9 comments

[ 0.25 ms ] story [ 5.0 ms ] thread
IN DSCI CI artifacts are easy, every job can create a file inside ~/artifacts directory and all subsequent will see it. If any job remove a file for ~/artifacts all the subsequent jobs won't se it.

So artifacts works as a pipeline data buffer

Very easy. No need for explicit links via YAML/whatever ...

The example contradicts itself in such a short blog post (the example creates and reads ~/artifacts.txt, the prose claims repeatedly artifacts are anything in ~/artifacts/).

This is also an even worse design than path artifacts in Gitlab CI.

Oh. It’s just a typo. Why it’s worse ?
It's a typo that carried over to the second code example of trying to read the file too? Is the sample code completely untested?
the example is fixed, thanks for spotting that
What's difficult about artifacts ?
In the systems I am familiar with one has to explicitly describe jobs artifacts via YAML files as dependencies, for the most cases it’s not needed as all I need is to pass files between jobs. Don’t want explicit syntax for that
This feels like it’s not built for massively parallelized CI steps