feat: add durable asset synchronization worker

This commit is contained in:
jb
2026-09-19 22:15:23 -04:00
parent 2675ea8114
commit fe7d218195
36 changed files with 2218 additions and 189 deletions
+222
View File
@@ -0,0 +1,222 @@
import { ConvexStore, type RefreshLease } from '../src/lib/server/store';
import { TokenCipher } from '../src/lib/server/security';
import { EveSso, type VerifiedGrant } from '../src/lib/server/sso';
import { EsiClient, EsiError, EsiHttpError, EsiRateLimitError } from '../src/lib/server/esi';
/** Conservative prototype budgets; these are intentionally configurable and bounded. */
export const SYNC_WORKER_DEFAULTS = {
pollMs: 1_000,
leaseMs: 10 * 60_000,
heartbeatMs: 60_000,
maxIdlePolls: 0,
} as const;
export type WorkerConfig = {
store: ConvexStore;
cipher: TokenCipher;
sso: EveSso;
esi: EsiClient;
workerId: string;
pollMs: number;
leaseMs: number;
heartbeatMs: number;
sleep: (ms: number) => Promise<void>;
};
function fixedFailure(): string { return 'Asset synchronization failed'; }
const DEFAULT_SYNC_COOLDOWN_GROUP = 'esi:assets';
function cooldownContext(userId: string, characterId: number): string { return `character:${userId}:${characterId}`; }
export class MutationUncertainError extends Error {
constructor(cause: unknown) { super('Synchronization mutation acknowledgement was uncertain'); this.name = 'MutationUncertainError'; this.cause = cause; }
}
function permanentMutationFailure(error: unknown): boolean {
const message = error instanceof Error ? error.message : String(error);
return /lease is not valid|authorization (?:is )?stale|authorization changed|generation is not current|already leased|Invalid asset chunk|Asset character mismatch|chunks must be staged|chunk already exists|Snapshot is not staging|Snapshot is incomplete|Invalid snapshot|not active|missing|Invalid refreshed scopes/.test(message);
}
async function retryMutation<T>(operation: () => Promise<T>, renew?: () => Promise<boolean>): Promise<T> {
let lastError: unknown;
for (let attempt = 0; attempt < 2; attempt += 1) {
try { return await operation(); }
catch (error) {
lastError = error;
if (permanentMutationFailure(error)) throw error;
if (attempt === 0 && renew) {
try { if (!(await renew())) throw new Error('Synchronization lease is not valid'); }
catch (renewError) {
if (permanentMutationFailure(renewError)) throw renewError;
lastError = renewError;
}
}
}
}
throw new MutationUncertainError(lastError);
}
/** Process one at-least-once delivery. Every write is fenced by the durable lease. */
export async function processSyncJob(config: WorkerConfig): Promise<boolean> {
const claimed = await config.store.claimSyncJob(config.workerId, config.leaseMs);
if (!claimed) return false;
let lostLease = false;
let refreshLease: RefreshLease | null = null;
let refreshAttempted = false;
const heartbeat = setInterval(() => {
void config.store.renewSyncJob(claimed.jobId, claimed.leaseToken, config.workerId, config.leaseMs).then((ok) => {
if (!ok) lostLease = true;
}).catch(() => { lostLease = true; });
}, config.heartbeatMs);
const stopHeartbeat = () => clearInterval(heartbeat);
try {
const refresh = await config.store.acquireRefreshLease(claimed.userId, claimed.characterId, config.workerId);
refreshLease = refresh;
if (!refresh) {
await config.store.deferSyncJob(claimed.jobId, claimed.leaseToken, Date.now() + 1_000);
return true;
}
let previous: VerifiedGrant;
try {
previous = await config.cipher.open<VerifiedGrant>(refresh.encryptedTokens, `character:${claimed.characterId}`);
if (previous.characterId !== claimed.characterId) throw new Error('authorization mismatch');
} catch {
await config.store.failSyncJob(claimed.jobId, claimed.leaseToken).catch(() => {});
return true;
}
const needsRefresh = refresh.tokenExpiresAt <= Date.now() + 60_000;
let grant: VerifiedGrant = previous;
let authorityVersion = { grantRevision: refresh.grantRevision, authorizationEpoch: refresh.authorizationEpoch, credentialVersion: refresh.credentialVersion, scopes: refresh.scopes };
if (!needsRefresh) {
// A lease acquired for a still-valid token is only a serialization guard;
// release it before paging so another job is not blocked for two minutes.
