jigold

I think I addressed the changes in the client, but I'm sure it's not 100% correct yet. Will deal with any bugs later. Can we move on to the implementation?...

> I think I'm seeing more where this approach is coming from, specifically we put batches as they exist today in a special category of having no updates and avoid...

All tables that are written to in `_create_jobs` must be tokenized or the job insertion will be serial. This is the reason for the tokens in the staging tables. I...

Also, I'm not opposed to doing a migration of the existing batches to having rows in the updates table. It will just take some work to write a chunking script...

I'm adding an additional migration PR that will put in the backwards compatibility for the updates table etc.

Sorry for not getting this done quicker. There's two new soon to be PRs in the stack that you can see as commits here: - [Add infrastructure for updates](https://github.com/hail-is/hail/pull/12010/commits/72ff68e628b97bae439d04d4cb45e8508941e8bb) -...

Also, having the migration be online and in multiple stages was making my head hurt a lot. Maybe on Monday we should sit down and white board this and make...

> Mainly that the commit procedure branches on whether the start id is 1 I was trying to avoid an expensive query to update all of the n_pending_parents when we...

I was having trouble figuring out how to handle the token and the attributes in hailtop.batch_client.aioclient.Batch. When we create an update from a Batch that already existed perhaps in a...

Closing in favor of #12199