Partitioned cookies are keyed by the top-level site. This does not affect the security guarantees of __Host, being that your cookie is valid for roughly the origin of the requestor, and will not be overriden using separate subdomains / paths. CHIPS used to require __Host to motivate better security on the web but that requirement was removed again, see #43.

Why do you think that in the above example the hostname is included in the partition key?

Or maybe put differently, in your application, what makes you believe that __Host is not working correctly?

The partition key for the cookie is displayed in the dev tools (Application -> Cookies). Go to chips-site-a, and view the partitioned cookie for chips-site-b. The Partition Key reads: "https://chips-site-a.glitch.me".

For my experiment, it would read "glitch.me" even when I use the __Host prefix (I am not actually using the same hostnames, I am only using them here to illustrate the point). The quote I sent above sounds like the prefix should resolve the issue. Though, the prefix may not be a possibility for me, I wanted to know if it resolves the issue. It seems it does not.

Although the glitch example doesn't clearly demonstrate this, I believe this could be a problem for my use case since both chips-site-a and chips-site-b have access to the chips-site-b cookie created when chips-site-a is the top-level domain. If you go back to the article and scroll to the What is CHIPS?: Partitioned section, the second figure would be violated, since both A and C sites share a domain and the partition key is using that domain only and not the hostname.

The concern for our use case is if we have customers (A and B) using the same domain. A user for customer A goes to their site, cust-a.example.com, and obtains a partitioned cookie for our service (separate domain). If that user then navigates to cust-b.example.com, they will be able to use that same partitioned cookie since the partition key is only matching example.com at the top-level. (Which is a violation of the first figure in the What is CHIPS?: Partitioned section.)

Cookies are always partitioned by top-level site, which is scheme + registrable domain, e.g. https://example.com. However, in the Glitch example, the top-level site is https://chips-site-a.glitch.me because glitch.me is on the PSL for security reasons (doesn't want to allow sharing cookies between its subdomains). So, yes, it is a bit misleading in that instance but unfortunately it's hard to host these examples without this kind of protection.

Hostname-bound cookies, on the other hand, do not affect the partition key. What __Host does is it prevents other hosts under your own registrable domain from overriding your cookie and thus potentially attacking you.

Unfortunately I think for your use case this means you'll have to find a solution that is tolerant of the partition sharing you described above. But I'm curious why this is a problem for you, until very recently CHIPS didn't exist at all, shouldn't your service already be used to third-party cookies being shared by default?

@johannhof Thank you for the clarification! That explains what I am observing.

As for our current solution, the goal is to improve the current scheme. CHIPS may not get us there if this is a valid production concern (I have an inquiry to determine if this is possibly a case unique to our dev/test environments). If it is a production concern, then we need a different solution as you suggest.

Ok, that's good to hear! With that resolved, we can either close this issue or you could change it to "Feature request: Opt into origin-partitioned cookies", which seems to be what you're looking for. To be clear I don't think this is something we'd consider doing right now, but it might help other developers add their support for the idea.

Just an update on our specific situation. It does look like it is a valid concern for us and thus CHIPS will not be a valid long-term solution for us without a way to create origin-partitioned cookies. Thanks!

Are you specifically asking for a tighter partition?

Though if you have a a.example.com and b.example.com setup where a/b are not fully under your control, shouldn't example.com be registered as public suffix?

Yes, the request is for a way to opt-in for the partition key to be set to the entire top-level hostname and not just the top-level domain.

Ideally, it would be nice if they were all registered. We don't control our customers domains, so I am not sure we can rely on all services that provide the sub-domains to register their domains as public suffixes.