Namespaced Attributes for Pino and Consola Integrations

[s1gr1d]

Problem

Currently, when logging objects with Pino and Consola, object properties are extracted as top-level attributes, making it unclear where they originate from:

Current behavior:

// Pino
pino.info({ user: 'user-id', something: { more: 3 } });

// Results in:
attributes: { 
 'pino.logger.level': { value: 30, type: 'integer' }, 

 // object keys added as top-level attributes
 user: { value: 'user-id', type: 'string' }, 
 something: { value: '{"more":3}', type: 'string' },
}

Right now, the Consola integration only stringifies the values and doesn't add them as attribute but there is a PR for that: #18602

But while working on this PR, I discovered that adding the object keys as top-level attributes might not work well as this log would result like this in Sentry:

consola.info(
 "Mixed data",
 { userId: 123, nestedMetadata: { id: 789, name: "Jane", source: "api" } },
 new Date("2023-06-15T12:00:00.000Z")
)

Proposed Solution

Use namespaced attributes to make the origin clear:

• consola.args.0, pino.args.0 → for args without keys (arrays, dates, etc.)
• consola.args.userId, pino.args.userId → for object properties
• The body could be Mixed data [object] [date]

This would:

1. Make it clear which attributes come from logged arguments
2. Prevent potential conflicts with system/integration attributes
3. Provide consistent behavior between Pino and Consola integrations

[linear]

JS-1354

[philprime]

(Posting for awareness)

We are having a similar discussion in the sentry-cocoa so we should consider finding a universal approach.

[ckblockit]

Subscribing to this issue.
@philprime Is the plan to update Sentry server-side handling rather than updating individual SDK?
I'm quickly hacking a solution to use dot notation to mimic nested object

[philprime]

Hey @ckblockit, I am not driving this topic so I can't really answer that for you. Maybe @s1gr1d knows what's best to do here.

[s1gr1d]

@ckblockit The plan is to update this (or rather other integrations - see below) in the SDKs. But this would be a breaking change as the attributes are sent differently to Sentry. So we will only release this in a future major version.

[s1gr1d]

Closing this PR as we won't add name-spaced attributes as described above. The way the Pino integration adds attributes is the way we want to add attributes for other logger integrations as well.

In the future, also variadic loggers that allow unlimited arguments (like console or consola) will also send their object attributes as structured top-level attributes to Sentry if the first argument is an object. I'll post another issue with the details on that soon.