[SEA-NodeJS] Kernel-parity batch: mTLS + custom headers/UA + retry/backoff + operation-status fields

KERNEL_REV

@@ -1 +1 @@
-80b68e1eef3b613910183a50dfa4dace854d50dd
+fcc459bbf3f39bf57e2ee02f14b99c0ec7a70123

lib/DBSQLClient.ts

@@ -152,6 +152,8 @@ export default class DBSQLClient extends EventEmitter implements IDBSQLClient, I
 
       useLZ4Compression: true,
 
+      preserveBigNumericPrecision: false,
+
       // Telemetry defaults are sourced from DEFAULT_TELEMETRY_CONFIG so
       // every component reads from the same single frozen const. Mapping the
       // unprefixed TelemetryConfiguration keys to the `telemetry`-prefixed
@@ -604,6 +606,11 @@ export default class DBSQLClient extends EventEmitter implements IDBSQLClient, I
       this.config.enableMetricViewMetadata = options.enableMetricViewMetadata;
     }
 
+    // Opt-in: preserve DECIMAL (string) / BIGINT (bigint) precision in results.
+    if (options.preserveBigNumericPrecision !== undefined) {
+      this.config.preserveBigNumericPrecision = options.preserveBigNumericPrecision;
+    }
+
     // Override telemetry config if provided in options. Per-key narrowed copy
     // preserves the structural type system: `ConnectionOptions` and
     // `ClientConfig` declare identical types for these knobs, so a user

lib/DBSQLParameter.ts

@@ -29,6 +29,37 @@ export enum DBSQLParameterType {
   INTERVALDAY = 'INTERVAL DAY',
 }
 
+// 32-bit signed integer bounds — the range of the Spark `INT` type.
+const INT32_MIN = -2147483648;
+const INT32_MAX = 2147483647;
+
+/**
+ * Infer the Spark parameter type for a JS `number` when the caller didn't set
+ * one explicitly.
+ *
+ * A JS `number` is an IEEE-754 double, so a whole-number value can still be far
+ * outside the `INT` range (e.g. `1e30`). Typing such a value as `INTEGER`
+ * makes the server reject it (`invalid INT literal "1e+30"`). Pick the
+ * narrowest type that actually fits:
+ *   - non-integer / non-finite → `DOUBLE`
+ *   - integer within INT (i32) range → `INTEGER`
+ *   - integer within the safe-integer range → `BIGINT`
+ *   - anything larger → `DOUBLE` (can't be represented exactly as an integer
+ *     anyway; callers needing exact 64-bit integers should pass a `bigint`).
+ */
+function inferNumberType(value: number): DBSQLParameterType {
+  if (!Number.isInteger(value)) {
+    return DBSQLParameterType.DOUBLE;
+  }
+  if (value >= INT32_MIN && value <= INT32_MAX) {
+    return DBSQLParameterType.INTEGER;
+  }
+  if (Number.isSafeInteger(value)) {
+    return DBSQLParameterType.BIGINT;
+  }
+  return DBSQLParameterType.DOUBLE;
+}
+
 interface DBSQLParameterOptions {
   type?: DBSQLParameterType;
   value: DBSQLParameterValue;
@@ -78,7 +109,7 @@ export class DBSQLParameter {
     if (typeof this.value === 'number') {
       return new TSparkParameter({
         name,
-        type: wireType ?? (Number.isInteger(this.value) ? DBSQLParameterType.INTEGER : DBSQLParameterType.DOUBLE),
+        type: wireType ?? inferNumberType(this.value),
         value: new TSparkParameterValue({
           stringValue: Number(this.value).toString(),
         }),
@@ -96,11 +127,17 @@ export class DBSQLParameter {
     }
 
