Docs feedback: decision_cache is documented as a block but behaves as an attribute

[clayton-cornell]

URL

https://grafana.com/docs/alloy/latest/reference/components/otelcol/otelcol.processor.tail_sampling/#otelcolprocessortail_sampling

Feedback

Slack discussion: https://raintank-corp.slack.com/archives/CSN5HV0CQ/p1727891950877629

According to the documentation, decision_cache is a block, and would be added to the alloy.config file as:

otelcol.processor.tail_sampling "LABEL" {
	decision_cache {
		sampled_cache_size = 100000
		}
...

This configuration generates an error similar to:

327 | decision_cache {
| _____^^^^^^^^^^^^^^^^
328 | | sampled_cache_size = 100000
329 | | }
| |_________^

The configuration works as expected if you add decision_cache formatted as an attribute like this:

	decision_cache = {
		sampled_cache_size = 100000,
		}

And confirmed here:

If this is the correct behaviour, the documentation needs to be updated.

[jdbaldry]

I'm not sure about the "argument" terminology, I think the syntax reference calls these "attributes": https://grafana.com/docs/alloy/latest/get-started/configuration-syntax/syntax/#attributes

[alexwulf]

Adding full error message received for future searchability:

Error: C:\Program Files\GrafanaLabs\Alloy\config.alloy:327:2: "decision_cache" must be an attribute, but is used as a block

327 | decision_cache {
| _____^^^^^^^^^^^^^^^^
328 | | sampled_cache_size = 100000
329 | | }
| |_________^

@jdbaldry is correct - I was saying argument in the slack thread by mistake, but was meaning to say attribute. I corrected this in the title/description.

[mattdurham]

This should have been a block, but its encoded in the config now and we cant change it without breaking backwards compat.

[clayton-cornell]

@mattdurham I've opened a PR to update the docs to reflect how this was implemented in the code. Do we want to change the docs, keeping it as an attribute, or fix decision_cache in the code to promote it to a block.

[mattdurham]

IMO changing the code breaks the compatibility, which we could justify saying it’s a bug but I don’t feel good with that route.

[mattdurham]

Let’s change the docs and I will see if I can think of a new name that is correct and we mark the old one as deprecated ?