@optimystic/db-p2p 0.2.3 → 0.7.0

package/src/cluster/block-transfer-service.ts

@@ -0,0 +1,231 @@
+import type { Startable, Stream } from '@libp2p/interface';
+import type { IRepo, PeerId, IPeerNetwork } from '@optimystic/db-core';
+import { pipe } from 'it-pipe';
+import * as lp from 'it-length-prefixed';
+import { fromString as u8FromString } from 'uint8arrays/from-string';
+import { toString as u8ToString } from 'uint8arrays/to-string';
+import { ProtocolClient } from '../protocol-client.js';
+import { createLogger } from '../logger.js';
+
+const log = createLogger('block-transfer-service');
+
+/** Protocol path */
+const BLOCK_TRANSFER_PREFIX = '/db-p2p/block-transfer/';
+const BLOCK_TRANSFER_VERSION = '1.0.0';
+
+export const buildBlockTransferProtocol = (protocolPrefix: string = ''): string =>
+	`${protocolPrefix}${BLOCK_TRANSFER_PREFIX}${BLOCK_TRANSFER_VERSION}`;
+
+/** Request to transfer blocks */
+export interface BlockTransferRequest {
+	type: 'pull' | 'push';
+	/** Block IDs being transferred */
+	blockIds: string[];
+	/** Reason for transfer */
+	reason: 'rebalance' | 'replication' | 'recovery';
+	/** For push: base64-encoded block data per block ID */
+	blockData?: Record<string, string>;
+}
+
+/** Response with block data */
+export interface BlockTransferResponse {
+	/** Blocks successfully transferred: blockId → base64-encoded data */
+	blocks: Record<string, string>;
+	/** Block IDs that couldn't be found/transferred */
+	missing: string[];
+}
+
+// --- Service (server-side handler) ---
+
+export interface BlockTransferServiceInit {
+	protocolPrefix?: string;
+}
+
+export interface BlockTransferServiceComponents {
+	registrar: { handle: (...args: any[]) => Promise<void>; unhandle: (...args: any[]) => Promise<void> };
+	repo: IRepo;
+}
+
+/**
+ * Libp2p service that handles incoming block transfer requests.
+ *
+ * Responds to pull requests by reading blocks from local storage.
+ * Handles push requests by accepting block data and storing it locally.
+ */
+export class BlockTransferService implements Startable {
+	private running = false;
+	private readonly protocol: string;
+	private readonly repo: IRepo;
+	private readonly registrar: BlockTransferServiceComponents['registrar'];
+
+	constructor(
+		private readonly components: BlockTransferServiceComponents,
+		init: BlockTransferServiceInit = {}
+	) {
+		this.protocol = buildBlockTransferProtocol(init.protocolPrefix ?? '');
+		this.repo = components.repo;
+		this.registrar = components.registrar;
+	}
+
+	async start(): Promise<void> {
+		if (this.running) return;
+		await this.registrar.handle(this.protocol, async (data: any) => {
+			await this.handleRequest(data.stream);
+		});
+		this.running = true;
+		log('started on %s', this.protocol);
+	}
+
+	async stop(): Promise<void> {
+		if (!this.running) return;
+		await this.registrar.unhandle(this.protocol);
+		this.running = false;
+		log('stopped');
+	}
+
+	private async handleRequest(stream: Stream): Promise<void> {
+		try {
+			const request = await this.readRequest(stream);
+			log('request type=%s blocks=%d reason=%s', request.type, request.blockIds.length, request.reason);
+
+			let response: BlockTransferResponse;
+			if (request.type === 'pull') {
+				response = await this.handlePull(request);
+			} else {
+				response = await this.handlePush(request);
+			}
+
+			await this.sendResponse(stream, response);
+			log('response blocks=%d missing=%d', Object.keys(response.blocks).length, response.missing.length);
+		} catch (error) {
+			log('error: %s', (error as Error).message);
+			try {
+				await this.sendResponse(stream, { blocks: {}, missing: [] });
+			} catch {
+				// ignore send errors
+			}
+		} finally {
+			try { await stream.close(); } catch { /* ignore */ }
+		}
+	}
+
+	private async handlePull(request: BlockTransferRequest): Promise<BlockTransferResponse> {
+		const blocks: Record<string, string> = {};
+		const missing: string[] = [];
+
+		const result = await this.repo.get({ blockIds: request.blockIds });
+
+		for (const blockId of request.blockIds) {
+			const blockResult = result[blockId];
+			if (blockResult?.block) {
+				blocks[blockId] = Buffer.from(JSON.stringify(blockResult.block)).toString('base64');
+			} else {
+				missing.push(blockId);
+			}
+		}
+
+		return { blocks, missing };
+	}
+
+	// TODO: handlePush validates incoming block data but does not persist it.
