pages when you refetch an infinite query, from the start. At least this is what we’re doing and I’m trying to talk sense into that approach 😂 |
@TkDodo that sounds like you are much closer to the "multiple cache entries" than the "single cache entry" I'm arguing against - in our case, I'd use a selector to combine them into a single "outside visible cache entry", so I don't think we're far away from each other at all :) But yeah, let's definitely talk about this in London! |
This draft is in a rougher state (especially around types) as it was mostly me forcing some things to get it working, but given the activity on it again; I'll hop in and clean it up, set some tests and update the discussion on the state of it. Honestly need to refresh myself with the problem space as I haven't done anything here since opening it :D |
This is a conceptual display of an API and how it would work inside RTKQ for an infinite Query. This is not a final implementation.
It is derived from the API for react query Infinite Query but with RTKQ's useInfiniteQuery hook etc and implementation.
disclaimer:
The typing and actual code for a library and implementation is bad, it basically just takes and repeats 90% of the query hook logic, the final implementation would more likely be an extension of the Query definition, but I wanted to completely separate it to make it more clear for feedback on the API.
Summary
useInfiniteQuery
hook - works almost the same asuseQuery
New args:
getNextPage
getPreviousPage
New Returns:
fetchNextPage
trigger function, similar to a lazyQuery trigger but combined with the querySubscriptionfetchPreviousPage
hasNextPage
- I haven't implemented it yet :DhasPrevPage
- I haven't implemented it yet :DisFetchingNextPage
isFetchingPrevPage
InfiniteQuery
is a new EndpointDefinitionUses its own initiator, and then initiates a typical
QueryThunk
Additional logic added to
querySlice
that adds direction/param to the querySubState and acts as the discriminator for an InfiniteQuery (different to arg which acts as the set cache key)ExecuteEndpoint
is changed to fetch every page from the pageParams that hasn't been fetched yet and add to the data object in the direction specified.Still needs to be done:
hasPrevPage
&hasNextPage
stateOpen Questions