instrument-distributed-tracing
정보
이 스킬은 마이크로서비스의 레이턴시 디버깅과 요청 흐름 추적을 위한 OpenTelemetry 분산 추적 기능을 제공합니다. 컨텍스트 전파, 샘플링 및 Jaeger/Tempo 연동을 지원하며 자동 및 수동 계측을 모두 사용할 수 있습니다. 트레이스를 로그/메트릭과 연관시키거나 성능 병목 현상을 식별하거나 기존 추적 시스템에서 OpenTelemetry로 마이그레이션하는 데 활용하세요.
빠른 설치
Claude Code
추천npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/instrument-distributed-tracingClaude Code에서 이 명령을 복사하여 붙여넣어 스킬을 설치하세요
문서
Instrument Distributed Tracing
OTel tracing → track req across microservices → find perf bottleneck.
Use When
- Latency debug → multi-service distributed sys
- Req flow + deps between microservices
- Slow DB query / external API in txn
- Correlate traces w/ logs+metrics → root cause
- E2E latency (user req → res)
- Legacy tracing (Zipkin, Jaeger) → OTel migration
- SLO compliance → latency percentile
In
- Req: Svc list (lang + framework)
- Req: Backend (Jaeger, Tempo, Zipkin, SaaS)
- Opt: Existing lib (OpenTracing, Zipkin)
- Opt: Sampling strat (%, rate limit)
- Opt: Custom span attrs → biz metadata
Do
See Extended Examples for complete config files.
Step 1: Backend
Deploy Jaeger / Grafana Tempo → receive+store traces.
Opt A: Jaeger all-in-one (dev/test):
# docker-compose.yml
version: '3.8'
services:
jaeger:
image: jaegertracing/all-in-one:1.51
ports:
- "5775:5775/udp" # Zipkin compact thrift
- "6831:6831/udp" # Jaeger compact thrift
- "6832:6832/udp" # Jaeger binary thrift
- "5778:5778" # Serve configs
- "16686:16686" # Jaeger UI
- "14268:14268" # Jaeger HTTP thrift
- "14250:14250" # Jaeger GRPC
- "9411:9411" # Zipkin compatible endpoint
environment:
- COLLECTOR_ZIPKIN_HOST_PORT=:9411
- COLLECTOR_OTLP_ENABLED=true
restart: unless-stopped
Opt B: Grafana Tempo (prod, scalable):
# docker-compose.yml
version: '3.8'
services:
tempo:
image: grafana/tempo:2.3.0
command: ["-config.file=/etc/tempo.yaml"]
volumes:
- ./tempo.yaml:/etc/tempo.yaml
- tempo-data:/tmp/tempo
ports:
- "3200:3200" # Tempo HTTP
- "4317:4317" # OTLP gRPC
- "4318:4318" # OTLP HTTP
- "9411:9411" # Zipkin
restart: unless-stopped
volumes:
tempo-data:
Tempo config (tempo.yaml):
server:
http_listen_port: 3200
distributor:
receivers:
jaeger:
# ... (see EXAMPLES.md)
Prod w/ S3 storage:
storage:
trace:
backend: s3
s3:
bucket: tempo-traces
endpoint: s3.amazonaws.com
region: us-east-1
wal:
path: /tmp/tempo/wal
pool:
max_workers: 100
queue_depth: 10000
→ Backend up, OTLP ready, Jaeger UI / Grafana → "no traces" initially.
If err:
- Ports busy?
netstat -tulpn | grep -E '(4317|16686|3200)' - Container logs:
docker logs jaeger/docker logs tempo - Test OTLP:
curl http://localhost:4318/v1/traces -v - Tempo:
tempo -config.file=/etc/tempo.yaml -verify-config
Step 2: Instrument Apps (Auto)
OTel auto-instrument → common frameworks → min code change.
Python w/ Flask:
pip install opentelemetry-distro opentelemetry-exporter-otlp
opentelemetry-bootstrap -a install
# app.py
from flask import Flask
from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
# ... (see EXAMPLES.md)
Go w/ Gin:
go get go.opentelemetry.io/otel
go get go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracegrpc
go get go.opentelemetry.io/otel/sdk/trace
go get go.opentelemetry.io/contrib/instrumentation/github.com/gin-gonic/gin/otelgin
package main
import (
"context"
"github.com/gin-gonic/gin"
"go.opentelemetry.io/otel"
# ... (see EXAMPLES.md)
Node.js w/ Express:
npm install @opentelemetry/api \
@opentelemetry/sdk-node \
@opentelemetry/auto-instrumentations-node \
@opentelemetry/exporter-trace-otlp-grpc
// tracing.js
const { NodeSDK } = require('@opentelemetry/sdk-node');
const { OTLPTraceExporter } = require('@opentelemetry/exporter-trace-otlp-grpc');
const { getNodeAutoInstrumentations } = require('@opentelemetry/auto-instrumentations-node');
const { Resource } = require('@opentelemetry/resources');
const { SemanticResourceAttributes } = require('@opentelemetry/semantic-conventions');
# ... (see EXAMPLES.md)
→ Traces from svcs appear in Jaeger UI / Grafana. HTTP req → spans auto.
