Files
pikasTech-unidesk/scripts/src/hwlab-node/control-plane-status-observed.ts
T

370 lines
16 KiB
TypeScript

import { readFileSync } from "node:fs";
import { runCommandObserved, type CommandProgress, type CommandResult } from "../command";
import { repoRoot, rootPath } from "../config";
import type { parseNodeScopedDelegatedOptions } from "./plan";
import { record } from "./utils";
export const NODE_RUNTIME_STATUS_WORKER_ENV = "UNIDESK_HWLAB_CONTROL_PLANE_STATUS_WORKER";
const WORKER_PROGRESS_PREFIX = "UNIDESK_HWLAB_CONTROL_PLANE_STATUS_PROGRESS ";
const STATUS_CONFIG_PATH = "config/hwlab-node-control-plane.yaml";
const WORKER_CAPTURE_CHARS = 2 * 1024 * 1024;
const WORKER_STDERR_TAIL_CHARS = 1600;
const WORKER_STDOUT_TAIL_CHARS = 1200;
export const NODE_RUNTIME_STATUS_PROBES = [
"source",
"control-plane",
"pipeline-run",
"argo",
"runtime",
"public",
"git-mirror",
] as const;
export type NodeRuntimeStatusProbe = typeof NODE_RUNTIME_STATUS_PROBES[number];
export type NodeRuntimeStatusProbeState = "started" | "completed" | "failed" | "skipped";
export interface NodeRuntimeStatusWorkerEvent {
probe: NodeRuntimeStatusProbe;
state: NodeRuntimeStatusProbeState;
at: string;
durationMs?: number;
reason?: string;
}
export interface NodeRuntimeStatusPolicy {
configPath: string;
totalBudgetSeconds: number;
heartbeatSeconds: number;
terminationGraceSeconds: number;
probes: Record<NodeRuntimeStatusProbe, boolean>;
}
type ScopedStatus = ReturnType<typeof parseNodeScopedDelegatedOptions>;
export function readNodeRuntimeStatusPolicy(): NodeRuntimeStatusPolicy {
const parsed = requiredRecord(Bun.YAML.parse(readFileSync(rootPath(STATUS_CONFIG_PATH), "utf8")) as unknown, STATUS_CONFIG_PATH);
const status = requiredRecord(parsed.status, `${STATUS_CONFIG_PATH}.status`);
const probesRaw = requiredRecord(status.probes, `${STATUS_CONFIG_PATH}.status.probes`);
const probes = Object.fromEntries(NODE_RUNTIME_STATUS_PROBES.map((probe) => [probe, requiredBoolean(probesRaw[probe], `${STATUS_CONFIG_PATH}.status.probes.${probe}`)])) as Record<NodeRuntimeStatusProbe, boolean>;
if (probes["pipeline-run"] && !probes.source) throw new Error(`${STATUS_CONFIG_PATH}.status.probes.source must be true when pipeline-run is enabled`);
return {
configPath: STATUS_CONFIG_PATH,
totalBudgetSeconds: requiredPositiveInteger(status.totalBudgetSeconds, `${STATUS_CONFIG_PATH}.status.totalBudgetSeconds`),
heartbeatSeconds: requiredPositiveInteger(status.heartbeatSeconds, `${STATUS_CONFIG_PATH}.status.heartbeatSeconds`),
terminationGraceSeconds: requiredPositiveInteger(status.terminationGraceSeconds, `${STATUS_CONFIG_PATH}.status.terminationGraceSeconds`),
probes,
};
}
export function isNodeRuntimeStatusWorker(): boolean {
return process.env[NODE_RUNTIME_STATUS_WORKER_ENV] === "1";
}
export function emitNodeRuntimeStatusWorkerEvent(event: NodeRuntimeStatusWorkerEvent): void {
process.stderr.write(`${WORKER_PROGRESS_PREFIX}${JSON.stringify(event)}\n`);
}
export async function runObservedNodeRuntimeStatus(scoped: ScopedStatus): Promise<{ full: Record<string, unknown>; summary: Record<string, unknown> }> {
const policy = readNodeRuntimeStatusPolicy();
const explicitBudget = scoped.originalArgs.includes("--timeout-seconds");
const budgetSeconds = explicitBudget ? scoped.timeoutSeconds : policy.totalBudgetSeconds;
const startedAtMs = Date.now();
const startedAt = new Date(startedAtMs).toISOString();
const tracker = createProbeTracker(policy);
const abortController = new AbortController();
let interruptSignal: "SIGINT" | "SIGTERM" | null = null;
const interrupt = (signal: "SIGINT" | "SIGTERM") => {
if (interruptSignal !== null) return;
interruptSignal = signal;
abortController.abort();
};
const onSigint = () => interrupt("SIGINT");
const onSigterm = () => interrupt("SIGTERM");
process.on("SIGINT", onSigint);
process.on("SIGTERM", onSigterm);
emitStatusProgress(scoped, {
state: "started",
phase: tracker.currentProbe,
