@databricks/zerobus-ingest-sdk 1.0.0 → 1.0.2

package/index.d.ts

@@ -0,0 +1,485 @@
+/* tslint:disable */
+/* eslint-disable */
+
+/* auto-generated by NAPI-RS */
+
+/**
+ * Record serialization format.
+ *
+ * Specifies how records should be encoded when ingested into the stream.
+ */
+export const enum RecordType {
+  /** JSON encoding - records are JSON-encoded strings */
+  Json = 0,
+  /** Protocol Buffers encoding - records are binary protobuf messages */
+  Proto = 1
+}
+/**
+ * Configuration options for the Zerobus stream.
+ *
+ * These options control stream behavior including recovery, timeouts, and inflight limits.
+ */
+export interface StreamConfigurationOptions {
+  /**
+   * Maximum number of unacknowledged requests that can be in flight.
+   * Default: 10,000
+   */
+  maxInflightRequests?: number
+  /**
+   * Enable automatic stream recovery on transient failures.
+   * Default: true
+   */
+  recovery?: boolean
+  /**
+   * Timeout for recovery operations in milliseconds.
+   * Default: 15,000 (15 seconds)
+   */
+  recoveryTimeoutMs?: number
+  /**
+   * Delay between recovery retry attempts in milliseconds.
+   * Default: 2,000 (2 seconds)
+   */
+  recoveryBackoffMs?: number
+  /**
+   * Maximum number of recovery attempts before giving up.
+   * Default: 4
+   */
+  recoveryRetries?: number
+  /**
+   * Timeout for flush operations in milliseconds.
+   * Default: 300,000 (5 minutes)
+   */
+  flushTimeoutMs?: number
+  /**
+   * Timeout waiting for server acknowledgments in milliseconds.
+   * Default: 60,000 (1 minute)
+   */
+  serverLackOfAckTimeoutMs?: number
+  /**
+   * Record serialization format.
+   * Use RecordType.Json for JSON encoding or RecordType.Proto for Protocol Buffers.
+   * Default: RecordType.Proto (Protocol Buffers)
+   */
+  recordType?: number
+  /**
+   * Maximum wait time during graceful stream close in milliseconds.
+   * When the server signals stream closure, this controls how long to wait
+   * for in-flight records to be acknowledged.
+   * - None (undefined): Wait for full server-specified duration
+   * - Some(0): Immediately trigger recovery without waiting
+   * - Some(x): Wait up to min(x, server_duration) milliseconds
+   * Default: None (wait for full server duration)
+   */
+  streamPausedMaxWaitTimeMs?: number
+}
+/**
+ * Properties of the target Delta table for ingestion.
+ *
+ * Specifies which Unity Catalog table to write to and optionally the schema descriptor
+ * for Protocol Buffers encoding.
+ */
+export interface TableProperties {
+  /** Full table name in Unity Catalog (e.g., "catalog.schema.table") */
+  tableName: string
+  /**
+   * Optional Protocol Buffer descriptor as a base64-encoded string.
+   * If not provided, JSON encoding will be used.
+   */
+  descriptorProto?: string
+}
+/**
+ * JavaScript headers provider callback wrapper.
+ *
+ * Allows TypeScript code to provide custom authentication headers
+ * by implementing a getHeaders() function.
+ */
+export interface JsHeadersProvider {
+  /** JavaScript function: () => Promise<Array<[string, string]>> */
+  getHeadersCallback: (...args: any[]) => any
+}
+/**
+ * Custom error type for Zerobus operations.
+ *
+ * This error type includes information about whether the error is retryable,
+ * which helps determine if automatic recovery can resolve the issue.
+ */
+export declare class ZerobusError {
+  /** Returns true if this error can be automatically retried by the SDK. */
+  get isRetryable(): boolean
+  /** Get the error message. */
+  get message(): string
+}
+/**
+ * A stream for ingesting data into a Databricks Delta table.
+ *
+ * The stream manages a bidirectional gRPC connection, handles acknowledgments,
+ * and provides automatic recovery on transient failures.
+ *
+ * # Example
+ *
+ * ```typescript
+ * const stream = await sdk.createStream(tableProps, clientId, clientSecret, options);
+ * const ackPromise = await stream.ingestRecord(Buffer.from([1, 2, 3]));
+ * const offset = await ackPromise;
+ * await stream.close();
+ * ```
+ */
+export declare class ZerobusStream {
+  /**
+   * Ingests a single record into the stream.
