It seems related to us poor single user pods not getting the trust to share twts.. which it seems to still untrust on restart for me.

@xuu@txt.sour.is I added some logging when a “dead” peer is removed as I suspect this to be a hot candidate for all the trouble. https://git.mills.io/yarnsocial/yarn/commit/21538951f9dc71b9366db6dbb784a8078096a4c8 Does this yield anything?

@xuu@txt.sour.is The Pod.LastSeen and Pod.LastUpdated fields are only ever updated in the Cache.DetectPodFromUserAgent(…) function as far as I can tell. This function is called in Cache.DetectClientFromRequest(…) and Cache.DetectClientFromResponse(…).

Cache.DetectClientFromRequest(…) is only invoked when the twtxt.txt is requested and looks at the User-Agent HTTP request header.

Cache.DetectClientFromResponse(…) is only called in Cache.FetchFeeds(…) and looks at the Powered-By HTTP response header. This header would be set in twtxt.txt HTTP responses from yarnd. A bunch of places invoke Cache.FetchFeeds(…), including a periodic job (UpdateFeedsJob.Run()). Maybe something is iffy around these locations.

@lyse@lyse.isobeef.org i trusted all pods yesterday and now when i pull it up they are all untrusted.