I am currently testing Flyte and I have a few questions: 1. Adding secrets is mostly done via command line and start with

kubectl create secrets

. The docs have a broken link to "different secret managers" a. I am coming from Airflow and found it quite useful to see which secrets are available and even be able to add/edit them via the UI b. The secrets are in namespaces

<projectId-domain>

which could look like

my-project-development

, but when not using a secret manager this can get very messy to maintain. Also, only people with access to the kubernetes config can add/change secrets, right? 2. When a later task in a workflow fails and I fix it, then I need to re-register the whole workflow which creates a new version which requires running all previous tasks, right? Is there a way to circumvent running earlier (possibly expensive) tasks? a. Use-case: I have a task that handles data and I add a

Resource

requirement. When the task fails due to an OOM error, I bump the requirement, but need to re-register the task.

The docs link should be fixed now,

You don’t need to rerun the entire workflow - why not use caching

1. My assumption is that when I change something in my code of

task2(task1_output)

a new version of the workflow is created and flyte does not recognize that output from

workflow-version-1.task1

can be fed to

workflow-version-2.task2

. Is my assumption wrong? 2. If my assumption is wrong, then caching would help. Since I am using polars (instead of pandas) I would need to implement the hashing function differently (e.g. like this)