How can I configure Polymer's platinum-sw-* to NOT cache one URL path?

Question

How can I configure Polymer's platinum-sw-cache or platinum-sw-fetch to cache all URL paths except for /_api, which is the URL for Hoodie's API? I've configured a platinum-sw-fetch element to handle the /_api path, then platinum-sw-cache to handle the rest of the paths, as follows:

<platinum-sw-register auto-register
                      clients-claim
                      skip-waiting
                      on-service-worker-installed="displayInstalledToast">
  <platinum-sw-import-script href="custom-fetch-handler.js"></platinum-sw-import-script>
  <platinum-sw-fetch handler="HoodieAPIFetchHandler"
                 path="/_api(.*)"></platinum-sw-fetch>
  <platinum-sw-cache default-cache-strategy="networkFirst"
                     precache-file="precache.json"/>
  </platinum-sw-cache>
</platinum-sw-register>

custom-fetch-handler.js contains the following. Its intent is simply to return the results of the request the way the browser would if the service worker was not handling the request.

var HoodieAPIFetchHandler = function(request, values, options){
  return fetch(request);
}

What doesn't seem to be working correctly is that after user 1 has signed in, then signed out, then user 2 signs in, then in Chrome Dev Tools' Network tab I can see that Hoodie regularly continues to make requests to BOTH users' API endpoints like the following:

http://localhost:3000/_api/?hoodieId=uw9rl3p
http://localhost:3000/_api/?hoodieId=noaothq

Instead, it should be making requests to only ONE of these API endpoints. In the Network tab, each of these URLs appears twice in a row, and in the "Size" column the first request says "(from ServiceWorker)," and the second request states the response size in bytes, in case that's relevant.

The other problem which seems related is that when I sign in as user 2 and submit a form, the app writes to user 1's database on the server side. This makes me think the problem is due to the app not being able to bypass the cache for the /_api route.

Should I not have used both platinum-sw-cache and platinum-sw-fetch within one platinum-sw-register element, since the docs state they are alternatives to each other?

Answer 1

In general, what you're doing should work, and it's a legitimate approach to take.

If there's an HTTP request made that matches a path defined in <platinum-sw-fetch>, then that custom handler will be used, and the default handler (in this case, the networkFirst implementation) won't run. The HTTP request can only be responded to once, so there's no chance of multiple handlers taking effect.

I ran some local samples and confirmed that my <platinum-sw-fetch> handler was properly intercepting requests. When debugging this locally, it's useful to either add in a console.log() within your custom handler and check for those logs via the chrome://serviceworker-internals Inspect interface, or to use the same interface to set some breakpoints within your handler.

What you're seeing in the Network tab of the controlled page is expected—the service worker's network interactions are logged there, whether they come from your custom HoodieAPIFetchHandler or the default networkFirst handler. The network interactions from the perspective of the controlled page are also logged—they don't always correspond one-to-one with the service worker's activity, so logging both does come in handy at times.

So I would recommend looking deeper into the reason why your application is making multiple requests. It's always tricky thinking about caching personalized resources, and there are several ways that you can get into trouble if you end up caching resources that are personalized for a different user. Take a look at the line of code that's firing off the second /_api/ request and see if it's coming from an cached resource that needs to be cleared when your users log out. <platinum-sw> uses the sw-toolbox library under the hood, and you can make use of its uncache() method directly within your custom handler scripts to perform cache maintenance.