stacklok / toolhive / 31626671343

70%

Build Type

push

github

Committed by web-flow

Commit Message

Bound client-controlled metric label length (#6279)

* Bound client-controlled metric label length

Cumulative metric readers (Prometheus, OTLP PeriodicReader) keep every
distinct attribute set resident for the process lifetime. The OTEL SDK's
2000-series cap bounds the count of attribute sets but not the byte length
of any single label value. The parsed MCP method, tool, and prompt names
are taken verbatim from the request and bounded only by the 8 MB body
limit, so a single label value can approach 8 MB and stay resident until
the process exits. A small burst of large values is enough to OOM a pod.

Add truncateLabelValue, capping client-controlled metric label values at
128 bytes on a UTF-8 rune boundary with a truncation marker, and apply it
to every client-controlled label site in the telemetry middleware. Span
attributes keep the full value on purpose: spans are sampled and ephemeral
and the OTEL MCP semconv wants the real value.

Part of #6271

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* Assert mcp_resource_id label in truncation test

The integration test verified gen_ai.tool.name and tool were observed
but never confirmed mcp_resource_id was, leaving a false-negative gap:
if that label key were dropped or renamed, the test would pass
silently. Assert it was checked alongside the others.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Coverage Stats

16 of 16 new or added lines in 1 file covered. (100.0%)

81 existing lines in 7 files now uncovered.

84475 of 120233 relevant lines covered (70.26%)

99.09 hits per line