Race condition in timeseries creation

[Lun4m]

If a timeseries is not yet present in LARD, two threads separately ingesting data for that timeseries can create two different timeseries IDs. Discovered from #46.

[intarga]

Perhaps a unique constraint on the native labels could solve this?

[Lun4m]

On the MET labels? In principle, that's what makes the most sense to me, but can we have a deactivated and an active timeseries sharing the same label? Also, this would disallow the current approach we have taken with migrations and ts reconciliation?

Edit: another problem is that some param codes don't have an associated param id, so it might not work for those. I'm still not sure how to handle them properly.

[intarga]

I meant only on the native labels (i.e labels.obsinn). We need duplicates on the met label for the reconciliation strategy, but each of those should have different native labels

[Lun4m]

Ah yeah, we are also inserting into that table, that should work 👌