+	// The pushed data is a serialized IBlock, not a full BlockArchive with revisions.
+	// Persistence should be wired when RebalanceMonitor integrates with BlockStorage.saveRestored().
+	private async handlePush(request: BlockTransferRequest): Promise<BlockTransferResponse> {
+		const blocks: Record<string, string> = {};
+		const missing: string[] = [];
+
+		if (!request.blockData) {
+			return { blocks: {}, missing: request.blockIds };
+		}
+
+		// Accept pushed blocks — for each block, validate we received parseable data
+		for (const blockId of request.blockIds) {
+			const data = request.blockData[blockId];
+			if (data) {
+				// Verify we received valid data
+				try {
+					JSON.parse(Buffer.from(data, 'base64').toString('utf8'));
+					blocks[blockId] = data;
+				} catch {
+					missing.push(blockId);
+				}
+			} else {
+				missing.push(blockId);
+			}
+		}
+
+		return { blocks, missing };
+	}
+
+	private async readRequest(stream: Stream): Promise<BlockTransferRequest> {
+		const messages: Uint8Array[] = [];
+		await pipe(
+			stream,
+			lp.decode,
+			async (source) => {
+				for await (const msg of source) {
+					messages.push(msg.subarray());
+				}
+			}
+		);
+
+		if (messages.length === 0) {
+			throw new Error('No request received');
+		}
+
+		return JSON.parse(u8ToString(messages[0]!, 'utf8')) as BlockTransferRequest;
+	}
+
+	private async sendResponse(stream: Stream, response: BlockTransferResponse): Promise<void> {
+		const bytes = u8FromString(JSON.stringify(response), 'utf8');
+		const encoded = pipe([bytes], lp.encode);
+		for await (const chunk of encoded) {
+			stream.send(chunk);
+		}
+	}
+}
+
+/** Factory for creating BlockTransferService following the libp2p service pattern. */
+export const blockTransferService = (init: BlockTransferServiceInit = {}) =>
+	(components: BlockTransferServiceComponents) => new BlockTransferService(components, init);
+
+// --- Client ---
+
+/**
+ * Client for sending block transfer requests to remote peers.
+ */
+export class BlockTransferClient extends ProtocolClient {
+	private readonly protocol: string;
+
+	constructor(
+		peerId: PeerId,
+		peerNetwork: IPeerNetwork,
+		protocolPrefix: string = ''
+	) {
+		super(peerId, peerNetwork);
+		this.protocol = buildBlockTransferProtocol(protocolPrefix);
+	}
+
+	/** Pull blocks from the remote peer. */
+	async pullBlocks(
+		blockIds: string[],
+		reason: BlockTransferRequest['reason'] = 'rebalance'
+	): Promise<BlockTransferResponse> {
+		const request: BlockTransferRequest = { type: 'pull', blockIds, reason };
+		return await this.processMessage<BlockTransferResponse>(request, this.protocol);
+	}
+
+	/** Push blocks to the remote peer. */
+	async pushBlocks(
+		blockIds: string[],
+		blockDataBuffers: Uint8Array[],
+		reason: BlockTransferRequest['reason'] = 'rebalance'
+	): Promise<BlockTransferResponse> {
+		const blockData: Record<string, string> = {};
+		for (let i = 0; i < blockIds.length; i++) {
+			blockData[blockIds[i]!] = Buffer.from(blockDataBuffers[i]!).toString('base64');
+		}
+		const request: BlockTransferRequest = { type: 'push', blockIds, reason, blockData };
+		return await this.processMessage<BlockTransferResponse>(request, this.protocol);
+	}
+}

package/src/cluster/block-transfer.ts

@@ -0,0 +1,265 @@
+import type { IRepo, IPeerNetwork } from '@optimystic/db-core';
+import { peerIdFromString } from '@libp2p/peer-id';
+import type { PartitionDetector } from './partition-detector.js';
+import type { RestorationCoordinator } from '../storage/restoration-coordinator-v2.js';
+import { BlockTransferClient } from './block-transfer-service.js';
+import type { RebalanceEvent } from './rebalance-monitor.js';
+import { createLogger } from '../logger.js';
+
+const log = createLogger('block-transfer');
+
+export interface BlockTransferConfig {
+	/** Max concurrent transfers. Default: 4 */
+	maxConcurrency?: number;
+	/** Timeout per block transfer (ms). Default: 30000 */
+	transferTimeoutMs?: number;
+	/** Retry attempts for failed transfers. Default: 2 */
+	maxRetries?: number;
+	/** Whether to push blocks to new owners proactively. Default: true */
+	enablePush?: boolean;
+}
+
+/**
+ * Coordinates block transfers in response to rebalance events.
