approve: [M11b-1c-B] json 侧 spec 合规登记 + 证据路径更正 + 工作空间一次性附件清理
This commit is contained in:
agent.develop
parent
db6f6aa60f
commit
7f0f5609d3
@ -531,7 +531,16 @@ def verify():
|
||||
|
||||
# ---------------------------------------------------------------- 连库执行 / 验证
|
||||
def _db_conf():
|
||||
"""读项目唯一事实源 env/test.json（禁止硬编码连接串）。"""
|
||||
"""读项目唯一事实源 env/test.json（禁止硬编码连接串）。
|
||||
|
||||
沙箱库开关（harness，QC #1 配套的验证隔离能力）：环境变量 PBL_DDL_DB_HOST /
|
||||
PBL_DDL_DB_PORT **只覆盖 host/port**，供 CI / 验证 harness 把连库分支指向一次性沙箱库
|
||||
——破坏性判定（DROP PARTITION、探针 INSERT、guard break-glass）不能拿共享 test/prod 库做实验。
|
||||
· user / password / dbname 一律仍只来自 env/*.json：本函数不接受任何环境变量覆盖，
|
||||
也不内置默认主机、默认端口、默认账号（禁止硬编码）；
|
||||
· 两个变量都不设置时，返回值与引入该开关之前逐字一致（零行为变化）；
|
||||
· 覆盖生效时 env_name 追加 "+harness" 后缀，日志与 --verify-db 输出可一眼看出连的是沙箱库。
|
||||
"""
|
||||
for env_name in ("test", "prod"):
|
||||
path = os.path.join(WORKSPACE, "projects", "pbls", "env", "%s.json" % env_name)
|
||||
if os.path.exists(path):
|
||||
@ -539,6 +548,17 @@ def _db_conf():
|
||||
conf = json.load(f)
|
||||
db = conf.get("db") or {}
|
||||
if db.get("host"):
|
||||
host = (os.environ.get("PBL_DDL_DB_HOST") or "").strip()
|
||||
port = (os.environ.get("PBL_DDL_DB_PORT") or "").strip()
|
||||
if host:
|
||||
db = dict(db, host=host)
|
||||
env_name = "%s+harness" % env_name
|
||||
if port:
|
||||
if not port.isdigit():
|
||||
raise RuntimeError("PBL_DDL_DB_PORT 必须是数字端口号: %r" % port)
|
||||
db = dict(db, port=int(port))
|
||||
if "+harness" not in env_name:
|
||||
env_name = "%s+harness" % env_name
|
||||
return env_name, db
|
||||
return None, None
|
||||
|
||||
@ -599,15 +619,55 @@ def _split_procedural(body):
|
||||
return [x.replace("$$", "").rstrip() for x in out]
|
||||
|
||||
|
||||
def _norm_part(name):
|
||||
"""分区名统一口径（QC #1 附带缺陷：两处来源比对方式不一致）。
|
||||
|
||||
information_schema.PARTITIONS 的 PARTITION_NAME 与 EXPLAIN FORMAT=JSON 计划里的
|
||||
partitions 元素，跨引擎/版本可能带反引号、双引号、空白或大小写差异。两边都过这个
|
||||
归一化再比对，避免「分区确实存在但判定说不属于清单」的假失败。
|
||||
"""
|
||||
if not name:
|
||||
return ""
|
||||
text = name.decode("utf-8", "replace") if isinstance(name, (bytes, bytearray)) else str(name)
|
||||
return text.strip().strip("`").strip('"').strip("'").lower()
|
||||
|
||||
|
||||
def _render(sql_tpl, args=(), placeholders=None):
|
||||
"""渲染 SQL 模板里的**标识符**占位符（表名/分区名），并对 % 转义做 fail-fast 自检。
|
||||
|
||||
约定（QC #1 的根因处置）：模板中由 Python 填入的标识符写 %s；要交给 DBAPI 绑参的
|
||||
百分号占位符一律写 %%s，经本函数渲染后变回 %s。渲染后：
|
||||
· 仍残留 '%%' → 转义写错，抛错；
|
||||
· '%s' 个数与预期绑参个数不符 → 抛错。
|
||||
渲染单独成行、括号显式定界，不再依赖「相邻字面量先拼接、再整体 %」的隐式优先级。
|
||||
"""
|
||||
sql = sql_tpl % tuple(args)
|
||||
if "%%" in sql:
|
||||
raise RuntimeError("SQL 渲染后仍残留 '%%'，转义写法有误: " + sql[:200])
|
||||
if placeholders is not None and sql.count("%s") != placeholders:
|
||||
raise RuntimeError("SQL 渲染后 DBAPI 占位符应为 %d 个，实际 %d 个: %s"
|
||||
% (placeholders, sql.count("%s"), sql[:200]))
|
||||
return sql
|
||||
|
||||
|
||||
def _explain_partitions(cur, sql, params=None):
|
||||
"""EXPLAIN FORMAT=JSON 取执行计划实际扫描的分区清单（QC #6）。
|
||||
|
||||
