First, thank you for all the hard work you do on GORM!
I’m encountering an issue when using Preload("Blah.Blah") in combination with prepared statements. In our setup, we deploy multiple instances of our application and perform rolling deployments. When deploying a new version that includes a migration, the migration is applied before starting the new pods, which function as expected.
However, this causes issues for the old pods that are still running and handling traffic. Specifically, if the migration adds a new column to a table, it invalidates the prepared statement caches in the old pods that use SELECT * generated from the Preload() calls on the affected tables. As a result, the old pods break with the error cached plan must not change result type (SQLSTATE 0A000), while the new ones are fine.
Proposal/Question: Would it be possible for Preload(...) to inspect the struct’s fields and generate a query that selects only those fields, instead of using SELECT *? This way, the old pods would continue to have valid query caches while the new pods begin selecting the additional fields introduced by the migration / changes to the structs.
Description
Hi,
First, thank you for all the hard work you do on GORM!
I’m encountering an issue when using
Preload("Blah.Blah")in combination with prepared statements. In our setup, we deploy multiple instances of our application and perform rolling deployments. When deploying a new version that includes a migration, the migration is applied before starting the new pods, which function as expected.However, this causes issues for the old pods that are still running and handling traffic. Specifically, if the migration adds a new column to a table, it invalidates the prepared statement caches in the old pods that use
SELECT *generated from thePreload()calls on the affected tables. As a result, the old pods break with the errorcached plan must not change result type (SQLSTATE 0A000), while the new ones are fine.Proposal/Question: Would it be possible for
Preload(...)to inspect the struct’s fields and generate a query that selects only those fields, instead of usingSELECT *? This way, the old pods would continue to have valid query caches while the new pods begin selecting the additional fields introduced by the migration / changes to the structs.Thanks again!