Files
pikasTech-unidesk/scripts/src/platform-infra-sub2api-ops/query.ts
T

409 lines
20 KiB
TypeScript

import type { UniDeskConfig } from "../config";
import { parseJsonOutput } from "../platform-infra-ops-library";
import { runSshCommandCapture } from "../ssh";
import { readSub2ApiConfig } from "../platform-infra/config";
import { resolveTarget } from "../platform-infra/manifest";
import type { Sub2ApiOpsOptions, Sub2ApiOpsTarget } from "./types";
export async function querySub2ApiOps(config: UniDeskConfig, options: Sub2ApiOpsOptions): Promise<Record<string, unknown>> {
const target = resolveOpsTarget(options.targetId);
const result = await runSshCommandCapture(config, target.route, ["sh"], remoteQueryScript(target, options));
const parsed = parseJsonOutput(result.stdout);
if (result.exitCode !== 0 || parsed === null) {
const stderr = result.stderr.trim().slice(-1200);
const stdout = result.stdout.trim().slice(-1200);
return {
ok: false,
sourceStatus: "unavailable",
error: stderr || stdout || "remote Sub2API ops query returned no JSON",
remote: { exitCode: result.exitCode, stdoutTail: stdout, stderrTail: stderr },
target,
valuesPrinted: false,
};
}
return { ...parsed, target, valuesPrinted: false };
}
function resolveOpsTarget(targetId: string): Sub2ApiOpsTarget {
const config = readSub2ApiConfig();
const target = resolveTarget(config, targetId);
return {
id: target.id,
route: target.route,
namespace: target.namespace,
runtimeMode: target.runtimeMode,
hostDockerAppPort: target.hostDocker?.appPort ?? null,
hostDockerEnvPath: target.hostDocker?.envPath ?? null,
appSecretName: config.runtime.database.secretName,
};
}
function pyJson(value: unknown): string {
return `json.loads(${JSON.stringify(JSON.stringify(value))})`;
}
function remoteQueryScript(target: Sub2ApiOpsTarget, options: Sub2ApiOpsOptions): string {
return `
set -u
python3 - <<'PY'
import base64
import json
import subprocess
import sys
from datetime import datetime
from urllib.parse import urlencode
TARGET = ${pyJson(target)}
OPTIONS = ${pyJson(options)}
def run(command, input_bytes=None):
return subprocess.run(command, input=input_bytes, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
def docker(args):
proc = run(["docker", *args])
if proc.returncode == 0:
return proc
sudo_proc = run(["sudo", "-n", "docker", *args])
return sudo_proc if sudo_proc.returncode == 0 else proc
def runtime_logs(since="24h", tail=20000):
if TARGET["runtimeMode"] == "host-docker":
return docker(["logs", f"--since={since}", f"--tail={tail}", "sub2api-app"])
return run(["kubectl", "-n", TARGET["namespace"], "logs", "deployment/sub2api", f"--since={since}", f"--tail={tail}"])
def tail_text(value, limit=800):
if isinstance(value, bytes):
value = value.decode("utf-8", errors="replace")
return str(value)[-limit:]
def kube_json(args, label):
proc = run(["kubectl", *args, "-o", "json"])
if proc.returncode != 0:
raise RuntimeError(f"{label} failed: {tail_text(proc.stderr)}")
return json.loads(proc.stdout.decode("utf-8"))
def read_host_env():
path = TARGET.get("hostDockerEnvPath")
try:
with open(path, "r", encoding="utf-8") as handle:
lines = handle.read().splitlines()
except PermissionError:
proc = run(["sudo", "-n", "cat", path])
if proc.returncode != 0:
raise RuntimeError("read host env failed: " + tail_text(proc.stderr))
lines = proc.stdout.decode("utf-8", errors="replace").splitlines()
values = {}
for line in lines:
text = line.strip()
if not text or text.startswith("#") or "=" not in text:
continue
key, value = text.split("=", 1)
value = value.strip()
if len(value) >= 2 and value[0] == value[-1] and value[0] in ("'", '"'):
value = value[1:-1]
values[key.strip()] = value
return values
def select_app_pod():
if TARGET["runtimeMode"] == "host-docker":
return "sub2api-app"