elapsedMs: 0,
budgetSeconds,
heartbeatSeconds: policy.heartbeatSeconds,
configPath: policy.configPath,
});
let stderrBuffer = "";
let commandResult: CommandResult;
try {
commandResult = await runCommandObserved(workerCommand(scoped, budgetSeconds), repoRoot, {
timeoutMs: budgetSeconds * 1000,
heartbeatMs: policy.heartbeatSeconds * 1000,
killAfterMs: policy.terminationGraceSeconds * 1000,
maxCaptureChars: WORKER_CAPTURE_CHARS,
signal: abortController.signal,
env: {
...process.env,
[NODE_RUNTIME_STATUS_WORKER_ENV]: "1",
UNIDESK_CLI_OUTPUT_DUMP_DISABLED: "1",
},
onStderr: (text) => {
stderrBuffer += text;
stderrBuffer = consumeWorkerProgress(stderrBuffer, (event) => {
tracker.accept(event);
emitStatusProgress(scoped, tracker.progress(event.state, budgetSeconds));
});
},
onProgress: (progress) => emitStatusProgress(scoped, tracker.heartbeat(progress, budgetSeconds)),
});
} finally {
process.off("SIGINT", onSigint);
process.off("SIGTERM", onSigterm);
}
const workerPayload = commandResult.timedOut || commandResult.aborted === true || commandResult.stdoutTruncated === true
? null
: parseWorkerPayload(commandResult.stdout);
const closeout = statusCloseout(commandResult, workerPayload);
const observation = tracker.observation({
state: closeout.state,
reason: closeout.reason,
signal: interruptSignal,
startedAt,
startedAtMs,
budgetSeconds,
budgetSource: explicitBudget ? "cli-override" : "yaml",
policy,
commandResult,
});
emitStatusProgress(scoped, {
state: "closeout",
phase: observation.currentPhase,
elapsedMs: observation.durationMs,
reason: observation.reason,
completedProbes: observation.completedProbes,
unfinishedProbes: observation.unfinishedProbes,
processGroupCleaned: observation.processGroup.cleaned,
});
if (workerPayload !== null) {
const summary = { ...record(workerPayload.summary), observation };
return { full: { ...workerPayload, observation, summary }, summary };
}
const summary = partialStatusSummary(scoped, closeout.reason, observation);
return { full: { ...summary, mode: "node-scoped-runtime-status-closeout", summary }, summary };
}
function workerCommand(scoped: ScopedStatus, budgetSeconds: number): string[] {
const args: string[] = [];
for (let index = 0; index < scoped.originalArgs.length; index += 1) {
const arg = scoped.originalArgs[index] ?? "";
if (arg === "--raw" || arg === "--full" || arg === "--json") continue;
if (arg === "--timeout-seconds") {
index += 1;
continue;
}
args.push(arg);
}
return [process.execPath, "scripts/cli.ts", "hwlab", "nodes", "control-plane", ...args, "--timeout-seconds", String(budgetSeconds), "--raw"];
}
function consumeWorkerProgress(buffer: string, onEvent: (event: NodeRuntimeStatusWorkerEvent) => void): string {
let remaining = buffer;
while (true) {
const newline = remaining.indexOf("\n");
if (newline === -1) return remaining.slice(-8192);
const line = remaining.slice(0, newline).trimEnd();
remaining = remaining.slice(newline + 1);
if (!line.startsWith(WORKER_PROGRESS_PREFIX)) continue;
try {
const parsed = JSON.parse(line.slice(WORKER_PROGRESS_PREFIX.length)) as NodeRuntimeStatusWorkerEvent;
if (NODE_RUNTIME_STATUS_PROBES.includes(parsed.probe) && ["started", "completed", "failed", "skipped"].includes(parsed.state)) onEvent(parsed);
} catch {
continue;
}
}
}
function parseWorkerPayload(stdout: string): Record<string, unknown> | null {
try {
const envelope = record(JSON.parse(stdout) as unknown);
const data = record(envelope.data);
return Object.keys(data).length === 0 ? null : data;
} catch {
return null;
}
}
function statusCloseout(result: CommandResult, payload: Record<string, unknown> | null): { state: string; reason: string } {
if (result.processGroupAliveAfterCleanup === true) return { state: "failed", reason: "status-process-group-cleanup-incomplete" };
if (result.timedOut) return { state: "timed-out", reason: "status-total-budget-exceeded" };
if (result.aborted === true) return { state: "interrupted", reason: "status-interrupted" };