await retryMutation(() => config.store.releaseRefreshLease
? config.store.releaseRefreshLease(claimed.userId, claimed.characterId, refresh.leaseToken, config.workerId)
: Promise.resolve(true));
refreshLease = null;
}
if (needsRefresh) {
try {
const dispatched = await config.store.markRefreshDispatched(claimed.userId, claimed.characterId, refresh.leaseToken, config.workerId);
if (!dispatched) {
await config.store.failSyncJob(claimed.jobId, claimed.leaseToken).catch(() => {});
return true;
}
refreshAttempted = true;
// Exactly one remote refresh attempt. An error is treated as uncertain:
// rotating refresh tokens cannot safely be replayed after a lost result.
grant = await config.sso.refresh(previous.refreshToken);
} catch {
await config.store.markRefreshUncertain(claimed.userId, claimed.characterId, refresh.leaseToken, config.workerId, refresh.expectedVersion).catch(() => {});
return true;
}
if (grant.characterId !== claimed.characterId || grant.ownerHash !== previous.ownerHash) {
await config.store.markRefreshUncertain(claimed.userId, claimed.characterId, refresh.leaseToken, config.workerId, refresh.expectedVersion).catch(() => {});
return true;
}
const committed = await config.store.commitRefresh(
claimed.userId,
claimed.characterId,
refresh.leaseToken,
config.workerId,
refresh.expectedVersion,
{ encryptedTokens: await config.cipher.seal(grant, `character:${claimed.characterId}`), tokenExpiresAt: grant.expiresAt, scopes: grant.scopes, jobId: claimed.jobId, jobLeaseToken: claimed.leaseToken },
);
if (!committed) {
await config.store.markRefreshUncertain(claimed.userId, claimed.characterId, refresh.leaseToken, config.workerId, refresh.expectedVersion).catch(() => {});
await config.store.failSyncJob(claimed.jobId, claimed.leaseToken).catch(() => {});
return true;
}
authorityVersion = committed;
// A refreshed credential cannot publish into a generation fenced to the
// old credential version. Discard only that staging generation.
if (claimed.generation !== undefined) await retryMutation(
() => config.store.restartSyncJobGeneration(claimed.jobId, claimed.leaseToken),
() => config.store.renewSyncJob(claimed.jobId, claimed.leaseToken, config.workerId, config.leaseMs),
);
}
const currentAuthority = await retryMutation(
() => config.store.refreshSyncJobAuthority(claimed.jobId, claimed.leaseToken, authorityVersion),
() => config.store.renewSyncJob(claimed.jobId, claimed.leaseToken, config.workerId, config.leaseMs),
);
if (!grant.scopes.includes('esi-assets.read_assets.v1') || !currentAuthority.scopes.includes('esi-assets.read_assets.v1') || lostLease) {
await config.store.failSyncJob(claimed.jobId, claimed.leaseToken).catch(() => {});
return true;
}
authorityVersion = currentAuthority;
const generation = await retryMutation(
() => config.store.beginSyncJob(claimed.jobId, claimed.leaseToken),
() => config.store.renewSyncJob(claimed.jobId, claimed.leaseToken, config.workerId, config.leaseMs),
);
const context = cooldownContext(claimed.userId, claimed.characterId);
const cooldown = await config.store.cooldown(context, claimed.cooldownGroup);
if (cooldown && cooldown.until > Date.now()) await config.sleep(cooldown.until - Date.now());
const authority = {
userId: claimed.userId,
principal: grant.ownerHash,
characterId: claimed.characterId,
scopes: authorityVersion.scopes,
grantRevision: authorityVersion.grantRevision,
authorizationEpoch: authorityVersion.authorizationEpoch,
credentialVersion: authorityVersion.credentialVersion,
} as const;
const snapshot = await config.esi.getAssets(claimed.characterId, grant.accessToken, authority);