pods = kube_json(["-n", TARGET["namespace"], "get", "pods", "-l", "app.kubernetes.io/name=sub2api"], "sub2api pods").get("items") or []
running = [item for item in pods if (item.get("status") or {}).get("phase") == "Running"]
if not running:
raise RuntimeError("sub2api app pod not found")
return running[0]["metadata"]["name"]
APP_POD = select_app_pod()
def runtime_version():
if TARGET["runtimeMode"] == "host-docker":
proc = docker(["inspect", "--format", "{{.Config.Image}}", "sub2api-app"])
image = proc.stdout.decode("utf-8", errors="replace").strip() if proc.returncode == 0 else ""
else:
pod = kube_json(["-n", TARGET["namespace"], "get", "pod", APP_POD], "sub2api pod")
containers = ((pod.get("spec") or {}).get("containers") or [])
image = containers[0].get("image") if containers else ""
tag = image.rsplit(":", 1)[1] if isinstance(image, str) and ":" in image else None
return {"image": image or None, "version": tag}
def config_value(key):
if TARGET["runtimeMode"] == "host-docker":
return read_host_env().get(key)
data = kube_json(["-n", TARGET["namespace"], "get", "configmap", "sub2api-config"], "configmap/sub2api-config").get("data") or {}
return data.get(key)
def secret_value(key):
if TARGET["runtimeMode"] == "host-docker":
return read_host_env().get(key)
data = kube_json(["-n", TARGET["namespace"], "get", "secret", TARGET["appSecretName"]], "sub2api secret").get("data") or {}
return base64.b64decode(data[key]).decode("utf-8") if key in data else None
def parse_curl(proc):
output = proc.stdout.decode("utf-8", errors="replace")
marker = "\\n__HTTP_CODE__:"
position = output.rfind(marker)
if position < 0:
return {"ok": False, "httpStatus": 0, "json": None, "body": output, "stderr": tail_text(proc.stderr)}
body = output[:position]
try:
status = int(output[position + len(marker):].strip()[-3:])
except ValueError:
status = 0
try:
parsed = json.loads(body) if body.strip() else None
except json.JSONDecodeError:
parsed = None
return {"ok": proc.returncode == 0 and 200 <= status < 300, "httpStatus": status, "json": parsed, "body": body, "stderr": tail_text(proc.stderr)}
def curl_api(method, path, bearer=None, payload=None):
body = b"" if payload is None else json.dumps(payload, separators=(",", ":")).encode("utf-8")
script = r'''
set -eu
method="$1"
url="$2"
token="\${3:-}"
tmp="$(mktemp)"
trap 'rm -f "$tmp"' EXIT
cat > "$tmp"
if [ -n "$token" ] && [ "$method" = GET ] && [ ! -s "$tmp" ]; then
curl -sS --max-time 20 -w '\n__HTTP_CODE__:%{http_code}' -X "$method" -H "Authorization: Bearer $token" "$url"
elif [ -n "$token" ]; then
curl -sS --max-time 20 -w '\n__HTTP_CODE__:%{http_code}' -X "$method" -H 'Content-Type: application/json' -H "Authorization: Bearer $token" --data-binary @"$tmp" "$url"
elif [ "$method" = GET ] && [ ! -s "$tmp" ]; then
curl -sS --max-time 20 -w '\n__HTTP_CODE__:%{http_code}' -X "$method" "$url"
else
curl -sS --max-time 20 -w '\n__HTTP_CODE__:%{http_code}' -X "$method" -H 'Content-Type: application/json' --data-binary @"$tmp" "$url"
fi
'''
if TARGET["runtimeMode"] == "host-docker":
command = ["sh", "-c", script, "sh", method, f"http://127.0.0.1:{TARGET['hostDockerAppPort']}{path}", bearer or ""]
else:
command = ["kubectl", "-n", TARGET["namespace"], "exec", "-i", APP_POD, "--", "sh", "-c", script, "sh", method, f"http://127.0.0.1:8080{path}", bearer or ""]
return parse_curl(run(command, body))
def data_of(response, label):
parsed = response.get("json")
code = parsed.get("code") if isinstance(parsed, dict) else None
if response.get("ok") is not True or (code is not None and code != 0):
message = parsed.get("message") if isinstance(parsed, dict) else tail_text(response.get("body"), 400)
raise RuntimeError(f"{label} failed: http={response.get('httpStatus')} message={message}")
return parsed.get("data") if isinstance(parsed, dict) and "data" in parsed else parsed
def items_of(data):