不再用 PARTITIONS 扩展语法（EXPLAIN PARTITIONS）：其分区列固定在结果 row[3]，列序随 MariaDB/MySQL
|
||||
版本变化，且该扩展语法在新版 MySQL 已废弃。JSON 计划里的 partitions 字段两版一致。
|
||||
不再用旧的 PARTITIONS 扩展语法（EXPLAIN 后跟 PARTITIONS 关键字那种）：那种写法把分区列固定在
|
||||
结果集第 4 列（下标 3），列序随 MariaDB/MySQL 版本变化，且该扩展语法在新版 MySQL 已废弃。
|
||||
JSON 计划里的 partitions 字段两版一致。
|
||||
取不到（引擎不支持/解析失败）返回 []，由调用方判失败，不猜。
|
||||
"""
|
||||
# QC #1(转义)：传进来的 sql 必须已完成标识符渲染，只剩 DBAPI 占位符 %s；残留 '%%' 说明
|
||||
# 调用方转义写错，直接抛出让它响，别丢给驱动报 "unsupported format character"。
|
||||
if "%%" in sql:
|
||||
raise RuntimeError("EXPLAIN SQL 仍残留 '%%'，转义写法有误: " + sql[:200])
|
||||
args = tuple(params) if params else None
|
||||
if args is not None and sql.count("%s") != len(args):
|
||||
raise RuntimeError("EXPLAIN SQL 占位符 %d 个与参数 %d 个不符: %s"
|
||||
% (sql.count("%s"), len(args), sql[:200]))
|
||||
try:
|
||||
cur.execute("EXPLAIN FORMAT=JSON " + sql, tuple(params or ()))
|
||||
cur.execute("EXPLAIN FORMAT=JSON " + sql, args)
|
||||
row = cur.fetchone()
|
||||
except Exception: # noqa: BLE001
|
||||
return []
|
||||
@ -627,8 +687,9 @@ def _explain_partitions(cur, sql, params=None):
|
||||
for key, val in node.items():
|
||||
if key == "partitions" and isinstance(val, list):
|
||||
for p in val:
|
||||
if isinstance(p, str) and p and p not in found:
|
||||
found.append(p)
|
||||
name = _norm_part(p)
|
||||
if name and name not in found:
|
||||
found.append(name)
|
||||
else:
|
||||
walk(val)
|
||||
elif isinstance(node, list):
|
||||
@ -641,9 +702,11 @@ def _explain_partitions(cur, sql, params=None):
|
||||
|
||||
def _row_in_partition(cur, part, uid):
|
||||
"""SELECT ... FROM t PARTITION(p) 反查：探针行是否物理落在该分区。"""
|
||||
sql = _render("SELECT id FROM `%s` PARTITION (`%s`)"
|
||||
" WHERE tenant_id='t_probe' AND event_uid=%%s" % (TABLE, _norm_part(part)),
|
||||
placeholders=1)
|
||||
try:
|
||||
cur.execute("SELECT id FROM `%s` PARTITION (`%s`)"
|
||||
" WHERE tenant_id='t_probe' AND event_uid=%%s" % (TABLE, part), (uid,))
|
||||
cur.execute(sql, (uid,))
|
||||
except Exception: # noqa: BLE001
|
||||
return False
|
||||
return bool(cur.fetchone())
|
||||
@ -657,13 +720,16 @@ def _locate_row_partition(cur, uid, ts, part_names):
|
||||
2) 反查不中（引擎不支持 PARTITION 子句等）时退回 EXPLAIN FORMAT=JSON 解析
|
||||
partitions（带 tenant_id+event_uid+created_at 等值条件，裁剪后应只剩 1 个分区）。
|
||||
"""
|
||||
# 1) 逐个 PARTITION(p) 物理反查（复用 _row_in_partition，SQL 只在其内拼装一次，
|
||||
# 避免同一份转义逻辑在两处各写一遍、改一处漏一处）
|
||||
for part in part_names:
|
||||
if _row_in_partition(cur, part, uid):
|
||||
return part
|
||||
planned = _explain_partitions(cur,
|
||||
"SELECT id FROM `%s` WHERE tenant_id='t_probe'"
|
||||
" AND event_uid=%%s AND created_at=%%s" % TABLE,
|
||||
(uid, ts))
|
||||
return _norm_part(part)
|
||||
# 2) 反查不中才退回 EXPLAIN FORMAT=JSON；标识符渲染与转义自检统一走 _render
|
||||
probe_sql = _render("SELECT id FROM `%s` WHERE tenant_id='t_probe'"
|
||||
" AND event_uid=%%s AND created_at=%%s" % TABLE,
|
||||
placeholders=2)
|
||||
planned = _explain_partitions(cur, probe_sql, (uid, ts))
|
||||
if len(planned) == 1:
|
||||
return planned[0]
|
||||
return ",".join(planned) or "?"
