they probably return errors as long as the migration is not applied

migrations are guaranteed to run in one transaction and only once

I have inherited a kubernetes deployment that has a deployment specification defining “replace” as deployment strategy, which means all instances are killed and new instances are deployed instead (with migration running in an init container) This doesn allow HA deployments without downtime. My question was if it is safe to run “mixed” set of hydra versions in a cluster — having first instance of updated version to run migration scripts while old instances server workload requests?

But I would recommend testing in a cluster as ymmv

Btw. It seems v1 migrations are broken in the latest v1 release, as it seems to be re-running already applied migrations.

what is broken? do you have logs or an issue?

Okay, so I re-enabled the

ory/hydra:v1

image migrations in the environment we’ve been having trouble with and this is the error we get from

hydra migrate

command:

An error occurred while connecting to SQL: Could not migrate sql schema, applied 0 migrations: pq: column "created_at" of relation "hydra_jwk" already exists handling 2

(there’s a long table before that message, but that is the message that the init container errors out with.)

@steep-lamp-91158 - do you have any feedback on this? What is your proposed solution to this issue?