Permanence of streams

[tulshi]

Copying discussion from meeting on 9/24:

• [Tushar] the spec implicitly assumes that a stream is permanent once created. Should it support expirable streams?
• [Tushar] Possible solution: verification events on a cadence
• [Yair] A Rx (or Tx) can disable a stream
• [Yair] As long as PUSH endpoint is returning success, nothing is wrong. But if endpoint starts sending 500, 400 etc, then Tx can do something about it.
• [Yair] You cannot tell the difference between an Rx that has been sold, etc vs an Rx that is behaving well
• [Yair] We shouldn't make verification event requests mandatory, but can make it optional. Possibly include a time value that indicates how long between requests before Tx kills stream.
• [Shayne] I like the idea of an optional value in StreamConfiguration
• [Apoorva] We could also make it any call, not just verification
• [Shayne] Is that harder to implement?
• [Yair] Agree that it could be any endpoint - status endpoint would work
• [Shayne] That's nice because then you don't have to send unneccessary data

[tulshi]

Action item recorded on 9/24:

• [Tushar] Send PR to clarify that streams can optionally expire, and how to prevent that

[keylimesoda]

+1 on Yair's comments.

It can be expensive for a transmitter to continue to transmit indefinitely to defunct receivers. It is potentially insecure to allow streams to exist indefinitely.

An optional "time remaining" or "expiration timedate" property on a Stream Configuration communicates a finite lifecycle to the Receiver. The property could be updated by the Transmitter after successful verification event, or via PATCH request from receiver.

[TusharR-google]

I'd like to propose that this be added to v1Final instead of vFuture.

Reasoning
The main issue is security.

Permanent access to data (security events) for PUSH receivers is a potential security hole. We can contrast this with something like Oauth grants that are limited in both scope and duration.

The duration limit is missing, once the stream is opened by a PUSH Rx, the Tx is expected to continue sending events until the stream gets paused or deleted.

This also leads to asymmetry in the PUSH and PULL flows, where the PULL Rx is re-authed every time it pulls events, while the PUSH Rx may not communicate again for many years after creating the stream.

It would be great to have a way in the spec for Tx that are going to auto-expire streams (currently strongly preferred for the Google Tx implementation) to communicate this fact.

I'll send out a PR shortly.

Cost Estimate vs ID3
This only affects PUSH streams, and the Rx changes should be minor, considering that most affected Rx will already have an automated way of calling Tx methods.

Most Tx/Rx pairs will be ID3 backwards compatible. Some Tx will optionally communicate a stream expiry duration, in which case the Rx will periodically perform a stream operation or request a verification event to refresh the expiry.

[tulshi]

I've added this to the agenda of the meeting on Nov 19, 2024.