If err:
- Exporter endpoint reachable from app?
- Env vars:
OTEL_EXPORTER_OTLP_ENDPOINT=http://tempo:4317 - Debug log:
OTEL_LOG_LEVEL=debug(Py),OTEL_LOG_LEVEL=DEBUG(Node) - Test w/ simple span → verify export pipeline
- Ver conflicts between OTel pkgs?
Step 3: Manual Instrument
Custom spans → biz logic, DB query, external call.
Py manual spans:
from opentelemetry import trace
tracer = trace.get_tracer(__name__)
def process_order(order_id):
# Create a span for the entire operation
# ... (see EXAMPLES.md)
Go manual spans:
import (
"context"
"go.opentelemetry.io/otel"
"go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/codes"
"go.opentelemetry.io/otel/trace"
# ... (see EXAMPLES.md)
Span attrs best practice:
- Semantic conv:
http.method,http.status_code,db.system,db.statement - Biz ctx:
user.id,order.id,product.category - Resource id:
instance.id,region,availability_zone - Errs:
span.RecordError(err)+span.SetStatus(codes.Error, message) - Events:
span.AddEvent("cache_miss")
→ Custom spans in trace view. Parent-child correct. Attrs visible. Errs highlighted.
If err:
- Ctx propagation → parent span ctx → child?
- Span names descriptive + naming conv?
- Spans ended (
defer span.End()Go,withPy)? - Attr types: str, int, bool, float only
- Semantic conv: standard attr names
Step 4: Ctx Propagation
Trace ctx flows across svc boundary + async ops.
HTTP headers propagation (W3C Trace Context):
# Client side (Python with requests)
import requests
from opentelemetry import trace
from opentelemetry.propagate import inject
tracer = trace.get_tracer(__name__)
# ... (see EXAMPLES.md)
// Server side (Go with Gin)
import (
"go.opentelemetry.io/otel"
"go.opentelemetry.io/otel/propagation"
)
# ... (see EXAMPLES.md)
Msg queue propagation (Kafka):
# Producer
from opentelemetry.propagate import inject
from kafka import KafkaProducer
producer = KafkaProducer(bootstrap_servers=['kafka:9092'])
# ... (see EXAMPLES.md)
# Consumer
from opentelemetry.propagate import extract
def process_message(msg):
# Extract trace context from Kafka headers
headers = {k: v.decode('utf-8') for k, v in msg.headers}
ctx = extract(headers)
# Continue the trace
with tracer.start_as_current_span("process_order_event", context=ctx):
order_id = json.loads(msg.value)['order_id']
handle_order(order_id)
Async ops (Py asyncio):
import asyncio
from opentelemetry import trace, context
async def async_operation():
# Capture current context
token = context.attach(context.get_current())
try:
with tracer.start_as_current_span("async_database_query"):
await asyncio.sleep(0.1) # Simulated async work
return "result"
finally:
context.detach(token)
→ Traces span multi svcs. Trace IDs consistent. Parent-child preserved.
If err:
- W3C propagator set:
otel.propagation.set_global_textmap(TraceContextTextMapPropagator()) - Headers passed in HTTP req?
- Kafka: header support v0.11+
- Debug → log
traceparentvalue - Viz → find broken trace links
Step 5: Sampling
Sampling → reduce trace vol + cost, keep visibility.
Sampling strats:
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.sampling import (
ParentBased,
TraceIdRatioBased,
StaticSampler,
Decision
# ... (see EXAMPLES.md)
Tail-based sampling w/ Tempo:
tempo.yaml:
overrides:
defaults:
metrics_generator:
processors: [service-graphs, span-metrics]
storage:
path: /tmp/tempo/generator/wal
remote_write:
- url: http://prometheus:9090/api/v1/write
send_exemplars: true
# Tail sampling (requires tempo-query)
ingestion_rate_limit_bytes: 5000000
ingestion_burst_size_bytes: 10000000
Grafana Tempo TraceQL → dynamic sampling:
# Sample traces with errors
{ status = error }
# Sample slow traces (>1s)
{ duration > 1s }
# Sample specific services
{ resource.service.name = "checkout-service" }
→ Trace vol → target %. Err traces always sampled. Sampling in span metadata.
If err:
- Sampler applied before tracer provider init
- Sampling decision attr in exported spans?
- Tail sampling:
ingestion_burst_size_bytessufficient buffering - Dropped traces:
otel_traces_dropped_total - Test synth high-vol traffic → validate rate
Step 6: Correlate Traces w/ Metrics+Logs
Link traces → metrics+logs → unified obs.
