I was accidentally serving images _without_ `updatedAt` with the long-term cache headers, but only in prod, not in dev
I think this is probably because I made a mistake in my Vercel route config, by including an `$updatedAt` parameter that doesn't actually exist in this case.
I'm guessing the Vercel dev server has different behavior for my mistake than the prod server. I'm guessing the dev server makes it an empty string, and the prod server makes it the literal string `$updatedAt`. Oops!
I hope this fixes it!
Oops, right, I forgot for a while that GraphQL fields have a special syntax for docstrings, and it's not just comments! This will help stuff show up in our GraphQL Playground API docs correctly 🥰
Someone asked to use the DTI API for a small client-side project, so I'm making this change to support it!
As explained in the comment, I think this should be safe regarding CSRF attacks. But it _does_ increase the risk that someday we change something elsewhere that creates a problem, like using cookies to authorize something. So, let's remember to be careful! (as I would hope we would be when adding another auth mechanism!)
I'm not sure which image url is better to return from stuff like this, and I don't actually have a use case for it anymore, so let's just clear it out until we need something like it!
Gonna have the /outfit-urls page start returning these instead, for feature parity with before
I might change the strategy on this at some point, like have it get `updatedAt` and redirect instead of generating the image. But this is simpler for now (and the Vercel cache doesn't seem to be as aggressive as I want anyway), and I can change it later!
Oops, we get a _lot_ of outfit image requests, and it's pushing the limits of our free Honeycomb plan! But I don't really need all that much detail, because there's so many.
So, we here apply sampling! `api/outfitImage` is getting a 1/10 rate, and for GraphQL, `ApiOutfitImage` is getting 1/10, and `SearchPanel` is getting 1/5.
I had to add a `addTraceContext` call, to give all the child events awareness of what operation they're being called in, too!
I haven't actually tested that this is working-working, just that the endpoints still return good data. We'll see how it shakes out in prod!
But I did add `console.log(sampleRate, shouldSample, data);` to the `samplerHook` briefly, to see the data flow through, and I reloaded a `SearchPanel` request a few times and observed a plausibly 20% success rate.
Oops, I wasn't requesting `bodyId` for item layers, so the check for `layer.bodyId !== "0"` was always true—because it was always `undefined`, even when it should have been `"0"`.
This wasn't an issue on the client, because the client _does_ request `bodyId` for caching item appearances between pets of the same body, and I didn't realize that it needs to be part of this fragment too!
Mm right, when we first render the output, `imageUrl` is `undefined`, so the output textbox renders with `value={undefined}`, which is an "uncontrolled" component that the DOM is free to change.
In practice, this isn't an issue because the textbox has `isReadOnly`, so the user can't _actually_ change it. But it's still a good idea for consistency and clarity to use an empty string instead of `undefined`, and it removes warning spam from my console!
Oops, previously the MajorErrorMessage was willing to shrink the width of the cute Grundo Programmer icon, to allow error messages with long words to avoid word breaks.
Here, we switch `1fr` for `minmax(0, 1fr)`, which allows the text zone to get smaller. (`1fr` is short for `minmax(auto, 1fr)`, which isn't capable of shrinking smaller than the natural value.)
Now, the error text is more willing to shrink by word-wrapping, than the image is by shrinking the image. Success!
Oops, my cute API idea for `speciesPickerProps` breaks `React.memo`, of course!
We could fix this by having the caller memoize the `speciesPickerProps` object, but that's too unusual and error-prone. We could also fix this by writing a custom function for `React.memo` to determine whether props match, but that seems like overkill when there's only one actual prop we're using here in practice.
So yeah, I've updated `SpeciesColorPicker` to just accept `speciesTestId` and `colorTestId` props instead!
Note that actually this component _is_ still re-rendering too often, because of a Chakra bug I just discovered and reported! So this change won't immediately improve performance, but it should stop re-rendering too often once we _also_ upgrade Chakra after this bug is fixed. https://github.com/chakra-ui/chakra-ui/issues/4080
Oops, right, I meant to use the new `impress-outfit-images.openneo.net` host for this! It works just fine from `impress-2020.openneo.net` as the backing source right now, but I want these semi-permanent URLs to be a bit more decoupled.
This is mostly for decoupling's sake: I want pretty outfit URLs that aren't dependent on the `/api/outfitImage` path structure, so that we can serve them as ~permanent pretty URLs that don't actually indicate where or how the image is stored.
This reverts commit a0121f09ea.
I didn't realize that VERCEL_URL wouldn't use the canonical URL. And thinking further about the Open Graph semantics, I think having just one canonical host makes more sense, even if it makes testing a bit more annoying in some cases.
Right, the routes are a bit confusing, `/outfits/new` matches `/outfits/:id`, but shouldn't.
I fix this at the routing level, by creating a more specific `/outfits/new` route that matches first, and just serves `index.html` right from the CDN.
To check that it's exact, I check for either a `?` or the end-of-string after `new`. That way, `/outfits/newish` correctly redirects to the SSR. (This doesn't actually matter a lot, because outfit IDs are only ever numbers and therefore can't start with "new"? But I figure, better to have an unambiguous semantic, then leave random things relying on outfit ID format assumptions.)