if (payload === null) return { state: "failed", reason: "status-worker-failed" };
return { state: "completed", reason: "status-collected" };
}
function partialStatusSummary(scoped: ScopedStatus, reason: string, observation: Record<string, unknown>): Record<string, unknown> {
const status = `bun scripts/cli.ts hwlab nodes control-plane status --node ${scoped.node} --lane ${scoped.lane}`;
return {
ok: false,
command: `hwlab nodes control-plane status --node ${scoped.node} --lane ${scoped.lane}`,
mode: "node-scoped-runtime-status-summary",
mutation: false,
node: scoped.node,
lane: scoped.lane,
sourceCommit: null,
degradedReason: reason,
pipelineRun: {},
argo: {},
runtime: {},
publicProbe: {},
gitMirror: {},
observation,
nextAction: `${status} --json`,
next: { status, json: `${status} --json`, full: `${status} --full`, raw: `${status} --raw` },
};
}
function createProbeTracker(policy: NodeRuntimeStatusPolicy) {
const started = new Map<NodeRuntimeStatusProbe, number>();
const completed: NodeRuntimeStatusProbe[] = [];
const failed: NodeRuntimeStatusProbe[] = [];
const skipped: NodeRuntimeStatusProbe[] = [];
const timings: Record<string, unknown>[] = [];
let currentProbe: string = "starting";
const enabled = NODE_RUNTIME_STATUS_PROBES.filter((probe) => policy.probes[probe]);
const accept = (event: NodeRuntimeStatusWorkerEvent) => {
currentProbe = event.probe;
if (event.state === "started") started.set(event.probe, Date.now());
if (event.state === "completed" || event.state === "failed" || event.state === "skipped") {
const destination = event.state === "completed" ? completed : event.state === "failed" ? failed : skipped;
if (!destination.includes(event.probe)) destination.push(event.probe);
timings.push({ probe: event.probe, state: event.state, durationMs: event.durationMs ?? Math.max(0, Date.now() - (started.get(event.probe) ?? Date.now())), reason: event.reason ?? null });
}
};
const unfinished = () => enabled.filter((probe) => !completed.includes(probe) && !failed.includes(probe) && !skipped.includes(probe));
return {
get currentProbe() { return currentProbe; },
accept,
progress: (state: string, budgetSeconds: number) => ({ state: `probe-${state}`, phase: currentProbe, elapsedMs: null, budgetSeconds, completedProbes: [...completed], unfinishedProbes: unfinished() }),
heartbeat: (progress: CommandProgress, budgetSeconds: number) => ({ state: "heartbeat", phase: currentProbe, elapsedMs: progress.elapsedMs, budgetSeconds, completedProbes: [...completed], unfinishedProbes: unfinished(), stdoutBytes: progress.stdoutBytes, stderrBytes: progress.stderrBytes }),
observation: (input: { state: string; reason: string; signal: string | null; startedAt: string; startedAtMs: number; budgetSeconds: number; budgetSource: string; policy: NodeRuntimeStatusPolicy; commandResult: CommandResult }) => ({
state: input.state,
reason: input.reason,
signal: input.signal,
startedAt: input.startedAt,
finishedAt: new Date().toISOString(),
durationMs: Math.max(0, Date.now() - input.startedAtMs),
currentPhase: currentProbe,
totalBudgetSeconds: input.budgetSeconds,
budgetSource: input.budgetSource,
heartbeatSeconds: input.policy.heartbeatSeconds,
configPath: input.policy.configPath,
completedProbes: [...completed],
failedProbes: [...failed],
skippedProbes: [...skipped],
unfinishedProbes: unfinished(),
phaseTimings: timings.slice(0, NODE_RUNTIME_STATUS_PROBES.length),
worker: {
exitCode: input.commandResult.exitCode,
signal: input.commandResult.signal,
timedOut: input.commandResult.timedOut,
aborted: input.commandResult.aborted === true,
stdoutBytes: input.commandResult.stdoutBytes ?? 0,
stderrBytes: input.commandResult.stderrBytes ?? 0,
stdoutTruncated: input.commandResult.stdoutTruncated === true,
stderrTruncated: input.commandResult.stderrTruncated === true,
...nodeRuntimeStatusWorkerDiagnostics(input.commandResult, input.reason === "status-worker-failed"),
},