     if (this.value instanceof Date) {
+      // A `Date` bound as `DATE` must project a calendar date (`yyyy-mm-dd`),
+      // not a full ISO-8601 timestamp: the SEA wire rejects
+      // `2024-03-14T00:00:00.000Z` as a DATE literal ("trailing input"), and
+      // Thrift accepts the date-only form just as well. Without an explicit
+      // DATE type the value still binds as a TIMESTAMP from the full ISO string.
+      const isDateType = wireType === DBSQLParameterType.DATE;
       return new TSparkParameter({
         name,
         type: wireType ?? DBSQLParameterType.TIMESTAMP,
         value: new TSparkParameterValue({
-          stringValue: this.value.toISOString(),
+          stringValue: isDateType ? this.value.toISOString().slice(0, 10) : this.value.toISOString(),
         }),
       });
     }

lib/contracts/IClientContext.ts

@@ -26,6 +26,12 @@ export interface ClientConfig {
   useLZ4Compression: boolean;
   enableMetricViewMetadata?: boolean;
 
+  // When true, DECIMAL values are returned as exact strings and 64-bit
+  // integers as JS `bigint`, instead of being coerced to a lossy `number`.
+  // Off by default to preserve the long-standing representation on both the
+  // Thrift and SEA backends. See `ConnectionOptions.preserveBigNumericPrecision`.
+  preserveBigNumericPrecision?: boolean;
+
   // Telemetry configuration
   telemetryEnabled?: boolean;
   telemetryBatchSize?: number;

lib/contracts/IDBSQLClient.ts

@@ -55,6 +55,15 @@ export type ConnectionOptions = {
   proxy?: ProxyOptions;
   enableMetricViewMetadata?: boolean;
 
+  /**
+   * Preserve full numeric precision in results. When `true`, DECIMAL columns
+   * are returned as exact strings and 64-bit integers (BIGINT) as JS `bigint`,
+   * instead of the default lossy coercion to a JS `number` (which silently
+   * rounds DECIMALs and integers beyond 2^53). Applies to both the Thrift and
+   * SEA backends. Defaults to `false` to preserve the existing representation.
+   */
+  preserveBigNumericPrecision?: boolean;
+
   /**
    * Extra HTTP headers attached to driver-owned out-of-band requests
    * (telemetry POSTs and feature-flag GETs). Not applied to the primary

lib/contracts/InternalConnectionOptions.ts

@@ -29,16 +29,49 @@ export interface InternalConnectionOptions {
   /**
    * SEA-only: verify the server's TLS certificate. Secure-by-default — omit
    * to keep full chain + hostname verification; set `false` only to opt into
-   * the insecure accept-anything mode.
+   * the insecure accept-anything mode. This is the master verify toggle:
+   * `false` also subsumes the hostname check (see
+   * `checkServerCertificateHostname`). Mirrors the Python connector's
+   * `_tls_no_verify` (inverted).
    * @internal SEA path only.
    */
   checkServerCertificate?: boolean;
 
+  /**
+   * SEA-only: verify that the server certificate matches the host
+   * (hostname-vs-SNI check), independently of full chain validation. Omit
+   * to keep the secure default (on); set `false` to skip only the hostname
+   * check while still validating the chain — e.g. connecting via an IP
+   * literal or a host the cert wasn't issued for. No-op when
+   * `checkServerCertificate` is `false` (that disables everything). Mirrors
+   * the Python connector's `_tls_verify_hostname`.
+   * @internal SEA path only.
+   */
+  checkServerCertificateHostname?: boolean;
+
   /**
    * SEA-only: PEM-encoded CA certificate (string or `Buffer`) added to the
    * trust store on top of the system roots — for TLS-inspecting proxies or
    * on-prem internal CAs. Honoured regardless of `checkServerCertificate`.
    * @internal SEA path only.
    */
   customCaCert?: Buffer | string;
+
+  /**
+   * SEA-only: PEM-encoded client certificate (string or `Buffer`) for
+   * mutual TLS (mTLS). Must be supplied together with `clientKeyPem`; a
+   * leaf cert optionally followed by its intermediate chain is accepted.
+   * Mirrors the Python connector's `_tls_client_cert_file`.
+   * @internal SEA path only.
+   */
+  clientCertPem?: Buffer | string;
+
+  /**
+   * SEA-only: PEM-encoded private key (string or `Buffer`) for the mTLS
+   * client certificate. Must be supplied together with `clientCertPem`.
+   * For portability supply a PKCS#8 key (`BEGIN PRIVATE KEY`). Mirrors the
+   * Python connector's `_tls_client_cert_key_file`.
+   * @internal SEA path only.
+   */
+  clientKeyPem?: Buffer | string;
 }

