feat(observability): observer-grade telemetry hooks + NeMo-Relay plugin

Adds backend-neutral observer hooks for plugins: session, turn, API
request, tool, approval, and subagent lifecycle events with stable
correlation IDs (session_id, task_id, turn_id, api_request_id,
tool_call_id, parent/child subagent ids). Extends VALID_HOOKS with
api_request_error and subagent_start.

Hot path is zero-cost when no plugin subscribes: has_hook()/presence
checks gate all payload construction, request payloads are returned
by reference when no middleware rewrites, and the sanitized response
payload no longer embeds raw response objects.

Bundles the optional NeMo-Relay observability plugin
(plugins/observability/nemo_relay) as an in-repo consumer of the new
hooks, peer to the existing langfuse plugin. Fails open when the
optional nemo-relay package is not installed.

Authored-by: Bryan Bednarski <bbednarski@nvidia.com>
Salvaged from #29722 onto current main.
This commit is contained in:
Bryan Bednarski
2026-06-03 06:36:46 -07:00
committed by Teknium
parent a78c73f3aa
commit 0d9b7132ff
26 changed files with 3188 additions and 140 deletions
+24
View File
@@ -1146,6 +1146,7 @@ def _build_child_agent(
child._subagent_id = subagent_id
child._parent_subagent_id = parent_subagent_id
child._subagent_goal = goal
child._parent_turn_id = getattr(parent_agent, "_current_turn_id", "") or ""
# Share a credential pool with the child when possible so subagents can
# rotate credentials on rate limits instead of getting pinned to one key.
@@ -1171,6 +1172,21 @@ def _build_child_agent(
except Exception as exc:
logger.debug("spawn_requested relay failed: %s", exc)
try:
from hermes_cli.plugins import invoke_hook as _invoke_hook
_invoke_hook(
"subagent_start",
parent_session_id=getattr(parent_agent, "session_id", None),
parent_turn_id=getattr(parent_agent, "_current_turn_id", "") or "",
parent_subagent_id=parent_subagent_id,
child_session_id=getattr(child, "session_id", None),
child_subagent_id=subagent_id,
child_role=effective_role,
child_goal=goal,
)
except Exception:
logger.debug("subagent_start hook invocation failed", exc_info=True)
return child
@@ -2265,9 +2281,17 @@ def delegate_task(
if _invoke_hook is None:
continue
try:
_child_index = entry.get("task_index", -1)
_child_agent = (
children[_child_index][2]
if isinstance(_child_index, int) and 0 <= _child_index < len(children)
else None
)
_invoke_hook(
"subagent_stop",
parent_session_id=_parent_session_id,
parent_turn_id=getattr(parent_agent, "_current_turn_id", "") or "",
child_session_id=getattr(_child_agent, "session_id", None),
child_role=child_role,
child_summary=entry.get("summary"),
child_status=entry.get("status"),