prha

Should state that the biggest risk here is overwhelming the DB with queries to fetch concurrency information on every iteration of the run coordinator (every 1 second). Any timeouts here...

@gibsondan also, I changed the entry on the dagster run to be a new named tuple called `root_op_concurrency`... the tuple allows us to a layer of abstraction to change the...

I think one big original sin with the run tags join is that the join column is on `run.run_id` instead of `run.id` (primary key). I think probably we'll want to...

I think people are generally on board with applying schema migrations. This in particular would require a data migration to populate a new column to join on, which may take...

Yeah, 100% agree that we should definitely rename `AssetRecord`. In practice, it's something like `AssetLatestInfo` or something like that, since it's generally just getting updated with events or by wipes....

I can construct an event, regardless of whether it's recorded in the system or not. Once the event is recorded with dagster, it's a record. e.g. user code constructs an...

tangential (maybe dumb) question: are materialization/observation events mutually exclusive? I think I remember a time (over a year ago) where they could theoretically be recorded for the same asset key.

I don't love that there's a time connotation with `present`, especially when there are partitions in the mix. Available?

> Is the proposal that if we need to query that information without getting a full record we'll add specialized indexes for them? Specifically: For data version, I think that...

> Is the proposal that if we need to query that information without getting a full record we'll add specialized indexes for them? Yes