Running upgrade-schema (or install schema, but less likely in this case) concurrently yields wrong results. We should add a mechanism that would prevent that (ideally a lock on the tig_schema_versions table that would be taken off after the process finishes)