feat(oracle-driver): Improvements & fixes, fix cube-js#10834 (cube-js#11174)

Author: ovr

.github/workflows/drivers-tests.yml

@@ -265,6 +265,7 @@ jobs:
           - databricks-jdbc-export-bucket-gcs-prefix
           - mssql
           - mysql
+          - oracle
           - postgres
           - postgres-pre-agg-credentials
           - redshift
@@ -299,6 +300,8 @@ jobs:
             use_tesseract_sql_planner: true
           - database: mssql
             use_tesseract_sql_planner: true
+          - database: oracle
+            use_tesseract_sql_planner: true
       fail-fast: false
 
     steps:

packages/cubejs-oracle-driver/driver/OracleDriver.js

@@ -7,11 +7,28 @@
 const {
   getEnv,
   assertDataSource,
+  Pool,
 } = require('@cubejs-backend/shared');
-const { BaseDriver, TableColumn } = require('@cubejs-backend/base-driver');
+const { BaseDriver, TableColumn, createPoolName } = require('@cubejs-backend/base-driver');
 const oracledb = require('oracledb');
 const { reduce } = require('ramda');
 
+// Maps Oracle `metaData.dbTypeName` strings to Cube generic types. NUMBER and the
+// TIMESTAMP* family are handled separately (scale-based / prefix match) below.
+const OracleTypeToGenericType = {
+  varchar2: 'text',
+  nvarchar2: 'text',
+  char: 'text',
+  nchar: 'text',
+  clob: 'text',
+  nclob: 'text',
+  long: 'text',
+  binary_float: 'float',
+  binary_double: 'double',
+  date: 'timestamp',
+  'number': 'decimal',
+};
+
 const sortByKeys = (unordered) => {
   const ordered = {};
 
@@ -43,20 +60,11 @@ const reduceCb = (result, i) => {
   return sortByKeys(result);
 };
 
-/**
- * Oracle driver class.
- */
 class OracleDriver extends BaseDriver {
-  /**
-   * Returns default concurrency value.
-   */
   static getDefaultConcurrency() {
     return 2;
   }
 
-  /**
-   * Class constructor.
-   */
   constructor(config = {}) {
     super({
       testConnectionTimeout: config.testConnectionTimeout,
@@ -72,20 +80,51 @@ class OracleDriver extends BaseDriver {
     this.db.partRows = 100000;
     this.db.maxRows = 100000;
     this.db.prefetchRows = 500;
+
+    const { maxPoolSize, pool, ...connectionConfig } = config;
+
     this.config = {
       user: getEnv('dbUser', { dataSource, preAggregations }),
       password: getEnv('dbPass', { dataSource, preAggregations }),
       db: getEnv('dbName', { dataSource, preAggregations }),
       host: getEnv('dbHost', { dataSource, preAggregations }),
       port: getEnv('dbPort', { dataSource, preAggregations }) || 1521,
-      poolMin: 0,
-      poolMax:
-        config.maxPoolSize ||
-        getEnv('dbMaxPoolSize', { dataSource, preAggregations }) ||
-        50,
-      ...config
+      ...connectionConfig,
     };
     this.config.connectionString = this.config.connectionString || `${this.config.host}:${this.config.port}/${this.config.db}`;
+
+    const poolName = createPoolName('oracle', dataSource, preAggregations);
+    this.pool = new Pool(poolName, {
+      create: async () => {
+        const connection = await this.db.getConnection(this.config);
+        await OracleDriver.initConnection(connection);
+
+        return connection;
+      },
+      validate: async (connection) => {
+        try {
+          await connection.ping();
+        } catch (e) {
+          this.databasePoolError(e);
+          return false;
+        }
+
+        return true;
+      },
+      destroy: (connection) => connection.close(),
+    }, {
+      min: 0,
+      max:
+        maxPoolSize ||
+        getEnv('dbMaxPoolSize', { dataSource, preAggregations }) ||
+        50,
+      evictionRunIntervalMillis: 10000,
+      softIdleTimeoutMillis: 30000,
+      idleTimeoutMillis: 30000,
+      testOnBorrow: true,
+      acquireTimeoutMillis: 20000,
+      ...pool,
+    });
   }
 
   async tablesSchema() {
@@ -110,12 +149,33 @@ class OracleDriver extends BaseDriver {
     return reduce(reduceCb, {}, data);
   }
 
