Some notes of varying degrees of being on topic. So far as caching goes, it isn't what many people are thinking when they think caching. One cannot really share the results of a previously run query. The cache is largely tied to the sqllab tab that ran the query. One could share the download link and anyone with access to the data could download the same to csv until the cache expires, but that is all. The intention, as I understand it, is that caching is just for managing the tab that is running a query.

As for async queries, I get the feeling (through lots of tinkering with caching) that it is enabled, so perhaps we need to up the cache timeout.

This seems to be resolved by using the postgres db in k8s rather than the trove db. I say seems to, because I can run a select sleep(700) when using the internal db, and get the error you describe when trying to use the external db. In the logs "lost connection to mysql" is mentioned when trying to use the external.

Looks like it runs your query in the testing environment:
{F37398759}

There are plans to move back to the db in k8s T342699, T343527 and T343526 which appears to resolve this issue as well.