Files
electric-cherry/src/lib/server/integrations/plutonium.ts
T

122 lines
7.3 KiB
TypeScript

import { db } from '../db/index.js';
import { config } from '../config.js';
import { getServer, listServers } from '../fleet/servers.js';
import { hasActiveProvisioning } from '../fleet/provisioning.js';
import { latestGameSubmission } from '../fleet/status-evidence.js';
import { latestGameSubmissions } from '../fleet/lifecycle-evidence.js';
import { rconPlutoniumVersion } from './rcon.js';
import { nativeHealth } from '../backends/native.js';
import { restartRecoveryPending } from '../provisioning/index.js';
import { parsePlutoniumManifest, PLUTONIUM_MANIFEST_URL } from '../../plutonium.js';
import type { PlutoniumRelease, PlutoniumVersion } from '../../plutonium.js';
import { readSetting, writeSetting } from '../maintenance/settings.js';
export const HOUR = 60 * 60 * 1000;
const nowIso = () => new Date().toISOString();
export function release(): PlutoniumRelease {
return readSetting('plutonium_release', { revision: null, checkedAt: null, error: null });
}
export function freshRelease(): number | null {
const latest = release();
const age = latest.checkedAt ? Date.now() - Date.parse(latest.checkedAt) : NaN;
return !latest.error && age >= 0 && age < HOUR ? latest.revision : null;
}
export function runningVersions(serverName?: string): PlutoniumVersion[] {
const bindings = serverName === undefined ? [] : [serverName];
const versions = db().prepare(`SELECT * FROM plutonium_versions ${serverName === undefined ? '' : 'WHERE server = ?'}`)
.all(...bindings) as PlutoniumVersion[];
const target = freshRelease();
const byServer = new Map(versions.map(version => [version.server, version]));
const epochs = serverName === undefined ? latestGameSubmissions() : new Map([[serverName, latestGameSubmission(serverName)]]);
const recovering = new Set((db().prepare(`SELECT server FROM restart_rollout WHERE healthy_at IS NULL ${serverName === undefined ? '' : 'AND server = ?'}`)
.all(...bindings) as { server: string }[]).map(row => row.server));
const selected = serverName === undefined ? undefined : getServer(serverName);
const servers = serverName === undefined ? listServers() : selected && !selected.destroyed_at ? [selected] : [];
return servers.map(server => {
const value = byServer.get(server.name)
?? { server: server.name, revision: null, checked_at: null, error: null };
const age = value.checked_at ? Date.now() - Date.parse(value.checked_at) : NaN;
const epoch = epochs.get(server.name);
const stopped = server.desired_state === 'stopped';
const valid = !value.error && age >= 0 && age < HOUR && (!epoch || value.checked_at! > epoch) && !recovering.has(server.name);
const revision = valid && !stopped ? value.revision : null;
return { ...value, revision, stopped,
update_status: stopped ? 'stopped' : revision === null || target === null ? 'unknown'
: revision < target ? 'outdated' : revision > target ? 'ahead' : 'current' };
});
}
export async function probeVersion(server: string, options: { maintenance?: boolean; operationId?: string } = {}): Promise<number | null> {
// Startup verification calls this directly; only maintenance observers yield
// ownership, without replacing provisioning's version evidence with an error.
const suppressed = () => (options.maintenance && hasActiveProvisioning(server)) || (options.operationId !== undefined
&& !db().prepare(`SELECT 1 FROM server_deployments WHERE id=? AND server=? AND server_deleted_at IS NULL
AND action IN ('create','rebuild') AND state='awaiting-health' AND submitted_at IS NOT NULL`).get(options.operationId, server));
// The verifying operation keeps its recovery slot until every readiness
// check passes. Waiting on its own slot would deadlock native rebuilds.
const recoveryBlocked = () => restartRecoveryPending(server) && !(options.operationId
&& db().prepare('SELECT 1 FROM restart_rollout WHERE server=? AND operation_id=? AND healthy_at IS NULL').get(server, options.operationId));
if (suppressed()) return null;
const original = getServer(server);
if (!original) return null;
const epoch = latestGameSubmission(server);
const recovering = recoveryBlocked();
const health = !recovering && original.desired_state !== 'stopped'
? await nativeHealth(server).catch(() => ({ ok: false })) : { ok: false };
if (suppressed()) return null;
let result = !health.ok ? { revision: null, error: 'Waiting for a healthy running server.' }
: await rconPlutoniumVersion(server).catch(() => ({ revision: null, error: 'Version check failed.' }));
if (suppressed()) return null;
const current = getServer(server);
// An older probe must not clear evidence already recorded for a new process
// or replacement registration, including replacements with no submission yet.
if (!current || current.created_at !== original.created_at || epoch !== latestGameSubmission(server)) return null;
const checkedAt = nowIso();
if (recoveryBlocked()) {
result = { revision: null, error: 'Waiting for restart recovery.' };
} else if (current.desired_state === 'stopped') {
result = { revision: null, error: 'Server is intentionally stopped.' };
} else if (epoch && checkedAt <= epoch) {
result = { revision: null, error: 'Waiting for a version observation after startup.' };
}
// Clear the old revision on failure: it is no longer evidence of what is running.
db().prepare(`INSERT INTO plutonium_versions (server, revision, checked_at, error) VALUES (?, ?, ?, ?)
ON CONFLICT(server) DO UPDATE SET revision = excluded.revision, checked_at = excluded.checked_at, error = excluded.error`)
.run(server, result.revision, checkedAt, result.error);
return result.revision;
}
let checking: Promise<void> | null = null;
export function checkPlutonium(): Promise<void> {
if (checking) return checking;
checking = check().finally(() => { checking = null; });
return checking;
}
async function check(): Promise<void> {
await checkRelease();
const servers = listServers();
for (let i = 0; i < servers.length; i += 4) {
await Promise.all(servers.slice(i, i + 4).map(s => probeVersion(s.name, { maintenance: true })));
}
writeSetting('plutonium_versions_checked_at', nowIso());
}
let checkingRelease: Promise<void> | null = null;
/** Refresh the published target without probing unrelated fleet servers. */
export function checkRelease(): Promise<void> {
if (checkingRelease) return checkingRelease;
checkingRelease = fetchRelease().finally(() => { checkingRelease = null; });
return checkingRelease;
}
async function fetchRelease(): Promise<void> {
const previous = release();
try {
const response = await fetch(PLUTONIUM_MANIFEST_URL, { signal: AbortSignal.timeout(config.httpTimeoutMs), redirect: 'error' });
if (!response.ok) throw new Error(`Updater manifest request failed (HTTP ${response.status}).`);
const body = await response.text();
if (body.length > 2_000_000) throw new Error('Updater manifest response was too large.');
writeSetting('plutonium_release', { revision: parsePlutoniumManifest(JSON.parse(body)), checkedAt: nowIso(), error: null });
} catch (err) {
writeSetting('plutonium_release', { ...previous, checkedAt: nowIso(), error: (err as Error).message });
}
}