hail
hail copied to clipboard
Cloud-native genomic dataframes and batch computing
### What happened? The Auth service in GCP uses a user's google email as their unique identifier (`login_id`). This is a suboptimal form of user identification because emails can change...
### What happened? #13131 Adds the ability for users to authenticate with the hail service using access tokens from their hail identity provider (GCP IAM or Azure AAD) instead of...
### What happened? Hail's google/azure credential classes do not require the caller to specify scopes when requesting access tokens, and thus default to a [very wide set of scopes](https://github.com/hail-is/hail/blob/91f5a0bfc30927014b60b11a353a4d95db009427/hail/python/hailtop/aiocloud/aiogoogle/credentials.py#L140), making...
### What happened? Upon worker shutdown, we warn errors if there are any unclosed asyncio Tasks before the process exits. Lately, the worker logs have been noisy with such warnings....
After discussing with @ehigham, we decided to try a lighter-weight RFC process, where an RFC is simply a PR for a new dev doc. Discussion about the design can happen...
### What happened? Consider making the following changes to the names of tables and columns now that job groups are in. This will require a shutdown of Batch during the...
### What happened? Remove the `cancel_batch` stored procedure from the database now that the Batch server no longer uses that function. This migration can be online. Make sure that the...
### What happened? The current query code uses `EXISTS` and `IN` to check whether a batch / job / job group has a particular attribute. This is extremely slow especially...
### What happened? In the new job groups code, we had to add this complicated SQL code all over the place to check whether a job group is cancelled. I...