mattermost table report

This commit is contained in:
kyy
2025-09-03 13:54:08 +09:00
parent 86f6a6f1fe
commit 8185275ba6
5 changed files with 213 additions and 3 deletions

38
run_table.sh Normal file
View File

@@ -0,0 +1,38 @@
#!/usr/bin/env bash
set -a
source .env
set +a
LOG_DIR="logs/table"
mkdir -p "${LOG_DIR}"
ABSOLUTE_RANGE=7d # 24h
TS="$(date +%Y%m%d_%H%M%S)"
LOG_FILE="${LOG_DIR}/report_${ABSOLUTE_RANGE}_${TS}.log"
(cd src && python3 -m report_table \
--range "${ABSOLUTE_RANGE}") \
>> "${LOG_FILE}" 2>&1 &
echo "[OK] Started background job."
echo "[OK] Logging to ${LOG_FILE}"
ABSOLUTE_RANGE=24h # 24h
TS="$(date +%Y%m%d_%H%M%S)"
LOG_FILE="${LOG_DIR}/report_${ABSOLUTE_RANGE}_${TS}.log"
(cd src && python3 -m report_table \
--range "${ABSOLUTE_RANGE}") \
>> "${LOG_FILE}" 2>&1 &
echo "[OK] Started background job."
echo "[OK] Logging to ${LOG_FILE}"
# # crontab -e
# # 매일 09:00 KST에 지난 24시간 보고
# 0 9 * * * /usr/bin/env bash -lc 'cd /opt/monitor && /usr/bin/python3 grafana_dash_pull_and_alert.py --range 24h'
# # 매주 월요일 09:30 KST에 지난 7일 보고
# 30 9 * * 1 /usr/bin/env bash -lc 'cd /opt/monitor && /usr/bin/python3 grafana_dash_pull_and_alert.py --range 7d'

View File

@@ -4,7 +4,7 @@ set -a
source .env source .env
set +a set +a
LOG_DIR="logs" LOG_DIR="logs/text"
mkdir -p "${LOG_DIR}" mkdir -p "${LOG_DIR}"
ABSOLUTE_RANGE=7d # 24h ABSOLUTE_RANGE=7d # 24h
@@ -12,7 +12,7 @@ TS="$(date +%Y%m%d_%H%M%S)"
LOG_FILE="${LOG_DIR}/report_${ABSOLUTE_RANGE}_${TS}.log" LOG_FILE="${LOG_DIR}/report_${ABSOLUTE_RANGE}_${TS}.log"
(cd src && python3 -m main \ (cd src && python3 -m report_text \
--range "${ABSOLUTE_RANGE}") \ --range "${ABSOLUTE_RANGE}") \
>> "${LOG_FILE}" 2>&1 & >> "${LOG_FILE}" 2>&1 &

