Conversation

Notices

1. Embed this notice
   The Nexus of Privacy (thenexusofprivacy@infosec.exchange)'s status on Thursday, 21-Dec-2023 18:16:11 JST The Nexus of Privacy

   • Tokyo Outsider (337ppm)
   • Hrefna (DHC)

   I've heard that for large instances the performance hit can be significant -- it makes caching much harder (or maybe disables it completely). Also, it turns off "post forwarding" (whatever that is) which apparently makes the "missing replies" problem (where you only see part of a conversation due to federation weirdnesses) worse; Byron told me this was an issue on Univeseodon when he tried enabling auth-fetch. I'm not sure about the details of the incompatibiliies, I remember somebody (maybe @hrefna ?) metnioning that Mastodon uses an old draft version of the HTTP signature spec so maybe it has something to do with that.

   @tokyo_0 @Sibilant

   • Embed this notice
     The Nexus of Privacy (thenexusofprivacy@infosec.exchange)'s status on Saturday, 23-Dec-2023 01:36:46 JST The Nexus of Privacy

     in reply to

     • Tokyo Outsider (337ppm)

     @tokyo_0 Claire's comment on the PR btw mentions this still has downsides: "increased resources consumption because of systematic request caching and inability to have reverse proxy caching"

     @Sibilant

   • Embed this notice
     Matthew W. Thomas (mwt@econtwitter.net)'s status on Saturday, 23-Dec-2023 01:36:46 JST Matthew W. Thomas

     in reply to

     • Tokyo Outsider (337ppm)

     @thenexusofprivacy @tokyo_0 @Sibilant the only instance I know of that has it is also one that I am unable to fetch from (on Mastodon despite no blocks). So, I think that there is much more than a performance hit. It's also unreliable and difficult to debug.

   • Embed this notice
     The Nexus of Privacy (thenexusofprivacy@infosec.exchange)'s status on Saturday, 23-Dec-2023 01:36:47 JST The Nexus of Privacy

     • Tokyo Outsider (337ppm)

     @tokyo_0 that's certainly true: "we can't provide the basics of security because our implementation is too inefficient" is a very weak position.

     It seems plausible to me that auth fetch would cause problems for a caching mechanism that was originally written without taking auth fetch into account. It could just be that nobody's looked (yet) at how to make it more efficient. But it could also be that each request from a different account really does require separate checking when auth fetch is turned on in ways that aren't the case when it's turned off. I haven't looked at the code let alone run a profiler so dunno. With Claire's PR turning it on, I guess we'll find out!

     @Sibilant