@Nikerabbit, I don't understand the title of the ticket -- "unspecified itmeout error"

But, note that this fragment will not serialize on its own because <ref>.data-mw.body.id refers to a DOM element and the serializer needs to be able to look up what that is.

Sorry I should have been clearer. If I paste the above snippet to http://parsoid-lb.eqiad.wikimedia.org/enwiki/ it takes 30 seconds (I suppose this is some kind of timeout) until it returns Error 503 Service Unavailable.

This is a "minimal" snippet of a full html cxserver is trying to convert to wikitext, failing, and thus publishing of a translation fails.

I would like to see better error messages from parsoid, and preferable faster as well ;) That would make debugging our publishing errors much easier.

See T97206

The varnish req-restart thing is about handling various anomalous error cases in a better way. It doesn't absolve the underlying backends from trying to not do erroneous things in the first place...

@BBlack This is one of those "anomalous error cases". See https://phabricator.wikimedia.org/T97206#1257608

If I paste the above snippet in local parsoid instance, it returns immediately with a descriptive 500 error, as requested. In production however, the back-and-forth in Varnish takes the extra 30s and returns Varnish's 503 error page.

Can you please confirm that this is indeed blocked on that task?

@Nikerabbit, @santhosh, @Arlolra - is this still relevant? Is this in CX or in Parsoid?

I believe Santhosh has investigated and fixed this.