lib/contracts/OperationStatus.ts

@@ -53,4 +53,33 @@ export interface OperationStatus {
    * to `WaitUntilReadyOptions.callback` for the consumer to interpret.
    */
   progressUpdateResponse?: unknown;
+
+  /**
+   * Number of rows modified by a DML statement (UPDATE / INSERT / DELETE /
+   * MERGE). `undefined`/`null` for SELECT and on backends/warehouses that do
+   * not surface the counter. Mirrors Thrift's
+   * `TGetOperationStatusResp.numModifiedRows`.
+   */
+  numModifiedRows?: number | null;
+
+  /**
+   * Server-supplied user-facing message, when the backend exposes one. Mirrors
+   * Thrift's `TGetOperationStatusResp.displayMessage`. May contain SQL
+   * fragments or parameter values — treat as potentially sensitive.
+   */
+  displayMessage?: string | null;
+
+  /**
+   * Server-supplied diagnostic detail (multi-line operator / stack context),
+   * when available. Mirrors Thrift's `TGetOperationStatusResp.diagnosticInfo`.
+   * For support surfaces, not user-facing.
+   */
+  diagnosticInfo?: string | null;
+
+  /**
+   * Server-supplied JSON blob with extended error details, when available.
+   * Mirrors Thrift's `TGetOperationStatusResp.errorDetailsJson`. Pass-through
+   * string — callers parse with `JSON.parse` if they need structured access.
+   */
+  errorDetailsJson?: string | null;
 }

lib/result/ArrowResultConverter.ts

@@ -15,7 +15,7 @@ import {
   RecordBatchReader,
   util as arrowUtils,
 } from 'apache-arrow';
-import { TTableSchema, TColumnDesc } from '../../thrift/TCLIService_types';
+import { TTableSchema, TColumnDesc, TTypeId } from '../../thrift/TCLIService_types';
 import IClientContext from '../contracts/IClientContext';
 import HiveDriverError from '../errors/HiveDriverError';
 import IResultsProvider, { ResultsProviderFetchNextOptions } from './IResultsProvider';
@@ -169,13 +169,41 @@ function formatDayTimeFromTotal(totalNanos: bigint): string {
   return `${sign}${days.toString()} ${pad2(hours)}:${pad2(minutes)}:${pad2(seconds)}${fraction}`;
 }
 
+/**
+ * Render an Arrow `Decimal` value — supplied as its unscaled integer (from
+ * `bigNumToBigInt`) plus the column `scale` — as an exact decimal string,
+ * e.g. unscaled `1234567890` / scale `5` → `"12345.67890"`. Used by the
+ * precision-preserving path so high-precision DECIMALs survive the round-trip
+ * instead of being flattened to an IEEE-754 double.
+ */
+export function bigNumDecimalToString(unscaled: bigint, scale: number): string {
+  if (scale <= 0) {
+    return unscaled.toString();
+  }
+  const negative = unscaled < ZERO_BIGINT;
+  // `padStart(scale + 1)` guarantees at least one digit before the point
+  // (e.g. unscaled `5` / scale `2` → `"005"` → `"0.05"`).
+  const digits = (negative ? -unscaled : unscaled).toString().padStart(scale + 1, '0');
+  const cut = digits.length - scale;
+  return `${negative ? '-' : ''}${digits.slice(0, cut)}.${digits.slice(cut)}`;
+}
+
 export default class ArrowResultConverter implements IResultsProvider<Array<any>> {
   private readonly context: IClientContext;
 
   private readonly source: IResultsProvider<ArrowBatch>;
 
   private readonly schema: Array<TColumnDesc>;
 
