Running blend service Multie times per times per 30 minute period

[peterdudfield]

I think currently forecasts init time are unique per 30 minute period. I imagine this is good for speeding up querying.

We currerntly run blend service more than once every 30 mins, after uk-pvnet-app is finished and national_xg is finished. Sometimes if we trigger uk-pvnet-app manually, it would also run multiple times in a 30 min period.

I'm not sure there is a way round this witht he current set up. Perhaps there could be an UpdateForecast route?

[devsjc]

It is just unique, not unique per 30 minute period. Are you trying to write multiple forecasts for the same location with the same energy source and forecaster at the same init time?

[peterdudfield]

yea, potentially, imagine, the blend service runs, but then it runs again because a different forecast finished, or we trigger it manually.

maybe this isnt an issue, but it feels slightly troublesome if we have to wait another 30 mins before a new forecast is saved.

[devsjc]

I'm hesitant to remove the safety to the validity of our data that the uniquess brings. This seems like it's only going to be a probelm if the blend service is run erroneously - it's been triggered too early and all the forecasts haven't written yet. How I see it: f we're making a blend, we should ensure that everything it is blending is completed, present, and correct, prior to running it!

[peterdudfield]

We currently have
uk-pvnet-app triggers blend, and then
national-xg finished and triggers blend. Currently this would cause a problem.

I guess it just restricts us to only blending once every 30 mins. I'm kinda hoping we could not be restricted by that.

But i also understand the pros of having it unqiue. Could we have an option that

• tries to creat forecast
• if init time already exists, we update the forecast
• dumps old forecast to a table (so we dont delete it)

[devsjc]

Why not just run the blend after the national-xg only? (or, more generally, after the final forecaster due to run)

[peterdudfield]

I think now its no longer unique, so its not longer a problem

[devsjc]

I'll close this then.