const chunks = Math.ceil(snapshot.assets.length / 250);
for (let index = 0; index < chunks; index += 1) {
if (lostLease) throw new Error('lease lost');
await retryMutation(
() => config.store.stageSyncAssets(claimed.jobId, claimed.leaseToken, index, snapshot.assets.slice(index * 250, (index + 1) * 250)),
() => config.store.renewSyncJob(claimed.jobId, claimed.leaseToken, config.workerId, config.leaseMs),
);
}
await retryMutation(
() => config.store.commitSyncJob(claimed.jobId, claimed.leaseToken, chunks, snapshot.syncedAt, snapshot.expiresAt),
() => config.store.renewSyncJob(claimed.jobId, claimed.leaseToken, config.workerId, config.leaseMs),
);
return true;
} catch (error) {
if (refreshAttempted && refreshLease) await config.store.markRefreshUncertain(claimed.userId, claimed.characterId, refreshLease.leaseToken, config.workerId, refreshLease.expectedVersion).catch(() => {});
if (error instanceof MutationUncertainError) {
await config.store.deferSyncJob(claimed.jobId, claimed.leaseToken, Date.now() + 1_000).catch(() => {});
} else if (error instanceof EsiRateLimitError) {
const retryAt = Math.min(Date.now() + 24 * 60 * 60_000, Math.max(Date.now() + 1_000, error.retryAt));
await config.store.deferSyncJob(claimed.jobId, claimed.leaseToken, retryAt, { context: cooldownContext(claimed.userId, claimed.characterId), until: retryAt, group: error.cooldown.group ?? DEFAULT_SYNC_COOLDOWN_GROUP }).catch(() => {});
} else if (error instanceof EsiError && (error.code === 'network' || error.code === 'deadline' || (error instanceof EsiHttpError && (error.status === 408 || error.status === 420 || error.status === 425 || error.status === 429 || error.status >= 500)))) {
const retryAt = Math.min(Date.now() + 24 * 60 * 60_000, Date.now() + 1_000 * 2 ** Math.max(0, claimed.attempts - 1));
await config.store.deferSyncJob(claimed.jobId, claimed.leaseToken, retryAt).catch(() => {});
} else {
await config.store.failSyncJob(claimed.jobId, claimed.leaseToken).catch(() => {});
}
return true;
} finally {
stopHeartbeat();
}
}
function env(name: string): string {
const value = process.env[name];
if (!value) throw new Error(`Sync worker configuration missing: ${name}`);
return value;
}
export async function runSyncWorker(): Promise<void> {
const workerId = process.env.SYNC_WORKER_ID ?? `sync-worker-${crypto.randomUUID()}`;
const pollMs = Math.max(250, Math.min(30_000, Number(process.env.SYNC_WORKER_POLL_MS ?? SYNC_WORKER_DEFAULTS.pollMs)));
const leaseMs = Math.max(60_000, Math.min(60 * 60_000, Number(process.env.SYNC_WORKER_LEASE_MS ?? SYNC_WORKER_DEFAULTS.leaseMs)));
const heartbeatMs = Math.max(10_000, Math.min(10 * 60_000, Number(process.env.SYNC_WORKER_HEARTBEAT_MS ?? SYNC_WORKER_DEFAULTS.heartbeatMs)));
const store = new ConvexStore(env('CONVEX_URL'), undefined, env('CONVEX_SYNC_WORKER_SECRET'));
const cipher = new TokenCipher(env('TOKEN_ENCRYPTION_KEY'));
const userAgent = env('ESI_USER_AGENT');
const sso = new EveSso({ clientId: env('EVE_CLIENT_ID'), clientSecret: process.env.EVE_CLIENT_SECRET || undefined, redirectUri: process.env.EVE_CALLBACK_URL ?? 'http://localhost/unused-worker-callback', userAgent });
const esi = new EsiClient({ userAgent });
const config: WorkerConfig = { store, cipher, sso, esi, workerId, pollMs, leaseMs, heartbeatMs, sleep: (ms) => new Promise((resolve) => setTimeout(resolve, ms)) };
if (process.env.SYNC_WORKER_STARTUP_CHECK === '1') return;
for (;;) {
const handled = await processSyncJob(config);
if (!handled) await config.sleep(pollMs);
}
}
if (import.meta.main) await runSyncWorker();