+ *
+ * For gained blocks: delegates to RestorationCoordinator.restore() which
+ * already handles ring-based discovery and fetching.
+ *
+ * For lost blocks: proactively pushes block data to new responsible peers
+ * via the BlockTransfer protocol.
+ */
+export class BlockTransferCoordinator {
+	private readonly maxConcurrency: number;
+	private readonly transferTimeoutMs: number;
+	private readonly maxRetries: number;
+	private readonly enablePush: boolean;
+	private inFlight = new Set<string>();
+	private concurrency = 0;
+	private readonly waitQueue: Array<() => void> = [];
+
+	constructor(
+		private readonly repo: IRepo,
+		private readonly peerNetwork: IPeerNetwork,
+		private readonly restorationCoordinator: RestorationCoordinator,
+		private readonly partitionDetector: PartitionDetector,
+		private readonly protocolPrefix: string = '',
+		config: BlockTransferConfig = {}
+	) {
+		this.maxConcurrency = config.maxConcurrency ?? 4;
+		this.transferTimeoutMs = config.transferTimeoutMs ?? 30000;
+		this.maxRetries = config.maxRetries ?? 2;
+		this.enablePush = config.enablePush ?? true;
+	}
+
+	/**
+	 * Pull blocks that this node has gained responsibility for.
+	 * Uses RestorationCoordinator to discover holders and fetch block data.
+	 */
+	async pullBlocks(blockIds: string[]): Promise<{ succeeded: string[]; failed: string[] }> {
+		if (this.partitionDetector.detectPartition()) {
+			log('pull:partition-detected, skipping %d blocks', blockIds.length);
+			return { succeeded: [], failed: blockIds };
+		}
+
+		const succeeded: string[] = [];
+		const failed: string[] = [];
+
+		const ids = blockIds.filter(id => !this.inFlight.has(`pull:${id}`));
+
+		await Promise.all(ids.map(id => this.executePull(id, succeeded, failed)));
+
+		return { succeeded, failed };
+	}
+
+	/**
+	 * Push blocks that this node has lost responsibility for to new owners.
+	 */
+	async pushBlocks(
+		blockIds: string[],
+		newOwners: Map<string, string[]>
+	): Promise<{ succeeded: string[]; failed: string[] }> {
+		if (!this.enablePush) {
+			return { succeeded: [], failed: [] };
+		}
+		if (this.partitionDetector.detectPartition()) {
+			log('push:partition-detected, skipping %d blocks', blockIds.length);
+			return { succeeded: [], failed: blockIds };
+		}
+
+		const succeeded: string[] = [];
+		const failed: string[] = [];
+
+		const ids = blockIds.filter(id => !this.inFlight.has(`push:${id}`) && newOwners.has(id));
+
+		await Promise.all(ids.map(id => this.executePush(id, newOwners, succeeded, failed)));
+
+		return { succeeded, failed };
+	}
+
+	/**
+	 * Handle a complete rebalance event — pull gained, push lost.
