[shivamm-gupta]

Description

Closes #300

Implements structured, queryable logs at run and node granularity under the dynalogs pillar, building on dynavec's opt-in structured logging and secret redaction (#200).

Changes

1. Structured Models & Serialization:

   • Introduced \RunLogRecord\ representing agent/graph run and node execution events with ISO UTC timestamps, correlation identifiers (
     un_id,
     ode_id, \span_id), severity level, messages, contextual metadata dictionaries, and exception tracebacks.
   • Provides lossless dictionary and single-line JSON serialization with automatic secret redaction.

2. Context Propagation & Automatic Correlation:

   • Added
     un_context(run_id=...)\ and
     ode_context(node_id=..., span_id=...)\ context managers leveraging \contextvars\ for async and threaded context isolation.
   • Any log call executed within active contexts automatically inherits and stamps active correlation identifiers without requiring manual propagation.

3. Secret Redaction Enhancements:

   • Extended recursive secret sanitization across authorization headers, Bearer tokens, private keys, API keys, AWS credentials, and embedded exceptions.
   • Guarded AI metrics (\ okens, \ otal_tokens, \prompt_tokens, etc.) so execution token counts remain numeric and unredacted.

4. Pluggable Sinks:

   • \LogSink\ interface with \StdoutSink\ (streaming formatted JSON to stdout/streams), \InMemoryLogSink\ (filtering/inspecting buffered records in memory), and \DynamoDBLogSink\ (persisting partitioned log records into DynamoDB single-table schema).

5. Ergonomic Logging API:

   • Added \RunLogger\ with level filtering, sink dispatching, and standard methods (\debug, \info, \warning, \error, \exception).
   • Exposed \get_run_logger(), \set_default_run_logger(), and \log_run_event()\ helper.

Verification

• Added 10 comprehensive unit tests in \ ests/test_run_logging.py\ covering context propagation, auto-generated IDs, secret redaction, min-level filtering, exception tracebacks, \StdoutSink, \InMemoryLogSink, and buffered/unbuffered \DynamoDBLogSink.
• All 15 logging tests passed offline with 0 AWS dependencies (\pytest tests/test_run_logging.py tests/test_logging.py).
• Formatted and verified clean with
  uff check\ and
  uff format.

[codeforstartups]

Reviewed — strong dynalogs foundation (closes #300). 🎯

RunLogger with pluggable sinks (Stdout / InMemory / DynamoDBLogSink), run_context/node_context contextvar-based correlation (run_id + node_id + span_id), and genuinely careful secret redaction: redacts by key name (secret/password/token/key_id/access_key/bearer/…) and value patterns (AWS secrets, Bearer …, key=value), recursively across dicts/lists/sets — while deliberately preserving token-count metrics (tokens, *_tokens, token_count) so LLM usage numbers survive. That nuance is exactly right.

All 4 CI checks green incl. strict mypy. This is the first dynalogs pillar piece to land — nicely done @shivamm-gupta! Natural next steps on top of this: #299 (span tree) and #301 (run history). Merging.