-  async getConnectionFromPool() {
-    if (!this.pool) {
-      this.pool = await this.db.createPool(this.config);
-    }
+  /**
+   * Runs once per pooled session. Aligns the session NLS formats with the ISO-ish
+   * date strings Cube binds, so implicit string→DATE/TIMESTAMP conversions (e.g.
+   * the native planner's `CAST(? AS TIMESTAMP)` over a 'YYYY-MM-DD' filter bound)
+   * parse instead of failing with ORA-01843 under Oracle's default NLS. Explicit
+   * TO_DATE/TO_TIMESTAMP calls carry their own masks and are unaffected.
+   * @protected
+   */
+  static async initConnection(connection) {
+    await connection.execute(
+      "ALTER SESSION SET NLS_DATE_FORMAT = 'YYYY-MM-DD' NLS_TIMESTAMP_FORMAT = 'YYYY-MM-DD' NLS_TIMESTAMP_TZ_FORMAT = 'YYYY-MM-DD HH24:MI:SS.FF TZH:TZM'"
+    );
+  }
+
+  /**
+   * Acquires a connection from the pool, runs `fn`, and always releases the
+   * connection back to the pool (on both success and failure).
+   * @protected
+   */
+  async withConnection(fn) {
+    const connection = await this.pool.acquire();
 
-    return this.pool.getConnection()
+    try {
+      return await fn(connection);
+    } finally {
+      await this.pool.release(connection);
+    }
   }
 
   async testConnection() {
@@ -127,36 +187,93 @@ class OracleDriver extends BaseDriver {
       throw new Error('Oracle can not work with table names longer than 128 symbols. ' +
         `Consider using the 'sqlAlias' attribute in your cube definition for ${quotedTableName}.`);
     }
+
     return super.createTable(quotedTableName, columns);
   }
 
-  async query(query, values) {
-    const conn = await this.getConnectionFromPool();
+  static normalizeParams(query, values) {
+    if (!values || values.length === 0) {
+      return { sql: query, binds: {} };
+    }
 
-    try {
-      const res = await conn.execute(query, values || {});
+    const binds = {};
+    const valueToName = new Map();
+    let idx = 0;
+    let nextName = 0;
+
+    // `:"?"` must be matched as a whole before a lone `?`, so it appears first
+    // in the alternation; since it starts with `:`, its inner `?` is consumed
+    // as part of the match and never matched again on its own.
+    //
+    // Placeholders carrying the same value share a single named bind. This is
+    // semantically identical (the same value is bound) and keeps repeated
+    // expressions textually identical across clauses — required by Oracle, which
+    // otherwise rejects e.g. a CASE expression in both SELECT and GROUP BY when
+    // its param renders as two different bind names (ORA-00979).
+    const sql = query.replace(/:"\?"|\?/g, () => {
+      const value = values[idx];
+      idx += 1;
+      // A Map distinguishes values by SameValueZero, so 1 and '1' stay separate;
+      // the raw value works as the key without stringifying.
+      let name = valueToName.get(value);
+      if (name === undefined) {
+        name = `cb_param_${nextName}`;
+        nextName += 1;
+        valueToName.set(value, name);
+        binds[name] = value;
+      }
+      return `:${name}`;
+    });
+
+    return { sql, binds };
+  }
+
+  async query(query, values) {
+    return this.withConnection(async (conn) => {
+      const { sql, binds } = OracleDriver.normalizeParams(query, values);
+      const res = await conn.execute(sql, binds);
       return res && res.rows;
-    } catch (e) {
-      throw (e);
-    } finally {
-      try {
-        await conn.close();
-      } catch (e) {
-        throw e;
+    });
+  }
+
+  static metaDataToColumnTypes(metaData) {
+    return (metaData || []).map((column) => {
+      const dbTypeName = (column.dbTypeName || '').toLowerCase();
+      let type = 'text';
+
+      if (dbTypeName.startsWith('timestamp')) {
+        type = 'timestamp';
+      } else {
+        type = OracleTypeToGenericType[dbTypeName] || 'text';
       }
-    }
+
+      return { name: column.name, type };
+    });
+  }
+
+  async downloadQueryResults(query, values, _options) {
+    return this.withConnection(async (conn) => {
+      const { sql, binds } = OracleDriver.normalizeParams(query, values);
+      const res = await conn.execute(sql, binds);
+      return {
+        rows: (res && res.rows) || [],
+        types: OracleDriver.metaDataToColumnTypes(res && res.metaData),
+      };
+    });
   }
 
