prepared statement caching: are comments ignored?

[adriangb]

I want to used the new prepared statement functionality but I was wondering if SQLCommenter style comments containing trace information will break this cache. I'm guessing the answer is sadly yes, otherwise the query would have to be minimally parsed.

[JelteF]

If the comments are different, the prepared statement will be considered different. PgBouncer does not parse the query and thus does not remove the comments. So yes it sounds like that would break the cache.

[adriangb]

That makes sense. I imagine "can you please parse queries and remove comments" would be an unreasonable feature request (although in general it makes sense that the cache key would ignore comments).

Would a query_rewrite_regex or more targeted prepared_statement_cache_key_regex be a reasonable feature request?

[JelteF]

Why are you adding these comments to the queries in the first place? Seems like that's the problem here. Can you turn that off instead? Afaict even without pgbouncer in the middle that's going to problematic to get any perf benefit from prepared statements, because every statement will be new anyway.

[adriangb]

That’s a great point. I guess you get one or the other thinking about it from a protocol perspective. I guess we should resolve this.