[abelonogov-ld]

Summary

Adds global attributes and custom span/log processors to @launchdarkly/observability-react-native. This addresses the customer request for app-wide attributes (e.g. UTM parameters) on all telemetry, and is one of a set of matching PRs across RN, Android, iOS, Flutter, MAUI, Java and Python that share the same semantics (see #858, #859, #860, #861, launchdarkly/swift-launchdarkly-observability#290).

Global attributes

• LDObserve.setAttributes(attributes) / LDObserve.removeAttributes(keys) maintain one process-wide, in-memory attribute map (immutable snapshot rebuilt only on change).
• Applied to every span (manual, fetch/XHR, error, track), log record and metric recorded after the call. Spans take the values current when they start (onStart).
• Precedence: per-call attributes > globals > resource. Globals only fill missing keys.
• highlight.session_id and session.id are ignored (logged in debug mode), as are invalid attribute values.
• Calls before init write straight to the store (not the pre-init call buffer), so telemetry recorded during async init, including automatic fetch spans, already carries them. Nothing is persisted; values survive session changes.

Custom processors (init-time options)

• spanProcessors?: SpanProcessor[] and logProcessors?: LogRecordProcessor[] on ReactNativeOptions.
• Order per provider: global-attributes processor → user processors (in order) → SDK batch processor, so user processors see globals and their changes are exported.
• Every user processor call (including onEnding, forceFlush, shutdown) is wrapped in try/catch, so a failing processor never breaks the SDK or the export.
• SpanProcessor, ReadableSpan, SdkSpan, LogRecordProcessor, SdkLogRecord and ReadableLogRecord are re-exported so customers don't need matching OTel versions.

README documents both APIs.

How did you test this change?

• Unit tests: store (merge, remove, reserved keys, invalid values, snapshot immutability), global-attribute processors (per-call wins, values at span start), safe wrappers (throwing/rejecting processors still export), metric merging for all metric types.
• Integration test through LDObserve and a real ObservabilityClient with stubbed exporters: attributes set before init apply immediately and reach telemetry recorded before init; reserved keys ignored; manual, fetch and error spans get globals with per-call values winning; logs get globals; remove takes effect; user processors see globals and their changes are exported even when another processor throws.
• tsc --noEmit, vitest run (147 tests), prettier check and vite build pass. Removing the global-attributes span processor makes 5 span tests fail.

Are there any deployment considerations?

Additive; no behavior change unless the new APIs or options are used. The global-attributes processor adds a cheap per-span / per-log loop over the snapshot. Global attributes are added to metric data points, so they increase metric cardinality.

---

Note

Overview
Adds process-wide global attributes and optional custom OpenTelemetry processors to @launchdarkly/observability-react-native.

LDObserve.setAttributes / removeAttributes maintain an in-memory map merged onto spans (at start), logs, errors, auto-instrumented fetch spans, and metrics. Per-call attributes win over globals; reserved session keys (highlight.session_id, session.id) are ignored. Calls before async init update the store immediately (not the pre-init call buffer) so early telemetry still picks them up.

Init options spanProcessors and logProcessors run after the built-in global-attribute processors and before export, wrapped so throws cannot block export. Related OTel processor/span/log types are re-exported from the package. README and the legacy example app document and demo UTM-style globals.

^{Reviewed by Cursor Bugbot for commit 135c413. Bugbot is set up for automated code reviews on this repo. Configure here.}

[abelonogov-ld]

Closing in favor of #856, which already implements setAttributes/removeAttributes for React Native. The example-legacy repro from this PR has been added there.