+  // When true, DECIMAL and 64-bit integer values keep full precision —
+  // DECIMAL as an exact string and BIGINT as a JS `bigint` — instead of being
+  // coerced to a lossy `number`. Enabled by the SEA backend, which always
+  // receives native Arrow `Decimal128` / `Int64` from the kernel and has no
+  // server-side "send as string" escape hatch (the Thrift backend gets the
+  // string form via `useArrowNativeTypes=false`). Off by default so the Thrift
+  // path keeps its long-standing `number` representation unchanged.
+  private readonly preserveBigNumericPrecision: boolean;
+
   private recordBatchReader?: IterableIterator<RecordBatch<TypeMap>>;
 
   // Remaining rows in current Arrow batch (not the record batch!)
@@ -193,10 +221,16 @@ export default class ArrowResultConverter implements IResultsProvider<Array<any>
   // operation backend and the SEA backend's neutral `ResultMetadata` —
   // which both carry `schema?: TTableSchema` — can construct the converter
   // without an adapter at the call site.
-  constructor(context: IClientContext, source: IResultsProvider<ArrowBatch>, { schema }: { schema?: TTableSchema }) {
+  constructor(
+    context: IClientContext,
+    source: IResultsProvider<ArrowBatch>,
+    { schema }: { schema?: TTableSchema },
+    { preserveBigNumericPrecision = false }: { preserveBigNumericPrecision?: boolean } = {},
+  ) {
     this.context = context;
     this.source = source;
     this.schema = getSchemaColumns(schema);
+    this.preserveBigNumericPrecision = preserveBigNumericPrecision;
   }
 
   public async hasMore() {
@@ -374,6 +408,11 @@ export default class ArrowResultConverter implements IResultsProvider<Array<any>
     if (value instanceof Object && value[isArrowBigNumSymbol]) {
       const result = bigNumToBigInt(value);
       if (DataType.isDecimal(valueType)) {
+        // Preserve full precision as an exact string when requested (SEA);
+        // otherwise keep the historical lossy `number` form.
+        if (this.preserveBigNumericPrecision) {
+          return bigNumDecimalToString(result, valueType.scale);
+        }
         return Number(result) / 10 ** valueType.scale;
       }
       // A rewritten Duration Int64 surfaces as a raw `bigint`, not a BigNum
@@ -397,6 +436,12 @@ export default class ArrowResultConverter implements IResultsProvider<Array<any>
       if (durationUnit) {
         return formatDurationToIntervalDayTime(value, durationUnit);
       }
+      // Keep the exact `bigint` when precision must be preserved (SEA); the
+      // default path narrows to `number` for backward compatibility (the
+      // Thrift backend has always returned BIGINT as a JS `number`).
+      if (this.preserveBigNumericPrecision) {
+        return value;
+      }
       return Number(value);
     }
 
@@ -411,7 +456,23 @@ export default class ArrowResultConverter implements IResultsProvider<Array<any>
       const typeDescriptor = column.typeDesc.types[0]?.primitiveEntry;
       const field = column.columnName;
       const value = record[field];
-      result[field] = value === null ? null : convertThriftValue(typeDescriptor, value);
+      if (value === null) {
+        result[field] = null;
+        return;
+      }
+      // When preserving precision, DECIMAL and BIGINT values were already
+      // produced in their exact form by `convertArrowTypes` (string / bigint).
+      // `convertThriftValue` would narrow both back to a lossy `number`
+      // (DECIMAL_TYPE → `Number(value)`, BIGINT_TYPE → `convertBigInt`), so
+      // pass them through untouched on this path.
+      if (
+        this.preserveBigNumericPrecision &&
+        (typeDescriptor?.type === TTypeId.DECIMAL_TYPE || typeDescriptor?.type === TTypeId.BIGINT_TYPE)
+      ) {
+        result[field] = value;
+        return;
+      }
+      result[field] = convertThriftValue(typeDescriptor, value);
     });
 
     return result;