if isinstance(data, list):
return data
if isinstance(data, dict) and isinstance(data.get("items"), list):
return data["items"]
return []
def get(token, path, params=None, label="GET"):
suffix = "?" + urlencode(params) if params else ""
return data_of(curl_api("GET", path + suffix, bearer=token), label)
def login():
email = config_value("ADMIN_EMAIL")
password = secret_value("ADMIN_PASSWORD")
if not email or not password:
raise RuntimeError("Sub2API admin credentials are unavailable")
data = data_of(curl_api("POST", "/api/v1/auth/login", payload={"email": email, "password": password}), "admin login")
token = data.get("access_token") or data.get("token") if isinstance(data, dict) else None
if not token:
raise RuntimeError("admin login response has no token")
return token
def find_named(items, selector, label):
if selector is None:
return items
text = str(selector).strip().lower()
numeric = int(text) if text.isdigit() else None
matches = [item for item in items if (numeric is not None and item.get("id") == numeric) or str(item.get("id", "")).strip().lower() == text or str(item.get("name", "")).strip().lower() == text]
if not matches:
available = [f"{item.get('id')}:{item.get('name')}" for item in items]
raise RuntimeError(f"unknown {label} {selector}; available={available}")
return matches
def log_item(line):
json_start = line.find("{")
if json_start < 0:
return None
try:
item = json.loads(line[json_start:])
except Exception:
return None
if not isinstance(item, dict):
return None
prefix = line[:json_start].strip().split()
item["_at"] = prefix[0] if prefix else None
item["_message"] = " ".join(prefix[3:]) if len(prefix) >= 4 else " ".join(prefix[2:])
return item
def epoch_of(value):
if not isinstance(value, str) or not value:
return None
try:
return datetime.fromisoformat(value.replace("Z", "+00:00")).timestamp()
except Exception:
return None
def account_names(token, platform):
params = {"page": 1, "page_size": 500}
if platform:
params["platform"] = platform
rows = items_of(get(token, "/api/v1/admin/accounts", params, "list accounts for monitor correlation"))
return {item.get("id"): item.get("name") for item in rows if isinstance(item, dict) and item.get("id") is not None}
def correlate_record(token, record, monitor):
checked_at = record.get("checked_at")
checked_epoch = epoch_of(checked_at)
latency_ms = record.get("latency_ms")
if checked_epoch is None or not isinstance(latency_ms, (int, float)):
return {"status": "unsupported", "reason": "native record has no usable checked_at/latency_ms", "candidates": []}
expected_start = checked_epoch - latency_ms / 1000.0
record_model = str(record.get("model") or "")
request_rows = items_of(get(token, "/api/v1/admin/ops/requests", {"time_range": "24h", "page": 1, "page_size": 500, "kind": "all"}, "list request correlation candidates"))
request_candidates = []
for item in request_rows:
request_id = item.get("request_id")
created_epoch = epoch_of(item.get("created_at"))
model_match = not record_model or str(item.get("model") or "") == record_model
if not isinstance(request_id, str) or not request_id or created_epoch is None or not model_match:
continue
start_delta_ms = round(abs(created_epoch - expected_start) * 1000)
if start_delta_ms <= 15000:
request_candidates.append((request_id, item, start_delta_ms))
proc = runtime_logs()
names = account_names(token, monitor.get("provider"))
grouped = {}
candidate_ids = {item[0] for item in request_candidates}
for line in proc.stdout.decode("utf-8", errors="replace").splitlines() if proc.returncode == 0 else []:
item = log_item(line)
if item is None:
continue
request_id = item.get("request_id")
at_epoch = epoch_of(item.get("_at"))
if not isinstance(request_id, str) or not request_id:
continue
if request_id not in candidate_ids and (at_epoch is None or at_epoch < expected_start - 15 or at_epoch > checked_epoch + 90):