+   *
+   * **@deprecated** Use `ingestRecordOffset()` instead, which returns the offset directly
+   * after queuing. Then use `waitForOffset()` to wait for acknowledgment when needed.
+   *
+   * This method accepts either:
+   * - A Protocol Buffer encoded record as a Buffer (Vec<u8>)
+   * - A JSON string
+   *
+   * This method BLOCKS until the record is sent to the SDK's internal landing zone,
+   * then returns a Promise for the server acknowledgment. This allows you to send
+   * many records immediately without waiting for acknowledgments:
+   *
+   * ```typescript
+   * let lastAckPromise;
+   * for (let i = 0; i < 1000; i++) {
+   *     // This call blocks until record is sent (in SDK)
+   *     lastAckPromise = stream.ingestRecord(record);
+   * }
+   * // All 1000 records are now in the SDK's internal queue
+   * // Wait for the last acknowledgment
+   * await lastAckPromise;
+   * // Flush to ensure all records are acknowledged
+   * await stream.flush();
+   * ```
+   *
+   * # Arguments
+   *
+   * * `payload` - The record data. Accepts:
+   *   - Buffer (low-level proto bytes)
+   *   - string (low-level JSON string)
+   *   - Protobuf message object with .encode() method (high-level, auto-serializes)
+   *   - Plain JavaScript object (high-level, auto-stringifies to JSON)
+   *
+   * # Returns
+   *
+   * A Promise that resolves to the offset ID when the server acknowledges the record.
+   */
+  ingestRecord(payload: unknown): Promise<bigint>
+  /**
+   * Ingests multiple records as a single atomic batch.
+   *
+   * **@deprecated** Use `ingestRecordsOffset()` instead, which returns the offset directly
+   * after queuing. Then use `waitForOffset()` to wait for acknowledgment when needed.
+   *
+   * This method accepts an array of records (Protocol Buffer buffers or JSON strings)
+   * and ingests them as a batch. The batch receives a single acknowledgment from
+   * the server with all-or-nothing semantics.
+   *
+   * Similar to ingestRecord(), this BLOCKS until the batch is sent to the SDK's
+   * internal landing zone, then returns a Promise for the server acknowledgment.
+   *
+   * # Arguments
+   *
+   * * `records` - Array of record data (Buffer for protobuf, string for JSON)
+   *
+   * # Returns
+   *
+   * Promise resolving to:
+   * - `bigint`: offset ID for non-empty batches
+   * - `null`: for empty batches
+   *
+   * # Example
+   *
+   * ```typescript
+   * const buffers = records.map(r => Buffer.from(encode(r)));
+   * const offsetId = await stream.ingestRecords(buffers);
+   *
+   * if (offsetId !== null) {
+   *   console.log('Batch acknowledged at offset:', offsetId);
+   * }
+   * ```
+   */
+  ingestRecords(records: Array<unknown>): Promise<bigint | null>
+  /**
+   * Ingests a single record and returns a future that resolves to the offset ID after queuing.
+   *
+   * Unlike `ingestRecord()`, this method's Promise resolves immediately after
+   * the record is queued, without waiting for server acknowledgment. Use
+   * `waitForOffset()` to wait for acknowledgment when needed.
+   *
+   * This is the recommended API for high-throughput scenarios where you want to
+   * decouple record ingestion from acknowledgment tracking.
+   *
+   * # Arguments
+   *
+   * * `payload` - The record data (Buffer, string, protobuf message, or plain object)
+   *
+   * # Returns
+   *
+   * `Promise<bigint>` - Resolves to the offset ID immediately after the record is queued
+   * (does not wait for server acknowledgment).
+   *
+   * # Example
+   *
+   * ```typescript
+   * // Promise resolves immediately with offset (before server ack)
+   * const offset1 = await stream.ingestRecordOffset(record1);
+   * const offset2 = await stream.ingestRecordOffset(record2);
+   * // Wait for both to be acknowledged
+   * await stream.waitForOffset(offset2);
+   * ```
+   */
+  ingestRecordOffset(payload: unknown): Promise<bigint>
+  /**
+   * Ingests multiple records as a batch and returns a future that resolves to the offset ID after queuing.
+   *
+   * Unlike `ingestRecords()`, this method's Promise resolves immediately after
+   * the batch is queued, without waiting for server acknowledgment. Use
+   * `waitForOffset()` to wait for acknowledgment when needed.