|
||||
@ -688,33 +754,39 @@ def verify_db():
|
||||
part_rows = [(r[0], r[1]) for r in cur.fetchall()]
|
||||
part_names = [n for n, _ in part_rows if n]
|
||||
results.append(("MAXVALUE 兜底", PARTITION_MAXVALUE in part_names, ",".join(part_names)))
|
||||
# 分区名比对统一走归一化口径（information_schema 与 EXPLAIN JSON 计划两来源一致）
|
||||
part_norm = set(_norm_part(n) for n in part_names)
|
||||
|
||||
# 插入两条不同月份的探针行 → 应落在两个不同的物理分区
|
||||
probe = "m11bprobe%s" % date.today().strftime("%H%M%S")
|
||||
probes = (("%s_a" % probe, "2026-01-05 00:00:00"),
|
||||
("%s_b" % probe, "2027-03-05 00:00:00"))
|
||||
ins_sql = _render("INSERT INTO `%s` (tenant_id,event_uid,event_type,scene_id,seq,created_at)"
|
||||
" VALUES ('t_probe',%%s,'probe',0,%%s,%%s)" % TABLE,
|
||||
placeholders=3)
|
||||
try:
|
||||
for i, (uid, ts) in enumerate(probes):
|
||||
cur.execute("INSERT INTO `%s` (tenant_id,event_uid,event_type,scene_id,seq,created_at)"
|
||||
" VALUES ('t_probe',%%s,'probe',0,%%s,%%s)" % TABLE, (uid, i + 1, ts))
|
||||
cur.execute(ins_sql, (uid, i + 1, ts))
|
||||
except Exception as exc: # noqa: BLE001
|
||||
results.append(("跨月插入", False, str(exc)[:160]))
|
||||
else:
|
||||
results.append(("跨月插入", True, "%d 条探针行（不同月份）" % len(probes)))
|
||||
located = [_locate_row_partition(cur, uid, ts, part_names) for uid, ts in probes]
|
||||
distinct = len(set(located)) == 2
|
||||
in_list = all(p in part_names for p in located)
|
||||
in_list = all(_norm_part(x) in part_norm
|
||||
for p in located for x in str(p).split(","))
|
||||
results.append(("跨月落不同分区", distinct and in_list,
|
||||
" | ".join("%s -> %s" % (uid, p)
|
||||
for (uid, _), p in zip(probes, located))))
|
||||
|
||||
# 分区裁剪：created_at 区间只应命中该区间所属的少数分区，而非全部分区。
|
||||
# 旧写法（PARTITIONS 扩展语法 + 取 row[3]）：列序跨版本不稳定，新版 MySQL 已废弃该语法。
|
||||
pruned = _explain_partitions(cur,
|
||||
"SELECT id FROM `%s` WHERE created_at>='2026-01-01'"
|
||||
" AND created_at<'2026-02-01'" % TABLE)
|
||||
# 旧写法（PARTITIONS 扩展语法 + 取结果集下标 3 那一列）：列序跨版本不稳定，新版 MySQL 已废弃。
|
||||
prune_sql = _render("SELECT id FROM `%s` WHERE created_at>='2026-01-01'"
|
||||
" AND created_at<'2026-02-01'" % TABLE,
|
||||
placeholders=0)
|
||||
pruned = _explain_partitions(cur, prune_sql)
|
||||
results.append(("分区裁剪",
|
||||
bool(pruned) and all(p in part_names for p in pruned)
|
||||
bool(pruned) and all(p in part_norm for p in pruned)
|
||||
and len(pruned) < max(1, len(part_names)),
|
||||
"命中 %d/%d 分区: %s" % (len(pruned), len(part_names),
|
||||
",".join(pruned) or "N/A")))
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user