continue
grouped.setdefault(request_id, []).append(item)
candidates = []
rows_by_id = {item[0]: (item[1], item[2]) for item in request_candidates}
all_request_ids = set(grouped.keys()) | set(rows_by_id.keys())
for request_id in all_request_ids:
events = grouped.get(request_id, [])
events.sort(key=lambda item: epoch_of(item.get("_at")) or 0)
request_row, row_delta_ms = rows_by_id.get(request_id, ({}, None))
first = events[0] if events else None
models = [str(item.get("model")) for item in events if item.get("model")]
first_epoch = epoch_of(first.get("_at")) if first is not None else epoch_of(request_row.get("created_at"))
start_delta_ms = row_delta_ms if row_delta_ms is not None else round(abs((first_epoch or expected_start) - expected_start) * 1000)
model_match = not record_model or record_model in models or str(request_row.get("model") or "") == record_model
if start_delta_ms > 15000 or not model_match:
continue
failovers = [item for item in events if "upstream_failover_switching" in str(item.get("_message") or "")]
select_failures = [item for item in events if "account_select_failed" in str(item.get("_message") or "")]
upstream_errors = [item for item in events if "account_upstream_error" in str(item.get("_message") or "")]
finals = [item for item in events if "http request completed" in str(item.get("_message") or "")]
final = finals[-1] if finals else None
account_ids = []
for item in events:
account_id = item.get("account_id")
if isinstance(account_id, str) and account_id.isdigit():
account_id = int(account_id)
if isinstance(account_id, int) and account_id not in account_ids:
account_ids.append(account_id)
candidates.append({
"requestId": request_id,
"match": {"kind": "native-request-start-window", "startDeltaMs": start_delta_ms, "modelMatched": model_match},
"firstAt": first.get("_at") if first is not None else request_row.get("created_at"),
"lastAt": events[-1].get("_at") if events else request_row.get("created_at"),
"model": next((item for item in models if item), request_row.get("model")),
"apiKeyName": next((item.get("api_key_name") for item in events if item.get("api_key_name")), None),
"accounts": [{"id": account_id, "name": names.get(account_id)} for account_id in account_ids] or ([{"id": request_row.get("account_id"), "name": request_row.get("account_name") or names.get(request_row.get("account_id"))}] if request_row.get("account_id") is not None else []),
"upstreamErrors": [{"at": item.get("_at"), "accountId": item.get("account_id"), "error": item.get("error")} for item in upstream_errors],
"failovers": [{"at": item.get("_at"), "accountId": item.get("account_id"), "upstreamStatus": item.get("upstream_status"), "switchCount": item.get("switch_count"), "maxSwitches": item.get("max_switches")} for item in failovers],
"selectFailures": [{"at": item.get("_at"), "error": item.get("error"), "excludedAccountCount": item.get("excluded_account_count")} for item in select_failures],
"final": {"at": request_row.get("created_at"), "statusCode": request_row.get("status_code") or request_row.get("upstream_status_code"), "accountId": request_row.get("account_id"), "latencyMs": request_row.get("duration_ms"), "path": request_row.get("path")} if final is None else {"at": final.get("_at"), "statusCode": final.get("status_code"), "accountId": final.get("account_id"), "latencyMs": final.get("latency_ms"), "path": final.get("path")},
})
candidates.sort(key=lambda item: item["match"]["startDeltaMs"])
if not candidates:
return {"status": "unsupported", "reason": "no unique native request_id is available in the record; bounded log correlation found no candidate", "candidates": []}
best_delta = candidates[0]["match"]["startDeltaMs"]
best = [item for item in candidates if item["match"]["startDeltaMs"] == best_delta]