168
src/report_table.py Normal file
View File

@@ -0,0 +1,168 @@
import argparse
import logging
import re
from datetime import timedelta
from typing import Dict, List, Tuple
from clients.grafana_client import GrafanaClient
from clients.loki import is_loki
from services.dashboard import (
extract_targets,
flatten_panels,
panel_datasource_resolver,
)
from services.reporter import post_mattermost
from services.summarizer import compute_total_for_target
from setting.config import AppConfig
from utils.timeutils import now_kst, to_epoch
logging.basicConfig(level=logging.INFO, format="%(message)s")
logger = logging.getLogger(__name__)
def parse_args():
p = argparse.ArgumentParser()
p.add_argument("--range", choices=["7d", "24h", "1d"], required=True)
return p.parse_args()
# ✅ 추가: compute_total_for_target 가 반환한 문자열에서 total 값만 추출
_TOTAL_RX = re.compile(r"total\s+\*\*([^*]+)\*\*", re.IGNORECASE)
def _parse_total_str(line: str) -> str | None:
"""
' - `legend` (loki) → total **123**' 형태에서 123만 추출.
실패 시 None
"""
if not line:
return None
m = _TOTAL_RX.search(line)
return m.group(1).strip() if m else None
# ✅ 추가: 마크다운 테이블에서 안전하게 보이도록 간단 이스케이프
def _md_escape(s: str) -> str:
return (s or "").replace("|", r"\|").replace("\n", " ")
def main():
args = parse_args()
cfg = AppConfig.load() # env에서 URL, API KEY, 대시보드 UID, 웹훅
gf = GrafanaClient(cfg.grafana_url, cfg.grafana_api_key) # Grafana REST 호출용
# 시간 범위 설정
now = now_kst()
if args.range in ("24h", "1d"):
start = now - timedelta(days=1)
range_label = "지난 24시간"
else:
start = now - timedelta(days=7)
range_label = "지난 7일"
start_epoch, end_epoch = to_epoch(start), to_epoch(now)
# step = step_for_range(end_epoch - start_epoch)
step = 21600 # 6시간
# 대시보드에서 패널 추출
dash = gf.get_dashboard_by_uid(cfg.grafana_dashboard_uid)
panels = flatten_panels(dash)
skipped: List[str] = []
all_ds = gf.list_datasources()
grouped: Dict[str, List[Tuple[str, str]]] = {}
# 모든 패널 순회
for p in panels:
title = p.get("title") or "(제목 없음)"
# 패널에 연결된 데이터소스 해석
ds = panel_datasource_resolver(p, gf.get_datasource_by_uid, lambda: all_ds)
if not ds or ds.get("id") is None:
skipped.append(f"- `{title}` (데이터소스 없음)")
continue
# Loki 데이터소스가 아니면 건너뜀 (Prometheus 패널 제외)
if not is_loki(ds):
continue
# 패널에 정의된 쿼리(target) 추출 (expr + legend)
targets = extract_targets(p)
if not targets:
skipped.append(f"- `{title}` (쿼리 없음)")
continue
# 각 target 쿼리(expr) 순회
for t in targets:
expr = t["expr"]
legend = t["legend"] or ""
try:
# Loki 쿼리를 실행해 total 합계 계산 (문자열)
line = compute_total_for_target(
ds=ds,
ds_id=ds["id"],
legend=legend,
expr=expr,
start_epoch=start_epoch,
end_epoch=end_epoch,
step_sec=step,
# prom_query_range_fn=gf.prom_query_range,
loki_query_range_fn=gf.loki_query_range,
)
if not line:
continue
# ✅ total **...** 파싱해서 테이블 행으로 적재
total_str = _parse_total_str(line)
if total_str is not None:
grouped.setdefault(title, []).append(
(_md_escape(legend or "(no legend)"), total_str)
)
else:
# total 추출 실패한 경우 스킵 목록에 상세 기록
skipped.append(
f"- `{title}` / `{legend}`: total 파싱 실패 → {line}"
)
except Exception as e:
skipped.append(f"- `{title}` / `{legend}`: 오류 - {e}")
# ✅ Mattermost 메시지 본문 구성
lines: List[str] = []
lines.append(
f"**LLM Gateway Unified Monitoring 요약** \n기간: {range_label} "
f"({start.strftime('%Y-%m-%d %H:%M')} ~ {now.strftime('%Y-%m-%d %H:%M')} KST)"
)
lines.append("")
if grouped:
# 발견된 순서를 유지(파이썬 3.7+ dict는 삽입순서 유지)
for panel_title, rows in grouped.items():
if not rows:
continue
lines.append(f"**• {panel_title}**\n")
lines.append("| 타겟(legend) | 합계 |")
lines.append("|:--|--:|")
for legend, total_str in rows:
lines.append(f"| {legend} | {total_str} |")
lines.append("") # 섹션 간 공백
else:
lines.append("_표시할 데이터가 없습니다._")
lines.append("")
if skipped:
lines.append(
"<details><summary>건너뛴 항목</summary>\n\n"
+ "\n".join(skipped)
+ "\n\n</details>"
)
post_mattermost(cfg.mattermost_webhook, lines)
logger.info(f"[OK] Sent grouped tables for {len(grouped)} panels.")
if __name__ == "__main__":
main()

View File

@@ -1,4 +1,5 @@
import argparse import argparse
import logging
from datetime import timedelta from datetime import timedelta
from clients.grafana_client import GrafanaClient from clients.grafana_client import GrafanaClient
@@ -13,6 +14,9 @@ from services.summarizer import compute_total_for_target
from setting.config import AppConfig from setting.config import AppConfig
from utils.timeutils import now_kst, to_epoch from utils.timeutils import now_kst, to_epoch
logging.basicConfig(level=logging.INFO, format="%(message)s")
logger = logging.getLogger(__name__)
def parse_args(): def parse_args():
p = argparse.ArgumentParser() p = argparse.ArgumentParser()
@@ -116,7 +120,7 @@ def main():
) )
post_mattermost(cfg.mattermost_webhook, lines) post_mattermost(cfg.mattermost_webhook, lines)
print(f"[OK] Sent summary for {counted} panels.") logger.info(f"[OK] Sent summary for {counted} panels.")
if __name__ == "__main__": if __name__ == "__main__":