fix(anthropic): restore accidentally lost cache tokens attributes

[dinmukhamedm]

This was accidentally removed in #3138 and some subsequent changes. Restoring. More context in #3647.

Note: this PR does NOT fully close #3647, as there will be a new attribute convention soon, see open-telemetry/semantic-conventions#3163

• I have added tests that cover my changes.
• If adding a new instrumentation or changing an existing one, I've added screenshots from some observability platform showing the change.
• PR name follows conventional commits format: feat(instrumentation): ... or fix(instrumentation): ....
• (If applicable) I have updated the documentation accordingly.

---

Important

Restores cache token attributes in Anthropic instrumentation and updates tests to verify correct attribute settings.

• Behavior:
  • Restores cache token attributes in _aset_token_usage() and _set_token_usage() in __init__.py and _set_token_usage() in streaming.py.
  • Updates attribute names from GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS to LLM_USAGE_CACHE_READ_INPUT_TOKENS and GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS to LLM_USAGE_CACHE_CREATION_INPUT_TOKENS.
• Tests:
  • Adds assertions in test_prompt_caching.py to verify cache token attributes are correctly set for both sync and async operations.

^{This description was created by for daa6ff5. You can customize this summary. It will automatically update as commits are pushed.}

Summary by CodeRabbit

• New Features

  • Enhanced token usage metrics to track cache creation and cache read tokens separately in instrumentation spans.

• Tests

  • Added comprehensive test coverage validating cache token tracking across streaming and asynchronous operations.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[ellipsis-dev]

Important

Looks good to me! 👍

Reviewed everything up to daa6ff5 in 11 seconds. Click for details.

• Reviewed 459 lines of code in 3 files
• Skipped 0 files when reviewing.
• Skipped posting 0 draft comments. View those below.
• Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.

Workflow ID: wflow_ZIhOqCTqy5KUUHF1

^{You can customize by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR restores cache-related token usage attributes in Anthropic instrumentation by renaming attribute keys from GEN_AI_USAGE_* to LLM_USAGE_* format, adding span attribute assignments for cache metrics in streaming operations, and adding comprehensive test coverage validating cache token separation between creation and read spans.

Changes

Cohort / File(s)	Summary
Attribute Key Renaming packages/opentelemetry-instrumentation-anthropic/opentelemetry/instrumentation/anthropic/__init__.py	Renamed attribute key constants from GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS and GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS to LLM_USAGE_CACHE_READ_INPUT_TOKENS and LLM_USAGE_CACHE_CREATION_INPUT_TOKENS across two token usage methods.
Streaming Metrics Addition packages/opentelemetry-instrumentation-anthropic/opentelemetry/instrumentation/anthropic/streaming.py	Added span attribute assignments in _set_token_usage to record cache_read_input_tokens and cache_creation_input_tokens using the updated attribute keys.
Test Coverage packages/opentelemetry-instrumentation-anthropic/tests/test_prompt_caching.py	Expanded test assertions to validate cache token separation between creation and read spans across multiple test scenarios (legacy, with-content, with-no-content, stream, and async variants), including exact numeric value verification and read/write behavior assertions.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~15 minutes

Poem

🐰 Tokens cached with care so bright,
Renamed keys in telemetry's light,
Creation reads, reads read again,
Metrics flow through span and den! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: restoring accidentally lost cache token attributes in Anthropic instrumentation.
Linked Issues check	✅ Passed	The PR fully addresses issue #3647 by restoring cache_read_input_tokens and cache_creation_input_tokens attributes across Anthropic instrumentation files with corresponding test coverage.
Out of Scope Changes check	✅ Passed	All changes are scoped to restoring cache token attributes in Anthropic instrumentation; no unrelated modifications detected.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}