Files
duocthu/apps/ai-service/rag/metrics.py
T

98 lines
3.8 KiB
Python

"""Domain counters, defined here so the numbers on a dashboard are the
numbers the domain actually decided.
Kept behind a tiny protocol rather than importing `prometheus_client` into
`rag/`: the domain records that a generation was refused for an ungrounded
number, and the process that happens to expose Prometheus does the exporting.
`NullMetrics` is the default, so tests and any deployment without a metrics
stack run unchanged.
The counter that matters is `generation_rejected` — it is the measured form of
the claim that the answer layer cannot state a figure the book does not.
"""
from __future__ import annotations
from typing import Protocol
class Metrics(Protocol):
def increment(self, name: str, **labels: str) -> None: ...
def observe(self, name: str, value: float, **labels: str) -> None: ...
class NullMetrics:
def increment(self, name: str, **labels: str) -> None: # noqa: ARG002
# Deliberately inert: the default when no metrics stack is configured.
return None
def observe(
self, name: str, value: float, **labels: str # noqa: ARG002
) -> None:
return None
class InMemoryMetrics:
"""Reference implementation of the contract; also what tests assert on."""
def __init__(self) -> None:
self.counts: dict[tuple[str, tuple[tuple[str, str], ...]], int] = {}
self.observations: dict[
tuple[str, tuple[tuple[str, str], ...]], list[float]
] = {}
def increment(self, name: str, **labels: str) -> None:
key = (name, tuple(sorted(labels.items())))
self.counts[key] = self.counts.get(key, 0) + 1
def total(self, name: str, **labels: str) -> int:
if labels:
return self.counts.get((name, tuple(sorted(labels.items()))), 0)
return sum(count for (n, _), count in self.counts.items() if n == name)
def observe(self, name: str, value: float, **labels: str) -> None:
key = (name, tuple(sorted(labels.items())))
self.observations.setdefault(key, []).append(value)
def observed(self, name: str, **labels: str) -> tuple[float, ...]:
if labels:
return tuple(
self.observations.get((name, tuple(sorted(labels.items()))), [])
)
return tuple(
value
for (metric_name, _), values in self.observations.items()
if metric_name == name
for value in values
)
RETRIEVAL_ROUTE = "duocthu_retrieval_route_total"
ABSTENTION = "duocthu_abstention_total"
GENERATION_REJECTED = "duocthu_generation_rejected_total"
GENERATION_SERVED = "duocthu_generation_served_total"
ANSWER_EXTRACTIVE = "duocthu_answer_extractive_total"
# Conversational loop. `CLARIFY_ASKED` is the counter that shows the system
# asking instead of guessing — the behaviour a reviewer will probe first.
CLARIFY_ASKED = "duocthu_clarify_asked_total"
LOOP_ROUNDS = "duocthu_loop_retrieval_rounds_total"
LOOP_REFINED = "duocthu_loop_refined_total"
LOOP_REPAIRED = "duocthu_loop_repaired_total"
FOLLOWUP_INHERITED = "duocthu_followup_inherited_total"
# Trace persistence is fail-open (F-09): a Postgres outage must not turn an
# already-computed, safe answer into a 500. This counts how often that
# degradation actually happens, since a silent fail-open with no counter is
# indistinguishable from tracing quietly working.
TRACE_WRITE_FAILED = "duocthu_trace_write_failed_total"
# HTTP and request-pipeline telemetry. Every label is a bounded vocabulary:
# route is a route template (never a raw path), status is a class (2xx/4xx/5xx),
# and stage/provider/reason values are normalized by the Prometheus adapter.
REQUESTS = "duocthu_requests_total"
REQUEST_DURATION = "duocthu_request_duration_seconds"
STAGE_DURATION = "duocthu_stage_duration_seconds"
DECISION = "duocthu_decision_total"
PROVIDER_FAILURE = "duocthu_provider_failure_total"