💡 Cycle schema hydration is too slow

[rauanmayemir]

I have an idea!

Currently, spiral/roadrunner app bootstrapping is very slow, it takes a pod at least 1 minute to become ready, part of which is cycle fetching information_schema data and setting up its schema. I wish it were faster.

Upon debug dumping queries, I noticed that the data is not cached in any way, so information_schema is queried over and over again.

If I were to connect to a DB that is far from the app (say, there is a 60-100ms latency), then the app wouldn't be able to boot itself at all in reasonable time frame.

[guilhermeaiolfi]

In case I wanted to cache the schema data manually, where would be the best place to do it?

[roxblnfk]

Hi
I prefer to cache the ORM Schema in a PHP file. By using https://github.com/cycle/schema-provider, you can build a pipeline that retrieves the schema from a PHP file, and if it's not available, renders it from attributes and writes it to the cache.

[roxblnfk]

An update on the introspection side of this issue: cycle/database#262 has been merged into 2.x.

Table introspection now batches its catalogue queries: CHECK/DEFAULT constraints and enum ranges are read once per table instead of once per column, and the index query is shared between the index and primary-key lookups. Introspecting a 40-column table drops from dozens of round trips to a handful, so schema sync against a remote database (the 60–100 ms latency scenario above) speeds up roughly by the same factor.

The improvement ships with the 2.22.1 release.