-  release() {
-    return this.pool && this.pool.close();
+  async release() {
+    await this.pool.drain();
+    await this.pool.clear();
   }
 
   readOnly() {
     return true;
   }
 
   wrapQueryWithLimit(query) {
-    query.query = `SELECT * FROM (${query.query}) AS t WHERE ROWNUM <= ${query.limit}`;
+    // Oracle forbids the `AS` keyword for table/subquery aliases.
+    query.query = `SELECT * FROM (${query.query}) t WHERE ROWNUM <= ${query.limit}`;
   }
 }
 

packages/cubejs-oracle-driver/package.json

@@ -14,8 +14,12 @@
   "main": "driver/OracleDriver.js",
   "dependencies": {
     "@cubejs-backend/base-driver": "1.6.64",
+    "@cubejs-backend/shared": "1.6.64",
     "ramda": "^0.27.0"
   },
+  "devDependencies": {
+    "@types/oracledb": "^6.2.3"
+  },
   "optionalDependencies": {
     "oracledb": "^6.2.0"
   },

packages/cubejs-schema-compiler/src/adapter/OracleQuery.ts

@@ -11,6 +11,7 @@ const GRANULARITY_VALUE = {
   minute: 'mm',
   second: 'ss',
   month: 'MM',
+  quarter: 'Q',
   year: 'YYYY'
 };
 
@@ -94,6 +95,7 @@ export class OracleQuery extends BaseQuery {
     if (!granularity) {
       return dimension;
     }
+
     return `TRUNC(${dimension}, '${GRANULARITY_VALUE[granularity]}')`;
   }
 
@@ -179,6 +181,85 @@ export class OracleQuery extends BaseQuery {
     return res;
   }
 