Add trace IDs → logs (Py):
import logging
from opentelemetry import trace
# Custom log formatter with trace context
class TraceFormatter(logging.Formatter):
def format(self, record):
# ... (see EXAMPLES.md)
Gen metrics from traces (Tempo):
# tempo.yaml
metrics_generator:
registry:
external_labels:
cluster: production
storage:
# ... (see EXAMPLES.md)
Prometheus metrics produced:
traces_service_graph_request_total— req count between svcstraces_span_metrics_duration_seconds— span duration histogramtraces_spanmetrics_calls_total— span call counts
Query traces from metrics (Grafana):
Exemplar support → Prometheus datasource:
datasources:
- name: Prometheus
type: prometheus
url: http://prometheus:9090
jsonData:
exemplarTraceIdDestinations:
- name: trace_id
datasourceName: Tempo
Enable exemplars in dashboard:
{
"fieldConfig": {
"defaults": {
"custom": {
"showExemplars": true
}
}
}
}
→ Click metric exemplar → trace opens. Logs → trace IDs. Traces → logs. Unified debug.
If err:
- Exemplar support enabled Prometheus (v2.26+)
- Trace ID format match (32-char hex)
- Metrics gen enabled Tempo config
- Remote write reachable from Tempo
- Test:
histogram_quantile(0.95, rate(http_request_duration_seconds_bucket[5m])) and on() exemplar
Check
- Backend receives spans from all svcs
- Traces → parent-child correct across svcs
- Span attrs → semantic conv + biz ctx
- Ctx propagates across HTTP+MQ
- Sampling → target %
- Err traces always sampled
- Trace IDs in logs (correct format)
- Grafana → traces via exemplars
- Log panels → trace viewer links
- Retention matches storage policy
Traps
- Ctx not propagated: Forget pass
context→ broken trace. Pass explicit. - Spans never ended: Miss
defer span.End()/with→ mem leak - Over-instrument: Span per fn → trace bloat. Focus svc boundary, DB, external API.
- Miss err record: No
span.RecordError()→ lose debug info. Always record errs. - High-card attrs: Unbounded vals (user ID, bodies) → storage issues. Sample / aggregate.
- Wrong span kind: CLIENT vs SERVER vs INTERNAL → svc graph wrong. Follow semantic conv.
- Sample before ctx: Sampling respects parent. Use
ParentBased→ honor upstream.
→
correlate-observability-signals— unified debug w/ metrics+logs+tracessetup-prometheus-monitoring— gen metrics from traces (Tempo gen)configure-log-aggregation— trace IDs in logsbuild-grafana-dashboards— viz trace-derived metrics + exemplar links
GitHub 저장소
연관 스킬
evaluating-llms-harness
테스팅이 Claude Skill은 MMLU, GSM8K를 포함한 60개 이상의 표준화된 학술 과제에서 LLM 성능을 벤치마크하기 위해 lm-evaluation-harness를 실행합니다. 개발자들이 모델 품질을 비교하고, 학습 진행 상황을 추적하거나 학술 결과를 보고할 수 있도록 설계되었습니다. 이 도구는 HuggingFace와 vLLM 모델을 포함한 다양한 백엔드를 지원합니다.
cloudflare-cron-triggers
테스팅이 스킬은 cron 표현식을 사용하여 Worker를 스케줄링하기 위한 Cloudflare Cron Triggers 구현에 관한 포괄적인 지식을 제공합니다. 주기적 작업, 유지보수 작업, 자동화된 워크플로우 설정 방법을 다루며, 잘못된 cron 표현식이나 시간대 문제 같은 일반적인 이슈들을 해결하는 방법을 포함합니다. 개발자들은 이를 통해 스케줄된 핸들러 구성, cron 트리거 테스트, Workflows 및 Green Compute와의 연동 작업을 수행할 수 있습니다.
webapp-testing
테스팅이 Claude Skill은 Python 스크립트를 통해 로컬 웹 애플리케이션을 테스트하기 위한 Playwright 기반 툴킷을 제공합니다. 프론트엔드 검증, UI 디버깅, 스크린샷 캡처, 로그 확인 기능을 지원하며 서버 라이프사이클을 관리합니다. 브라우저 자동화 작업에 사용하되 컨텍스트 오염을 방지하기 위해 소스 코드를 읽지 않고 스크립트를 직접 실행하세요.
finishing-a-development-branch
테스팅이 스킬은 테스트 통과를 확인한 후 체계적인 통합 옵션을 제시하여 개발자가 완성된 작업을 마무리하도록 돕습니다. 구현이 완료된 후 머지, PR 생성, 브랜치 정리와 같은 워크플로우를 안내합니다. 코드가 준비되고 테스트가 완료되었을 때 개발 프로세스를 체계적으로 마무리하기 위해 사용하세요.
