Co-authored-by: Sam Saffron <sam.saffron@gmail.com>
Co-authored-by: David Taylor <david@taylorhq.com>
This gives more control over the request. In particular we can easily
lookup DNS dynamically, instead of only upon NGINX startup.
Previously, NGINX was looking up IP for the letter avatar service and
caching the CDN IP address, this caused issues if CDN changed IP, in
which letter avatars would be broken till a container restarted.
NGINX config has been updated to add caching. This change will require
a container rebuild.
The proxy will now function in development environments, so the patch
for `letter_avatar_proxy` has been removed.
This is crucial in multisite installations, because otherwise the nginx logs
are fairly useless, however it can also be quite handy to know what
hostnames are being sent to your site. The variable is quoted, because it
is untrusted input (it is taken directly from the HTTP request), but nginx
helpfully escapes the quoting character automagically, so we don't have to
worry about that.
For now, the log analysis plugin *recognises* the new log format
(and continues to recognise the previous format, for backwards
compatibility), but doesn't do anything with the new log entry field. This
means your multisite performance plugin data is still broken, but it's no
worse than it was before.
On sites that don't otherwise configure an avatar fallback, Discourse will
now tell the client to get its letter avatars from a location which nginx
proxies to the centralised `avatars.discourse.org` service. This alleviates
privacy concerns, whilst still providing some degree of performance benefit
(no need for every site to delay avatar response by 300ms for image
rendering).
It is still possible to gain the benefits of global image caching and the
lower latency of requesting directly from a CDN, by explicitly changing the
`external_system_avatars_url` site setting to
`https://avatars.discourse.org/letter/{first_letter}/{color}/{size}.png`.
