We had a high demand episode of schema modifications and we took long to solve all process.

We notice that the 6 workers were not being enough to process all schema notifications fast enough and there were consuming about 3-4GB of memory each.

We also notice we were setting as RUNNING only 20 schema modifications per time, even after scale to 10 workers. As soon as we improve the memory consumption we should solve this bottle neck.

One ideia also can be discard soon the schema modification that does not affect the BigQuery table, because we might have a lot of requests with no changes. Consider this strategy, but we don’t know if it will help at all.