Do you happen to know an example of this?

This is a similar question I think https://discord.com/channels/778216384475693066/778998585454755870/897056922690256896

If you do settle on a neat implementation we'd really appreciate a PR into the main project

Unfortunately due to priority issues this is not going to happen in a near future, since I am mostly looking into this as a preparation for an issue I know my team will face, but when I do get to make it I can try to tidy it up.

Absolutely - part of the reason we haven't got a 1st party way of doing this is that there are so many varied requirements with this sort of thing

Hey guys! I'm getting a weird error when running a pipeline. For context, I'm running a translation model (that might/or not leverage multithreading), when running it, this is the output..

bash
[1]    99263 segmentation fault  kedro run --pipeline translate
/Users/durc1211/.asdf/installs/python/3.8.9/lib/python3.8/multiprocessing/resource_tracker.py:216: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown        
  warnings.warn('resource_tracker: There appear to be %d '

And the process just stops.. Any ideas? Could not find much online or in the kedro repo..

If the underlying model uses parallelism you can’t use parallel runner with it

Is that something I can configure?

Because I've tried to run it with all of the

--runner <runner-name>

configurations, and it still fails..

Oh in which case it’s likely not a Kedro issue but part of the underlying library - It’s end of day here in London so i can’t help much more tonight. My only advice is to set up breakpoints and step up the debugger to try and work out where / why it fails

@User I'm still a bit convinced that it might be the two libraries playing together... Submitted an issue anywho 🙂 https://github.com/argosopentech/argos-translate/issues/213

I've commented on that issue - it's really bizarre, I can ask one of the devs to look into it

I have a feeling it's something in 3.8. Can you give me some simple instructions to run kedro with 3.9 support? (I know you're gonna release it soon)

kedro 0.17.5 with Python 3.9 unofficially probably works fine actually. The main gotchas are: * if you're using

pandas.ExcelDataSet

then you will need to add

engine: openpyxl

to the

load_args

* you might get some library version conflicts which you'll need to fiddle with a bit Probably the easiest way is to install kedro + its dependencies + your project dependencies and then install Python 3.9 (rather than doing starting out with Python 3.9 first). I just tried this out and managed to get the spaceflights starter running on kedro 0.17.5, Python 3.9 just by fiddling with requirements a bit so it should definitely be possible!

The other thing you can try here is running the nodes/pipeline outside kedro just as pure Python functions. This should help to see whether this is a problem involving the kedro runner at all

Hi Joel, I've been working on a custom dataset to suit my need. To that end, I explored the kedro datasets architecture quite deep and learned about the realease function. I noticed that release will not be called on datasets which are the pipeline's first inputs or last outputs, see https://github.com/quantumblacklabs/kedro/blob/master/kedro/runner/sequential_runner.py#L72 - could you please explain what is preventing the release of those datasets?

I'm not 100% sure - I'm going to ask the team. Initial guess it is a memory optimisation if you need to reuse the data multiple times. I would say that you could define an

after_node_run

hook if you wanted to do this explicitly

What issues are you experience as a result of this?

Reading about the doc on IncrementalDataset here: https://kedro.readthedocs.io/en/stable/05_data/02_kedro_io.html#incremental-dataset-confirm "Partitions that are created externally during the run will also not affect the dataset loads and won’t appear in the list of loaded partitions until the next run or until the release() method is called on the dataset object." So I was curious about this release function and when it was called. The issue is that if I override the release function, it may or may not be called depending on wether the ds is a pipeline's input/output

So I have two reactions: - I'd love to understand your implementation here to see if this is the best way of solving the problem - You can actually define your own runner class

I should also say the runner will be overhauled in a future version it's something we want to rewrite when we get a chance

Thanks! I eventually moved the logic to an after_node_run hook, happy to share the implementation I came up with once I have an MVP

Amazing! Great the hook approach worked, that is our preferred approach for custom logic

Adding parameters in Kedro Pipeline https://stackoverflow.com/questions/69889151/adding-parameters-in-kedro-pipeline

Hi! I'm running a pipeline on an HPC, scheduling jobs with slurm and I would like to run the same pipeline with different inputs. Until now I always used the TemplateConfigLoader to achieve this, changing the variables in the globals.yml, but now I would like to run the pipelines for all the inputs in parallel taking advantage of slurm. The problem I observed is that using the same catalog and changing the globals for a different pipeline run, if the jobs starts at the same time the catalog used during all runs may be the wrong one. Is there a way to use a catalog stored in a different folder passing the path from the cli to be sure that each pipeline is using a different globals.yml? Or how would be the best practice to avoid using the wrongly rendered catalog or the wrong globals.yml? Thanks!

You’re looking for modular pipelines! I’m currently working on revamped docs as we speak but the tutorial is up there

Modular pipelines allow you to reuse the same pipelines as different instances and namespaces with the ability to override certain inputs, outputs and parameters

Hi! Concerning the answer to my last question: I sow your answer suggesting the use of modular pipelines , but in my case I'm using always the same pipeline, just changing the input files. To give more context to the problem: I have a pipeline which I will call PIP. The catalog is rendered using the variable cohort in the globals.yml. I want to run the pipeline using different values for the variables cohort. This works fine as long as I'm running it sequentially like : update globals.yml -> kedro run for each cohort. If I want to run this in parallel the steps update globals -> kedro run run in parallel for all the cohorts. So globals can be overwritten by chance. Therefore I would like to be able to specify the globals.yml in another place or provide the variable from the command line to avoid this problem. Is there any way to achieve it ?

Azure Data Lake Storage Gen2 (ADLS Gen2) as a data source for Kedro pipeline https://stackoverflow.com/questions/69940562/azure-data-lake-storage-gen2-adls-gen2-as-a-data-source-for-kedro-pipeline

SOLVED: I tried to save multiple files to the same folder. Hello all. I'm trying to upgrade Kedro from 0.15.9 all the way to 0.17.5. I'm having issues saving parquet files in S3. I think it has to do with a "S3 metadata eventual consistency" issue. "Caused by: java.io.FileNotFoundException: No such file or directory: s3a://kedrobucket/supply_chain_data_asset_matt/data/03_primary/api/part-00000-09420dd5-677d-421f-9b15-555b2d648c05-c000.snappy.parquet It is possible the underlying files have been updated. You can explicitly invalidate the cache in Spark by running 'REFRESH TABLE tableName' command in SQL or by recreating the Dataset/DataFrame involved." I'm going package by package to figure out what is causing the issue. Anyone else ran into this?

kedro[pandas.CSVDataSet,spark.SparkDataSet]==0.17.5

in in my requirements.in. But

kedro build_reqs

results in

ImportError: cannot import name 'get_installed_distributions'

What is the latest version of pip supported by kedro==0.17.5