processGroup: {
id: input.commandResult.processGroupId ?? null,
isolated: process.platform !== "win32",
terminationReason: input.commandResult.terminationReason ?? null,
cleaned: input.commandResult.terminationReason === null || input.commandResult.processGroupAliveAfterCleanup === null
? null
: input.commandResult.processGroupAliveAfterCleanup === false,
},
}),
};
}
export function runNodeRuntimeStatusProbe<T>(
probe: NodeRuntimeStatusProbe,
policy: NodeRuntimeStatusPolicy,
onProgress: ((event: NodeRuntimeStatusWorkerEvent) => void) | undefined,
run: () => T,
skipped: () => T,
): T {
const startedAt = Date.now();
if (!policy.probes[probe]) {
onProgress?.({ probe, state: "skipped", at: new Date(startedAt).toISOString(), durationMs: 0, reason: "disabled-by-yaml" });
return skipped();
}
onProgress?.({ probe, state: "started", at: new Date(startedAt).toISOString() });
try {
const value = run();
onProgress?.({ probe, state: "completed", at: new Date().toISOString(), durationMs: Date.now() - startedAt });
return value;
} catch (error) {
onProgress?.({ probe, state: "failed", at: new Date().toISOString(), durationMs: Date.now() - startedAt, reason: error instanceof Error ? error.message.slice(0, 300) : String(error).slice(0, 300) });
throw error;
}
}
export function skippedNodeRuntimeStatusCommand(name: string): CommandResult {
return {
command: ["<skipped>", name],
cwd: repoRoot,
exitCode: 0,
stdout: "",
stderr: "",
signal: null,
timedOut: false,
};
}
export function nodeRuntimeStatusWorkerDiagnostics(result: CommandResult, includeStdoutTail: boolean): Record<string, unknown> {
return {
stderrTail: statusDiagnosticTail(result.stderr, WORKER_STDERR_TAIL_CHARS, true),
stdoutTail: includeStdoutTail ? statusDiagnosticTail(result.stdout, WORKER_STDOUT_TAIL_CHARS, false) : null,
valuesRedacted: true,
};
}
function statusDiagnosticTail(text: string, maxChars: number, filterProgress: boolean): string | null {
const filtered = filterProgress
? text.split(/\r?\n/u).filter((line) => !line.trimStart().startsWith(WORKER_PROGRESS_PREFIX)).join("\n")
: text;
const redacted = redactStatusDiagnostic(filtered).trim();
return redacted.length === 0 ? null : redacted.slice(-maxChars);
}
function redactStatusDiagnostic(text: string): string {
return text
.replace(/-----BEGIN [^-\r\n]*PRIVATE KEY-----[\s\S]*?-----END [^-\r\n]*PRIVATE KEY-----/giu, "<private-key:redacted>")
.replace(/((?:postgres(?:ql)?|https?|ssh):\/\/)[^@\s"']+@/giu, "$1<redacted>@")
.replace(/\b(Bearer|Basic)\s+[A-Za-z0-9._~+\/=:-]+/giu, "$1 <redacted>")
.replace(/\b(?:sk-[A-Za-z0-9_-]{16,}|(?:ghp|gho|ghu|ghs|ghr|github_pat)_[A-Za-z0-9_]{16,})\b/gu, "<credential:redacted>")
.replace(/([?&](?:access[_-]?token|api[_-]?key|apikey|authorization|cookie|password|secret|token)=)[^&\s"']+/giu, "$1<redacted>")
.replace(/(["']?(?:access[_-]?token|api[_-]?key|apikey|authorization|cookie|database[_-]?url|jwt[_-]?secret|n8n_encryption_key|passwd|password|private[_-]?key|secret|token)["']?\s*[:=]\s*)(?:"[^"\r\n]*"|'[^'\r\n]*'|[^\s,}\r\n]+)/giu, "$1<redacted>");
}
function emitStatusProgress(scoped: ScopedStatus, detail: Record<string, unknown>): void {
process.stderr.write(`${JSON.stringify({ event: "hwlab.control-plane.status.progress", at: new Date().toISOString(), node: scoped.node, lane: scoped.lane, ...detail })}\n`);
}
function requiredRecord(value: unknown, path: string): Record<string, unknown> {
if (typeof value !== "object" || value === null || Array.isArray(value)) throw new Error(`${path} must be an object`);
return value as Record<string, unknown>;
}
function requiredPositiveInteger(value: unknown, path: string): number {
if (typeof value !== "number" || !Number.isInteger(value) || value <= 0) throw new Error(`${path} must be a positive integer`);
return value;
}
function requiredBoolean(value: unknown, path: string): boolean {
if (typeof value !== "boolean") throw new Error(`${path} must be a boolean`);
return value;
}