Ah right, this is how we get the deployed API code to know its own hostname! This way, when we do a preview deploy, /api/outfitImage will use the preview version of the GraphQL endpoint, too.
Hmm, the built copy of the HTML isn't deployed with the API function on Vercel. Ok! Let's just do the same trick as we did for the dev server, and make an HTTP request.
This is fine enough for perf because we're caching this result locally to the function, plus it should be a fast CDN-cached response anyway.
Size upgrade, wowie! (I figure sharing services probably prefer a higher-resolution image, they're not very big, and to then scale them down for the application-specific use case & device.)
Meta tags are a bit tricky in apps built with `create-react-app`! While some bots like Google are able to render the full page when crawling, not all bots are. Most will just see the empty-ish index.html that would normally load up the application.
But we want outfit sharing to work! And be cool! And use our new outfit thumbnails!
In this change, we add a new server-side rendering API route to handle `/outfits/:id`.
It's very weak server-side rendering: it just loads index.html, and makes a few small tweaks inside the `<head>` tag. But it should be enough for sharing to work in clients that support the basics of Open Graph, which I think most major providers respect! (I know Twitter has their own tags, but it also respects the basics of OG, so let's see whether there's anything we end up _wanting_ to tweak or not!)
Not using this anywhere in-app yet! But might swap it into the user outfits page, and use it to server-side-render social sharing meta tags!
Also eyeing this as a way to replace our nearly 1TB of outfit image S3 storage, and save $20/mo…
This folder will include code shared by both the client-side app and the server!
The server isn't using it yet, but it will in a new API endpoint soon! I'm doing this in a separate commit to avoid lumping all the import-change noise into that commit.
I'm doing this in preparation for an API endpoint to build outfit images by ID. It'll need the same logic to decide which layers are visible, and the same GQL fragments to load the relevant data!
Oh right, labeling PB items as NP is confusing! Here, we add a "PB" case to the lil badge on the corner of the item thumbnail, in item search page & homepage Newest Items.
Oops, I forgot to grep outside `src` for this, lol! I'll keep the S3 domain support for now, because it's still fine to accept and some clients might be on old code, whatever!
Idk I was surprised to notice that /api/validPetPoses wasn't cached by the Vercel CDN… but I guess that makes sense, because it's only allowed to be an hour old, and client caches hold onto it, so I suppose it makes sense for the CDN cache to not bother.
But in practice, that means users loading the site are pretty _likely_ to see /api/validPetPoses taking its full load time, which can be up to 500ms.
So, I'm offering this additional cache hint, to be willing to serve /api/validPetPose responses up to a week old while reloading the latest data. My hope is that the cache algorithm is much more excited about holding onto that, and that it makes the 500ms delay very _rare_ in practice!
Oops, we did an in-place sort on the search variables we passed to Apollo! This meant that Apollo's first read of the variables wouldn't match later reads, so it would always decide the variables had changed, causing an infinite re-render loop.
Remember to copy existing arrays before sorting! 😅
Incidentally, this only happened for Markings, by coincidence: it's the only (I think) searchable zone label with multiple zone IDs, that don't sort alphabetically the same as they sort numerically. This `.sort()` sorts them alphabetically, whereas they come in numerical order in `allZones`, because that's the order the GQL server returns them in `build-cached-data.js`.
There have been usability problems with this search filter UI, and I think they mostly come down to people accidentally selecting filters when they don't mean to—sometimes pressing Enter to indicate that they're done typing, but accidentally selecting something.
Here, we remove that behavior, and additionally add a new behavior to clear the suggestions on pressing Enter.
We've been serving images directly from `impress-asset-images.s3.amazonaws.com` for a long time. While they serve with long-lasting HTTP cache headers, and the app requests them with the `updated_at` timestamp in the query string; each GET request still executes a full S3 ReadObject operation to get the latest version.
In the past, this was only relevant to users on Image Mode, not Flash Mode. But now that everyone's on Image Mode, this matters a lot more!
Now, we've configured a Fastly host at `impress-asset-images.openneo.net`, to sit in front of our S3 bucket. This should dramatically reduce the GET requests to S3 itself, as our cache warms up and gains copies of the most common asset PNGs.
That said, I'm not sure how much actual cost impact this change will have. Our AWS console isn't configured to differentiate cost by bucket yet—I've started this process, but it might take a few days to propagate. All I know is that our current costs are $35/mo data transfer + $20/mo storage, and that outfit images are responsible for most of the storage cost. I hypothesize that `impress-asset-images` is responsible for most of the reads and data transfers, but I'm not sure!
In the future, I think we'll be able to bring our AWS costs to near-zero, by:
- Obsolete `impress-asset-images`, by using the official Neopets PNGs instead, after the HTML5 conversion completes.
- Obsolete `impress-outfit-images`, by using a Node endpoint to generate the images, fronted by a CDN cache. (Transfer the actual data to a long-term storage backup, and replace the S3 objects with redirects, so that old S3 URLs will still work.)
I hope this will be a big slice of the costs though! 🤞
(Note: I'll be deploying this on a bit of a delay, because I want to see the DNS propagate across the globe before flipping to a new domain!)
