-
[OPIK-6321] [SDK][FE] feat: hide non-essential LangGraph internal spans by default (#7095)
frostbyte_neo 发布于
2026-06-16 10:28:17 +00:00 - [OPIK-6321] [SDK][FE] feat: hide non-essential LangGraph internal spans by default
SDK (langchain integration): tag LangGraph framework-plumbing spans with
metadata.opik = {framework: langgraph, category: internal}. Classification keys
on the langgraph* run metadata: a chain span inside a graph is internal unless its
run name equals langgraph_node (the node/sub-agent boundary). LLM and tool runs are
always kept; runs without langgraph_* metadata (plain LCEL) are out of scope.FE (v2 trace tree): trust the metadata, framework-agnostic. A small decision layer
(spanVisibility) decides hideability via composable rules; hiddenSpans drops hidden
spans and re-parents their children; a persisted per-user toggle and a displayed/total
span count are self-contained components. Generic tree viewer/store/helpers untouched.Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- fix(traces): address PR review on langgraph internal-span hiding
- VisibleSpanCount: count hidden over spans only so the trace root can't undercount displayed
- langchain: coerce a non-dict reserved _opik before tagging internal spans
- hiddenSpans.test: rename cases to the should…when convention
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- test(langchain): cover langgraph span classifier with real-run-driven tests
- drive is_internal_langgraph_run / get_run_metadata off run_dicts captured from a
real no-LLM LangGraph run instead of hand-built fixtures - keep synthetic cases only for markers and malformed inputs real runs can't surface
- integration test asserts the exact set of visible (non-internal) spans, so new
undetected plumbing breaks the test instead of leaking into the UI
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- fix(traces): drop langgraph framework tag from _opik metadata; fix FE lint
- _opik now stores only {category: internal}; the framework name was never read by the
UI (which keys solely on category), keeping the categorization framework-neutral - VisibleSpanCount: wrap useMemo callback to satisfy prettier
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- refactor(traces): use _opik.is_internal flag instead of category=internal
Replace the generic {category: internal} categorization with an explicit
{is_internal: true} flag across the SDK tagging, FE decision layer, and tests.Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
Co-authored-by: Claude Opus 4.8 (1M context) noreply@anthropic.com
下载附件