+   *
+   * # Arguments
+   *
+   * * `records` - Array of record data
+   *
+   * # Returns
+   *
+   * `Promise<bigint | null>` - Resolves to the offset ID immediately after the batch
+   * is queued (does not wait for server acknowledgment). Returns null for empty batches.
+   *
+   * # Example
+   *
+   * ```typescript
+   * // Promise resolves immediately with offset (before server ack)
+   * const offset = await stream.ingestRecordsOffset(batch);
+   * if (offset !== null) {
+   *   await stream.waitForOffset(offset);
+   * }
+   * ```
+   */
+  ingestRecordsOffset(records: Array<unknown>): Promise<bigint | null>
+  /**
+   * Waits for a specific offset to be acknowledged by the server.
+   *
+   * Use this method with `ingestRecordOffset()` and `ingestRecordsOffset()` to
+   * selectively wait for acknowledgments. This allows you to ingest many records
+   * quickly and then wait only for specific offsets when needed.
+   *
+   * # Arguments
+   *
+   * * `offset_id` - The offset ID to wait for (returned by ingestRecordOffset/ingestRecordsOffset)
+   *
+   * # Errors
+   *
+   * - Timeout if acknowledgment takes too long
+   * - Server errors propagated immediately (no waiting for timeout)
+   *
+   * # Example
+   *
+   * ```typescript
+   * const offsets = [];
+   * for (const record of records) {
+   *   offsets.push(await stream.ingestRecordOffset(record));
+   * }
+   * // Wait for the last offset (implies all previous are also acknowledged)
+   * await stream.waitForOffset(offsets[offsets.length - 1]);
+   * ```
+   */
+  waitForOffset(offsetId: bigint): Promise<void>
+  /**
+   * Flushes all pending records and waits for acknowledgments.
+   *
+   * This method ensures all previously ingested records have been sent to the server
+   * and acknowledged. It's useful for checkpointing or ensuring data durability.
+   *
+   * # Errors
+   *
+   * - Timeout errors if flush takes longer than configured timeout
+   * - Network errors if the connection fails during flush
+   */
+  flush(): Promise<void>
+  /**
+   * Closes the stream gracefully.
+   *
+   * This method flushes all pending records, waits for acknowledgments, and then
+   * closes the underlying gRPC connection. Always call this method when done with
+   * the stream to ensure data integrity.
+   *
+   * # Errors
+   *
+   * - Returns an error if some records could not be acknowledged
+   * - Network errors during the close operation
+   */
+  close(): Promise<void>
+  /**
+   * Gets the list of unacknowledged records.
+   *
+   * This method should only be called after a stream failure to retrieve records
+   * that were sent but not acknowledged by the server. These records can be
+   * re-ingested into a new stream.
+   *
+   * # Returns
+   *
+   * An array of Buffers containing the unacknowledged record payloads.
+   */
+  getUnackedRecords(): Promise<Array<Buffer>>
+  /**
+   * Gets unacknowledged records grouped by their original batches.
+   *
+   * This preserves the batch structure from ingestion:
+   * - Each ingestRecord() call → 1-element batch
+   * - Each ingestRecords() call → N-element batch
+   *
+   * Should only be called after stream failure. All records returned as Buffers
+   * (JSON strings are converted to UTF-8 bytes).
+   *
+   * # Returns
+   *
+   * Array of batches, where each batch is an array of Buffers
+   *
+   * # Example
+   *
+   * ```typescript
+   * try {
+   *   await stream.ingestRecords(batch1);
+   *   await stream.ingestRecords(batch2);
+   * } catch (error) {
+   *   const unackedBatches = await stream.getUnackedBatches();
+   *
+   *   // Re-ingest with new stream
+   *   for (const batch of unackedBatches) {
+   *     await newStream.ingestRecords(batch);
+   *   }
+   * }
+   * ```
+   */
+  getUnackedBatches(): Promise<Array<Array<Buffer>>>
+}
+/**
+ * The main SDK for interacting with the Databricks Zerobus service.
+ *
+ * This is the entry point for creating ingestion streams to Delta tables.
