Hi. We have a domain graph.cool that used to serve a specific app on Netlify. We shut down that service, and for that build a new “Sunset site” to replace the existing website and documentation (by redirecting to the homepage of the domain) - deployed as a new app on Netlify as well.
Yesterday evening I removed the domain from the old app and put it on the new app. On first look everything seemed fine: Homepage shows our Goodbye letter, and all subpages redirect to that homepage.
Problem: Today we got reports that some subpages are still showing the old content for some users.
More a problem: We can even reproduce that ourselves, sometimes and in some browsers.
Does not solve the problem unfortunately. More pages not show the correct new page for me, but Thank You for Using Graphcool 💚 for example still shows the old app.
Some background on the setup:
The old graph.cool used a rewrite to another app for the /docs routes.
I now removed all rewrites and redirects from the old deployment (which should not even be shown any more under the domain, but who knows) and it is still being delivered to my browser.
Proxying between Netlify sites without careful configurating, and moving hostnames between sites can have unexpected effects like these. I just looked at our traffic logs for www.graph.cool and saw and cleared up some of that problem (there were three CDN nodes that had some of the proxy’d content cached that was trumping your updates). I’m not quite sure I got all of it, but based on “which deploys we serve under that namespace” from a live traffic stream, things are pretty vastly improved.
For future reference, a redeploy of the proxy’d to site would have cleared the problem up in the same way I did, so that’s worth a try next time.
Can confirm, I personally can not reproduce the problem any more at all - all pages now redirect to the intended page. Still waiting for feedback from colleagues across the world, will report again if something does not work as intended.
In this case, I meant the docs site. The “wrong content” was the formerly-correct content, from the docs site. I simulated a redeploy, which clears its cache (which was stuck in the other site’s “namespace” since served at the same hostname), which resolved the issue.
Definitely do let us know if you see anything like this again!
