Start Flow from specific dataset

Tuong-Vi

Hello,

As dss developper, I would like to be able to run a part of the flow from a dataset I can choose (start point).

Actually, I can run one recipie ("build only this dataset") or restart all the flow from the beginning (it can be long sometimes). In the flow, I can select "Build flow outputs reachable from here". It will be useful to have also "Build flow from here until the last dataset".

The top will be the option "force rebuild dependencies with customizable start point / end point ",

Regards

Ashley

Hi @Tuong-Vi
,

I'm not sure you meant was the difference between "build flow outputs reachable from here" and "build flow from here until the last dataset". Could you clarify that?

Best,

Ashley

Note: I'll log the 'top option' you mentioned in which you'd like to be able to build a section of the Flow you've selected. Based on previous discussions we've had around this idea, it looks unlikely that we'll implement it, but I've added your idea to the existing group of requests.

Marlan

Hello @Tuong-Vi
and @AshleyW
. I've wanted to be able to do this as well.

I would describe it as building the datasets that are selected if I right click on a dataset and click "Select all downstream".

Maybe it could be implemented by offering an option to build selected parts of the flow. So one could "Select all downstream" and then "Build selected".

Note that since I typically work with SQL datasets I always use the force rebuild option.

Also I thought that "build flow outputs reachable from here" might do this (i.e., build downstream datasets) but it seems to want to rebuild more than that. At least when I select force rebuild dependencies. If I select build required dependencies then none the downstream SQL datasets are rebuilt. So it'll rebuild way more than I want it to or nothing at all. I don't use it.

It'd be nice to have the requested option in a Scenario step as well. I can work around this by including a bunch of build only this dataset (with force rebuild). So I can accomplish what I need, it's just more difficult to do than it needs to be.

Marlan

Tuong-Vi

Hello,

I'm agree with you @Marlan
, and in a Scenario, this option will be useful too. An another option I would like to see in scenario is action for "all datasets". It will avoid to select dataset sequentially for global action like synchronize hive metastore or build metrics.... The best : check/unckeck dataset in the list (dataset to compute)

Ashley

Hi @Tuong-Vi
,

This idea has been added to our backlog.

Tuong-Vi

Hello, thank you for your attention to this matter,

have a nice day

Turribeach

So not exactly what you ask for but we have come up with a solution to tackle the issue that Scenarios are not re-runable from the latest point of failure. We now design pour scenarios as follows:

1. What you would create as a scenario step we create as a separate scenario
2. We name each of this scenarios with a preffix number (ie 010_Build_Inputs, 020_Calculate_Metrics) leaving gaps so we can insert steps in the middle. This also makes them alpha sorted
3. We then have a "Full Run" scenario that calls/run each of the above scenarios as a step.

This design allows us to quickly resume running a long scenario from the last step that failed without having to enable / disable steps in the Full Run scenario. Using a design like the one I suggested will allow you to start running a flow from any arbitrary point.

tgb417

If it is OK with folks.

I'd like to do a +1 on the idea that from a Scenario one can choose a step type that "Build Flow outputs reachable from here". I have a number of projects that actually fan out at the end rather than coming down to a single dataset or a very small number of end datasets.

In my use case, the project is a data normalization project. It is a project from which we serve cleaned-up "normalized" datasets to provide other projects with useful data to start. Rather than having every project go after the data on its own.

Katie

Hello @Tuong-Vi
!

Thank you for your feedback. Wanted to let you know that this is something our dev team is currently working on

Stay tuned for more updates once it's released!

Katie

tgb417

@ktgross15
,

It is great to hear that there will be some work on this.

Turribeach

Available on v12!

https://doc.dataiku.com/dss/latest/release_notes/12.html#run-subsequent-recipes-and-on-the-fly-schema-propagation

Katie

Hello all!

As promised, we have just released improvements to make schema propagation & building behavior even more intuitive in V12! You can now run downstream & propagate schema from the "run" button within a recipe, or from the flow's "build" button.

See more detail in our reference docs & knowledge base article.

Let me know if you have any questions!

Katie

Tuong-Vi

Thanks a lot @ktgross15
, the teams will save a lot of time