The human-centric platform for production ML & AI

Outerbounds

Our users have a feature request to update task IDs with some human readable metadata - especially for foreach tasks where they need to hunt for which unit(s) are failing by clicking into task details one-by-one. I think there are a few different ways we can try to achieve this, has anyone looked into this before or have a best practice in mind?

For example imagine if we have a nested `foreach` over `['ae', 'de', 'at]` and `['v1', 'v2']` - they'd rather see a task ID like `ae__v1__t-d7c8c2ea` than `t-d7c8c2ea`

right, there are a few improvements in progress that will make this easier

for now you can open a notebook or a script and find the root cause like this:
```from metaflow import Flow
run = Flow('ForeachFlow').latest_run
for task in run['analyze_data']:
    try:
        if task.stderr in 'KILLED BY ORCHESTRATOR':
            continue
        else:
            print(f'failed task {task.pathspec}\nstderr: {task.stderr}')
    except:
        pass```
i.e. ignore tasks that we killed by the orchestrator due to a sibling task failing