[feature] Serve 4xx in response to requests during DB restart

[mirabilos]

Describe the bug with a clear and concise description of what the bug is.

During a PostgreSQL restart (e.g. due to a system security upgrade), requests are given a 500 instead of a 4xx code to try again:

timestamp="2024-03-09T20:53:25.124Z" func=server.glob..func1.Logger.func13.1 level=ERROR latency="356.15µs" userAgent="http.rb/5.1.1 (Mastodon/4.3.0-nightly.2024-02-23; +https://mastodon.social/)" method=POST statusCode=500 path=/users/XXX/inbox clientIP=X.X.X.X pubKeyID=https://mastodon.social/users/XXX#main-key errors="Error #01: PostInboxScheme: error authenticating post inbox: AuthenticatePostInbox: could not fetch receiving account XXX: FATAL: terminating connection due to administrator command (SQLSTATE 57P01)\n" requestID=0kqft94e040007tkjmbg msg="Internal Server Error: wrote 54B"

- and -

X.X.X.X - - [09/Mar/2024:20:53:25 +0000] TLSv1.3:TLS_AES_256_GCM_SHA384:IPv4"toot.XXX.org" "POST /users/XXX/inbox HTTP/1.1" 500 54 "-" "http.rb/5.1.1 (Mastodon/4.3.0-nightly.2024-02-23; +https://mastodon.social/)"

What's your GoToSocial Version?

0.13.1-SNAPSHOT+git-4b0eefb 🦥

GoToSocial Arch

amd64 binary

What happened?

No response

What you expected to happen?

The absence of a DB is handled gracefully; clients are given a 4xx error code to make them retry later.

How to reproduce it?

No response

Anything else we need to know?

Yes, I should probably stop GtS before upgrading PostgreSQL. I had not thought of it.

[tsmethurst]

This is a whole thing that would require significant additional code to implement for not much benefit, so I'm not really keen on doing it.

We'd have to consider things like how GtS should know that it's a temporary database issue and not a permanent one, how long GtS should report to clients that they should wait before trying again, etc.

Besides, most remote instance softwares (including GtS) will probably retry later on a 5xx anyways, since Service Unavailable is 503.

I'm gonna close this as "won't implement".

EDIT: I also changed the title to reflect that this is a feature request; it's not really a bug to serve Internal Server Error when the DB has gone missing.

[mirabilos]

tobi dixit:

Besides, most remote instance softwares (including GtS) will probably retry later on a 5xx anyways, since [Service Unavailable](https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/503) is 503.

Ah, okay. Fair.