+ *
+ * # Example
+ *
+ * ```typescript
+ * const sdk = new ZerobusSdk(
+ *   "https://workspace-id.zerobus.region.cloud.databricks.com",
+ *   "https://workspace.cloud.databricks.com"
+ * );
+ *
+ * const stream = await sdk.createStream(
+ *   { tableName: "catalog.schema.table" },
+ *   "client-id",
+ *   "client-secret"
+ * );
+ * ```
+ */
+export declare class ZerobusSdk {
+  /**
+   * Creates a new Zerobus SDK instance.
+   *
+   * # Arguments
+   *
+   * * `zerobus_endpoint` - The Zerobus API endpoint URL
+   *   (e.g., "https://workspace-id.zerobus.region.cloud.databricks.com")
+   * * `unity_catalog_url` - The Unity Catalog endpoint URL
+   *   (e.g., "https://workspace.cloud.databricks.com")
+   *
+   * # Errors
+   *
+   * - Invalid endpoint URLs
+   * - Failed to extract workspace ID from the endpoint
+   */
+  constructor(zerobusEndpoint: string, unityCatalogUrl: string)
+  /**
+   * Creates a new ingestion stream to a Delta table.
+   *
+   * This method establishes a bidirectional gRPC connection to the Zerobus service
+   * and prepares it for data ingestion. By default, it uses OAuth 2.0 Client Credentials
+   * authentication. For custom authentication (e.g., Personal Access Tokens), provide
+   * a custom headers_provider.
+   *
+   * # Arguments
+   *
+   * * `table_properties` - Properties of the target table including name and optional schema
+   * * `client_id` - OAuth 2.0 client ID (ignored if headers_provider is provided)
+   * * `client_secret` - OAuth 2.0 client secret (ignored if headers_provider is provided)
+   * * `options` - Optional stream configuration (timeouts, recovery settings, etc.)
+   * * `headers_provider` - Optional custom headers provider for authentication.
+   *   If not provided, uses OAuth with client_id and client_secret.
+   *
+   * # Returns
+   *
+   * A Promise that resolves to a ZerobusStream ready for data ingestion.
+   *
+   * # Errors
+   *
+   * - Authentication failures (invalid credentials)
+   * - Invalid table name or insufficient permissions
+   * - Network connectivity issues
+   * - Schema validation errors
+   *
+   * # Example
+   *
+   * ```typescript
+   * // OAuth authentication (default)
+   * const stream = await sdk.createStream(
+   *   { tableName: "catalog.schema.table" },
+   *   "client-id",
+   *   "client-secret"
+   * );
+   *
+   * // Custom authentication with headers provider
+   * const stream = await sdk.createStream(
+   *   { tableName: "catalog.schema.table" },
+   *   "", // ignored
+   *   "", // ignored
+   *   undefined,
+   *   {
+   *     getHeadersCallback: async () => [
+   *       ["authorization", `Bearer ${myToken}`],
+   *       ["x-databricks-zerobus-table-name", tableName]
+   *     ]
+   *   }
+   * );
+   * ```
+   */
+  createStream(tableProperties: TableProperties, clientId: string, clientSecret: string, options?: StreamConfigurationOptions | undefined | null, headersProvider?: JsHeadersProvider | undefined | null): Promise<ZerobusStream>
+  /**
+   * Recreates a stream with the same configuration and re-ingests unacknowledged batches.
+   *
+   * This method is the recommended approach for recovering from stream failures. It:
+   * 1. Retrieves all unacknowledged batches from the failed stream
+   * 2. Creates a new stream with identical configuration
+   * 3. Re-ingests all unacknowledged batches in order
+   * 4. Returns the new stream ready for continued ingestion
+   *
+   * # Arguments
+   *
+   * * `stream` - The failed or closed stream to recreate
+   *
+   * # Returns
+   *
+   * A Promise that resolves to a new ZerobusStream with all unacknowledged batches re-ingested.
+   *
+   * # Errors
+   *
+   * - Failed to retrieve unacknowledged batches from the original stream
+   * - Authentication failures when creating the new stream
+   * - Network connectivity issues during re-ingestion
+   *
+   * # Examples
+   *
+   * ```typescript
+   * try {
+   *   await stream.ingestRecords(batch);
+   * } catch (error) {
+   *   await stream.close();
+   *   // Recreate stream with all unacked batches re-ingested
+   *   const newStream = await sdk.recreateStream(stream);
+   *   // Continue ingesting with newStream
+   * }
+   * ```
+   */
+  recreateStream(stream: ZerobusStream): Promise<ZerobusStream>
+}