+	 */
+	async handleRebalanceEvent(event: RebalanceEvent): Promise<void> {
+		log('rebalance:start gained=%d lost=%d', event.gained.length, event.lost.length);
+
+		const [pullResult, pushResult] = await Promise.all([
+			event.gained.length > 0 ? this.pullBlocks(event.gained) : { succeeded: [], failed: [] },
+			event.lost.length > 0 && event.newOwners.size > 0
+				? this.pushBlocks(event.lost, event.newOwners)
+				: { succeeded: [], failed: [] }
+		]);
+
+		log('rebalance:done pull=%d/%d push=%d/%d',
+			pullResult.succeeded.length, event.gained.length,
+			pushResult.succeeded.length, event.lost.length);
+	}
+
+	private async executePull(
+		blockId: string,
+		succeeded: string[],
+		failed: string[]
+	): Promise<void> {
+		const key = `pull:${blockId}`;
+		if (this.inFlight.has(key)) return;
+		this.inFlight.add(key);
+
+		try {
+			for (let attempt = 0; ; attempt++) {
+				await this.acquireSemaphore();
+				let archive: Awaited<ReturnType<RestorationCoordinator['restore']>>;
+				try {
+					archive = await this.withTimeout(
+						this.restorationCoordinator.restore(blockId),
+						this.transferTimeoutMs
+					);
+				} finally {
+					this.releaseSemaphore();
+				}
+
+				if (archive) {
+					log('pull:ok block=%s', blockId);
+					succeeded.push(blockId);
+					return;
+				}
+				if (attempt < this.maxRetries) {
+					log('pull:retry block=%s attempt=%d', blockId, attempt + 1);
+					await this.delay(this.backoffMs(attempt));
+					continue;
+				}
+				log('pull:failed block=%s', blockId);
+				failed.push(blockId);
+				return;
+			}
+		} finally {
+			this.inFlight.delete(key);
+		}
+	}
+
+	private async executePush(
+		blockId: string,
+		newOwners: Map<string, string[]>,
+		succeeded: string[],
+		failed: string[]
+	): Promise<void> {
+		const key = `push:${blockId}`;
+		if (this.inFlight.has(key)) return;
+		this.inFlight.add(key);
+
+		try {
+			for (let attempt = 0; ; attempt++) {
+				await this.acquireSemaphore();
+				let pushed = false;
+				try {
+					const owners = newOwners.get(blockId);
+					if (!owners || owners.length === 0) {
+						failed.push(blockId);
+						return;
+					}
+
+					// Read block data from local storage
+					const result = await this.repo.get({ blockIds: [blockId] });
+					const blockResult = result[blockId];
+					if (!blockResult?.block) {
+						log('push:no-local-data block=%s', blockId);
+						failed.push(blockId);
+						return;
+					}
+
+					const blockData = new TextEncoder().encode(JSON.stringify(blockResult.block));
+
+					// Push to at least one new owner
+					for (const ownerPeerIdStr of owners) {
+						try {
+							const peerId = peerIdFromString(ownerPeerIdStr);
+							const client = new BlockTransferClient(peerId, this.peerNetwork, this.protocolPrefix);
+							const response = await this.withTimeout(
+								client.pushBlocks([blockId], [blockData]),
+								this.transferTimeoutMs
+							);
+
+							if (response && !response.missing.includes(blockId)) {
+								pushed = true;
+								log('push:ok block=%s peer=%s', blockId, ownerPeerIdStr);
+								break;
+							}
+						} catch (err) {
+							log('push:peer-error block=%s peer=%s err=%s',
+								blockId, ownerPeerIdStr, (err as Error).message);
+						}
+					}
+				} finally {
+					this.releaseSemaphore();
+				}
+
+				if (pushed) {
+					succeeded.push(blockId);
+					return;
+				}
+				if (attempt < this.maxRetries) {
+					log('push:retry block=%s attempt=%d', blockId, attempt + 1);
+					await this.delay(this.backoffMs(attempt));
+					continue;
+				}
+				log('push:failed block=%s', blockId);
+				failed.push(blockId);
+				return;
+			}
+		} finally {
+			this.inFlight.delete(key);
+		}
+	}
+
+	// --- Semaphore for concurrency limiting ---
+
+	private async acquireSemaphore(): Promise<void> {
+		if (this.concurrency < this.maxConcurrency) {
+			this.concurrency++;
+			return;
+		}
+		await new Promise<void>(resolve => this.waitQueue.push(resolve));
+		this.concurrency++;
+	}
+
+	private releaseSemaphore(): void {
+		this.concurrency--;
+		const next = this.waitQueue.shift();
+		if (next) next();
+	}
+
+	// --- Helpers ---
+
+	private backoffMs(attempt: number): number {
+		return Math.min(1000 * Math.pow(2, attempt), 10000);
+	}
+
+	private delay(ms: number): Promise<void> {
+		return new Promise(resolve => setTimeout(resolve, ms));
+	}
+
+	private withTimeout<T>(promise: Promise<T>, ms: number): Promise<T | undefined> {
+		return Promise.race([
+			promise,
+			new Promise<undefined>(resolve => setTimeout(() => resolve(undefined), ms))
+		]);
+	}
+}