return {"status": "inferred" if len(best) == 1 else "ambiguous", "reason": "heuristic correlation by record start window and model; the native monitor record has no request_id" if len(best) == 1 else "multiple requests have the same nearest start time", "selected": best[0] if len(best) == 1 else None, "candidates": candidates[:5]}
def diagnosis(token):
group_params = {"platform": OPTIONS["platform"]} if OPTIONS.get("platform") else None
groups = items_of(get(token, "/api/v1/admin/groups/all", group_params, "list groups"))
groups = find_named(groups, OPTIONS.get("group"), "group")
snapshots = []
selected = groups if groups else [{"id": None, "name": "all", "platform": OPTIONS.get("platform") or "all"}]
for group in selected:
params = {"time_range": OPTIONS["timeRange"]}
platform = OPTIONS.get("platform") or group.get("platform")
if platform and platform != "all":
params["platform"] = platform
if group.get("id"):
params["group_id"] = group["id"]
overview = get(token, "/api/v1/admin/ops/dashboard/overview", params, "ops dashboard overview")
snapshots.append({"group": {"id": group.get("id"), "name": group.get("name"), "platform": platform or "all"}, "overview": overview})
evidence = None
diagnosis_id = OPTIONS.get("diagnosisId")
if diagnosis_id:
scope, _, category = diagnosis_id.partition(":")
group = next((item["group"] for item in snapshots if ("g" + str(item["group"].get("id"))) == scope or scope == "all"), None)
if group is None:
raise RuntimeError(f"diagnosis id scope not found: {diagnosis_id}")
params = {"time_range": OPTIONS["timeRange"], "page": 1, "page_size": 20}
if group.get("id"):
params["group_id"] = group["id"]
if group.get("platform") and group.get("platform") != "all":
params["platform"] = group["platform"]
if category.startswith("upstream-"):
endpoint = "/api/v1/admin/ops/upstream-errors"
elif category.startswith("error-") or category.startswith("sla-"):
endpoint = "/api/v1/admin/ops/request-errors"
else:
endpoint = "/api/v1/admin/ops/requests"
params["kind"] = "all"
params["sort"] = "duration_desc"
evidence = {"diagnosisId": diagnosis_id, "records": items_of(get(token, endpoint, params, "diagnosis evidence"))}
return {"groups": snapshots, "evidence": evidence}
def channels(token):
monitors = items_of(get(token, "/api/v1/channel-monitors", None, "list channel monitors"))
if OPTIONS.get("platform"):
monitors = [item for item in monitors if str(item.get("provider", "")).lower() == OPTIONS["platform"].lower()]
monitors = find_named(monitors, OPTIONS.get("channel"), "channel")
output = []
for monitor in monitors:
detail = get(token, f"/api/v1/channel-monitors/{monitor['id']}/status", None, "channel monitor status")
output.append({"summary": monitor, "detail": detail})
history = None
correlation = None
if OPTIONS.get("channel") and output:
monitor_id = output[0]["summary"]["id"]
params = {"limit": 1000}
if OPTIONS.get("model"):
params["model"] = OPTIONS["model"]
history = items_of(get(token, f"/api/v1/admin/channel-monitors/{monitor_id}/history", params, "channel monitor history"))
record_id = OPTIONS.get("recordId")
if record_id:
selected = next((item for item in history if str(item.get("id")) == str(record_id)), None)
if selected is not None:
correlation = correlate_record(token, selected, output[0]["summary"])
return {"channels": output, "history": history, "correlation": correlation}
try:
token = login()
data = diagnosis(token) if OPTIONS["action"] == "diagnosis" else channels(token)
data["runtimeVersion"] = runtime_version()
print(json.dumps({"ok": True, "sourceStatus": "available", "data": data}, ensure_ascii=False, separators=(",", ":")))
except Exception as error:
print(json.dumps({"ok": False, "sourceStatus": "unavailable", "error": str(error), "data": None}, ensure_ascii=False, separators=(",", ":")))
sys.exit(1)
PY
`;
}