+  public dateBin(interval: string, source: string, origin: string): string {
+    const parsed = parseSqlInterval(interval);
+    const originTs = `TO_TIMESTAMP('${origin}', 'YYYY-MM-DD"T"HH24:MI:SS.FF3')`;
+
+    const totalMonths = (parsed.year || 0) * 12 + (parsed.quarter || 0) * 3 + (parsed.month || 0);
+    const totalSeconds = (parsed.week || 0) * 604800 + (parsed.day || 0) * 86400 +
+      (parsed.hour || 0) * 3600 + (parsed.minute || 0) * 60 + (parsed.second || 0);
+
+    // Pure month-based interval: bin with calendar-accurate month arithmetic.
+    if (totalMonths > 0 && totalSeconds === 0) {
+      return `ADD_MONTHS(${originTs}, FLOOR(MONTHS_BETWEEN(${source}, ${originTs}) / ${totalMonths}) * ${totalMonths})`;
+    }
+
+    // Pure fixed-length interval: bin with second arithmetic.
+    // (CAST(... AS DATE) - CAST(... AS DATE)) yields a day count; * 86400 → seconds.
+    if (totalSeconds > 0 && totalMonths === 0) {
+      const diffSeconds = `(CAST(${source} AS DATE) - CAST(${originTs} AS DATE)) * 86400`;
+      return `${originTs} + NUMTODSINTERVAL(FLOOR(${diffSeconds} / ${totalSeconds}) * ${totalSeconds}, 'SECOND')`;
+    }
+
+    throw new UserError(`Mixed month/second intervals are not supported for Oracle custom granularities: ${interval}`);
+  }
+
+  public seriesSql(timeDimension) {
+    const values = timeDimension.timeSeries().map(
+      ([from, to]) => `SELECT '${from}' f, '${to}' t FROM DUAL`
+    ).join(' UNION ALL ');
+    return `SELECT TO_TIMESTAMP(dates.f, 'YYYY-MM-DD"T"HH24:MI:SS.FF3') as ${this.escapeColumnName('date_from')}, ` +
+      `TO_TIMESTAMP(dates.t, 'YYYY-MM-DD"T"HH24:MI:SS.FF3') as ${this.escapeColumnName('date_to')} ` +
+      `FROM (${values}) dates`;
+  }
+
+  public sqlTemplates() {
+    const templates = super.sqlTemplates();
+    // Oracle forbids `AS` before a table/subquery alias.
+    templates.expressions.query_aliased = '{{ query }} {{ quoted_alias }}';
+    // Oracle does not support positional GROUP BY — group by expressions.
+    templates.statements.group_by_exprs = '{{ group_by | map(attribute=\'expr\') | join(\', \') }}';
+    // No `AS` before the FROM subquery alias, and Oracle row-limiting syntax.
+    templates.statements.select = '{% if ctes %} WITH \n' +
+      '{{ ctes | join(\',\n\') }}\n' +
+      '{% endif %}' +
+      'SELECT {% if distinct %}DISTINCT {% endif %}' +
+      '{{ select_concat | map(attribute=\'aliased\') | join(\', \') }} {% if from %}\n' +
+      'FROM (\n' +
+      '{{ from | indent(2, true) }}\n' +
+      ') {{ from_alias }}{% elif from_prepared %}\n' +
+      'FROM {{ from_prepared }}' +
+      '{% endif %}' +
+      '{% for join in joins %}\n{{ join }}{% endfor %}' +
+      '{% if filter %}\nWHERE {{ filter }}{% endif %}' +
+      '{% if group_by %}\nGROUP BY {{ group_by }}{% endif %}' +
+      '{% if having %}\nHAVING {{ having }}{% endif %}' +
+      '{% if order_by %}\nORDER BY {{ order_by | map(attribute=\'expr\') | join(\', \') }}{% endif %}' +
+      '{% if offset is not none %}\nOFFSET {{ offset }} ROWS{% endif %}' +
+      '{% if limit is not none %}\nFETCH NEXT {{ limit }} ROWS ONLY{% endif %}';
+    // Oracle has no `::` cast and no `VALUES` row-constructor table source. Build the
+    // series with TO_TIMESTAMP + UNION ALL SELECT ... FROM DUAL, and no `AS` before
+    // the derived-table alias (Oracle forbids it). `seria` items are [from, to] pairs.
+    templates.statements.time_series_select = 'SELECT TO_TIMESTAMP(dates.f, \'YYYY-MM-DD"T"HH24:MI:SS.FF3\') AS "date_from",\n' +
+      'TO_TIMESTAMP(dates.t, \'YYYY-MM-DD"T"HH24:MI:SS.FF3\') AS "date_to" \n' +
+      'FROM (\n' +
+      '{% for time_item in seria %}' +
+      'SELECT \'{{ time_item[0] }}\' f, \'{{ time_item[1] }}\' t FROM DUAL' +
+      '{% if not loop.last %} UNION ALL\n{% endif %}' +
+      '{% endfor %}' +
+      ') dates';
+
+    delete templates.expressions.ilike;
+    templates.tesseract.ilike = 'LOWER({{ expr }}) {% if negated %}NOT {% endif %}LIKE LOWER({{ pattern }})';
+
+    // Oracle has no `STRING` type (used by the default in CAST(... AS STRING),
+    // e.g. the multi-column count() concatenation). CAST to VARCHAR2 requires a
+    // length, so use the max standard VARCHAR2 size.
+    templates.types.string = 'VARCHAR2(4000)';
+
+    return templates;
+  }
+
   public newFilter(filter) {
     return new OracleFilter(this, filter);
   }