Currently the Code Insights queue system works based on time. Data points for recent time periods will populate before older time periods, without regard for the series they come from. This makes all insights backfill backwards through time at the same rate. This was a reasonable solution in the past, but the product has outgrown it in the following ways:
Any error causes every insight to stop processing for as long as one series is erroring
New series that are created cause other insights to stop backfilling until that series catches up
In the future it will probably be useful for series to queue up discretely, and follow some simple rules:
Deprioritize a series if it is constantly erroring
Currently the Code Insights queue system works based on time. Data points for recent time periods will populate before older time periods, without regard for the series they come from. This makes all insights backfill backwards through time at the same rate. This was a reasonable solution in the past, but the product has outgrown it in the following ways:
In the future it will probably be useful for series to queue up discretely, and follow some simple rules: