Impala Frequently Asked Questions

To look at the core features and functionality on Impala, the easiest way to try out Impala is to download the Cloudera QuickStart VM and start the Impala service through Cloudera Manager, then use impala-shell in a terminal window or the Impala Query UI in the Hue web interface.

To do performance testing and try out the management features for Impala on a cluster, you need to move beyond the QuickStart VM with its virtualized single-node environment. Ideally, download the Cloudera Manager software to set up the cluster, then install the Impala software through Cloudera Manager.

Cloudera offers a demonstration VM called the QuickStart VM, available in VMWare, VirtualBox, and KVM formats. For more information, see the Cloudera QuickStart VM. After booting the QuickStart VM, many services are turned off by default; in the Cloudera Manager UI that appears automatically, turn on Impala and any other components that you want to try out.

Starting with Impala 1.3.0, Impala documentation is integrated with the CDH 5 documentation, in addition to the standalone Impala documentation for use with CDH 4. For CDH 5, the core Impala developer and administrator information remains in the associated Impala documentation portion. Information about Impala release notes, installation, configuration, startup, and security is embedded in the corresponding CDH 5 guides.

• New features
• Known and fixed issues
• Incompatible changes
• Installing Impala
• Upgrading Impala
• Configuring Impala
• Starting Impala
• Security for Impala
• CDH Version and Packaging Information

Information about the latest CDH 4-compatible Impala release remains at the Impala for CDH 4 Documentation page.

More product information is available here:

• O'Reilly introductory e-book: Cloudera Impala: Bringing the SQL and Hadoop Worlds Together
• O'Reilly getting started guide for developers: Getting Started with Impala: Interactive SQL for Apache Hadoop
• Blog: Cloudera Impala: Real-Time Queries in Apache Hadoop, For Real
• Webinar: Introduction to Impala
• Product website page: Cloudera Enterprise RTQ

To see the latest release announcements for Impala, see the Cloudera Announcements forum.

• Join the Impala discussion forum and the Impala mailing list to ask questions and provide feedback.
• Use the Impala Jira project to log bug reports and requests for features.

For information on Impala requirements, see Impala Requirements. Note that there is often a minimum required level of Cloudera Manager for any given Impala version.

Although Impala is not an in-memory database, when dealing with large tables and large result sets, you should expect to dedicate a substantial portion of physical memory for the impalad daemon. Recommended physical memory for an Impala node is 128 GB or higher. If practical, devote approximately 80% of physical memory to Impala.

The amount of memory required for an Impala operation depends on several factors:

• The file format of the table. Different file formats represent the same data in more or fewer data files. The compression and encoding for each file format might require a different amount of temporary memory to decompress the data for analysis.

• Whether the operation is a SELECT or an INSERT. For example, Parquet tables require relatively little memory to query, because Impala reads and decompresses data in 8 MB chunks. Inserting into a Parquet table is a more memory-intensive operation because the data for each data file (potentially hundreds of megabytes, depending on the value of the PARQUET_FILE_SIZE query option) is stored in memory until encoded, compressed, and written to disk.

• Whether the table is partitioned or not, and whether a query against a partitioned table can take advantage of partition pruning.

• Whether the final result set is sorted by the ORDER BY clause. Each Impala node scans and filters a portion of the total data, and applies the LIMIT to its own portion of the result set. In Impala 1.4.0 and higher, if the sort operation requires more memory than is available on any particular host, Impala uses a temporary disk work area to perform the sort. The intermediate result sets are all sent back to the coordinator node, which does the final sorting and then applies the LIMIT clause to the final result set.

  For example, if you execute the query:

  select * from giant_table order by some_column limit 1000;

  and your cluster has 50 nodes, then each of those 50 nodes will transmit a maximum of 1000 rows back to the coordinator node. The coordinator node needs enough memory to sort (LIMIT * cluster_size) rows, although in the end the final result set is at most LIMIT rows, 1000 in this case.
  Likewise, if you execute the query:

  select * from giant_table where test_val > 100 order by some_column;

  then each node filters out a set of rows matching the WHERE conditions, sorts the results (with no size limit), and sends the sorted intermediate rows back to the coordinator node. The coordinator node might need substantial memory to sort the final result set, and so might use a temporary disk work area for that final phase of the query.

• Whether the query contains any join clauses, GROUP BY clauses, analytic functions, or DISTINCT operators. These operations all require some in-memory work areas that vary depending on the volume and distribution of data. In Impala 2.0 and later, these kinds of operations utilize temporary disk work areas if memory usage grows too large to handle. See SQL Operations that Spill to Disk for details.

• The size of the result set. When intermediate results are being passed around between nodes, the amount of data depends on the number of columns returned by the query. For example, it is more memory-efficient to query only the columns that are actually needed in the result set rather than always issuing SELECT *.

• The mechanism by which work is divided for a join query. You use the COMPUTE STATS statement, and query hints in the most difficult cases, to help Impala pick the most efficient execution plan. See Performance Considerations for Join Queries for details.

See Hardware Requirements for more details and recommendations about Impala hardware prerequisites.

Impala makes use of SSE 4.1 instructions.

• A large set of SQL statements, including SELECT and INSERT, with joins, Subqueries in Impala SELECT Statements, and Impala Analytic Functions. Highly compatible with HiveQL, and also including some vendor extensions. For more information, see Impala SQL Language Reference.
• Distributed, high-performance queries. See Tuning Impala for Performance for information about Impala performance optimizations and tuning techniques for queries.
• Using Cloudera Manager, you can deploy and manage your Impala services. Cloudera Manager is the best way to get started with Impala on your cluster.
• Using Hue for queries.
• Appending and inserting data into tables through the INSERT statement. See How Impala Works with Hadoop File Formats for the details about which operations are supported for which file formats.
• ODBC: Impala is certified to run against MicroStrategy and Tableau, with restrictions. For more information, see Configuring Impala to Work with ODBC.
• Querying data stored in HDFS and HBase in a single query. See Using Impala to Query HBase Tables for details.
• In Impala 2.2.0 and higher, querying data stored in the Amazon Simple Storage Service (S3). See Using Impala to Query the Amazon S3 Filesystem for details.
• Concurrent client requests. Each Impala daemon can handle multiple concurrent client requests. The effects on performance depend on your particular hardware and workload.
• Kerberos authentication. For more information, see Overview of Impala Security.
• Partitions. With Impala SQL, you can create partitioned tables with the CREATE TABLE statement, and add and drop partitions with the ALTER TABLE statement. Impala also takes advantage of the partitioning present in Hive tables. See Partitioning for Impala Tables for details.

• Querying streaming data.
• Deleting individual rows. You delete data in bulk by overwriting an entire table or partition, or by dropping a table.
• Indexing (not currently). LZO-compressed text files can be indexed outside of Impala, as described in Using LZO-Compressed Text Files.
• Full text search on text fields. The Cloudera Search product is appropriate for this use case.
• Custom Hive Serializer/Deserializer classes (SerDes). Impala supports a set of common native file formats that have built-in SerDes in CDH. See How Impala Works with Hadoop File Formats for details.
• Checkpointing within a query. That is, Impala does not save intermediate results to disk during long-running queries. Currently, Impala cancels a running query if any host on which that query is executing fails. When one or more hosts are down, Impala reroutes future queries to only use the available hosts, and Impala detects when the hosts come back up and begins using them again. Because a query can be submitted through any Impala node, there is no single point of failure. In the future, we will consider adding additional work allocation features to Impala, so that a running query would complete even in the presence of host failures.
• Hive indexes.
• Non-Hadoop data stores, such as relational databases.

For the detailed list of features that are different between Impala and HiveQL, see SQL Differences Between Impala and Hive.

Impala supports the HiveServer2 JDBC driver.

Yes, Avro is supported. Impala has always been able to query Avro tables. You can use the Impala LOAD DATA statement to load existing Avro data files into a table. Starting with Impala 1.4, you can create Avro tables with Impala. Currently, you still use the INSERT statement in Hive to copy data from another table into an Avro table. See Using the Avro File Format with Impala Tables for details.

Impala streams results whenever they are available, when possible. Certain SQL operations (aggregation or ORDER BY) require all of the input to be ready before Impala can return results.

There are many possible reasons why a given query could be slow. Use the following checklist to diagnose performance issues with existing queries, and to avoid such issues when writing new queries, setting up new nodes, creating new tables, or loading data.

• Immediately after the query finishes, issue a SUMMARY command in impala-shell. You can check which phases of execution took the longest, and compare estimated values for memory usage and number of rows with the actual values.
• Immediately after the query finishes, issue a PROFILE command in impala-shell. The numbers in the BytesRead, BytesReadLocal, and BytesReadShortCircuit should be identical for a specific node. For example:

  - BytesRead: 180.33 MB
  - BytesReadLocal: 180.33 MB
  - BytesReadShortCircuit: 180.33 MB

  If BytesReadLocal is lower than BytesRead, something in your cluster is misconfigured, such as the impalad daemon not running on all the data nodes. If BytesReadShortCircuit is lower than BytesRead, short-circuit reads are not enabled properly on that node; see Post-Installation Configuration for Impala for instructions.
• If the table was just created, or this is the first query that accessed the table after an INVALIDATE METADATA statement or after the impalad daemon was restarted, there might be a one-time delay while the metadata for the table is loaded and cached. Check whether the slowdown disappears when the query is run again. When doing performance comparisons, consider issuing a DESCRIBE table_name statement for each table first, to make sure any timings only measure the actual query time and not the one-time wait to load the table metadata.
• Is the table data in uncompressed text format? Check by issuing a DESCRIBE FORMATTED table_name statement. A text table is indicated by the line:

  InputFormat: org.apache.hadoop.mapred.TextInputFormat

  Although uncompressed text is the default format for a CREATE TABLE statement with no STORED AS clauses, it is also the bulkiest format for disk storage and consequently usually the slowest format for queries. For data where query performance is crucial, particularly for tables that are frequently queried, consider starting with or converting to a compact binary file format such as Parquet, Avro, RCFile, or SequenceFile. For details, see How Impala Works with Hadoop File Formats.
• If your table has many columns, but the query refers to only a few columns, consider using the Parquet file format. Its data files are organized with a column-oriented layout that lets queries minimize the amount of I/O needed to retrieve, filter, and aggregate the values for specific columns. See Using the Parquet File Format with Impala Tables for details.
• If your query involves any joins, are the tables in the query ordered so that the tables or subqueries are ordered with the one returning the largest number of rows on the left, followed by the smallest (most selective), the second smallest, and so on? That ordering allows Impala to optimize the way work is distributed among the nodes and how intermediate results are routed from one node to another. For example, all other things being equal, the following join order results in an efficient query:

  select some_col from
      huge_table join big_table join small_table join medium_table
    where
      huge_table.id = big_table.id
      and big_table.id = medium_table.id
      and medium_table.id = small_table.id;

  See Performance Considerations for Join Queries for performance tips for join queries.
• Also for join queries, do you have table statistics for the table, and column statistics for the columns used in the join clauses? Column statistics let Impala better choose how to distribute the work for the various pieces of a join query. See Table and Column Statistics for details about gathering statistics.
• Does your table consist of many small data files? Impala works most efficiently with data files in the multi-megabyte range; Parquet, a format optimized for data warehouse-style queries, uses large files (originally 1 GB, now 256 MB in Impala 2.0 and higher) with a block size matching the file size. Use the DESCRIBE FORMATTED table_name statement in impala-shell to see where the data for a table is located, and use the hadoop fs -ls or hdfs dfs -ls Unix commands to see the files and their sizes. If you have thousands of small data files, that is a signal that you should consolidate into a smaller number of large files. Use an INSERT ... SELECT statement to copy the data to a new table, reorganizing into new data files as part of the process. Prefer to construct large data files and import them in bulk through the LOAD DATA or CREATE EXTERNAL TABLE statements, rather than issuing many INSERT ... VALUES statements; each INSERT ... VALUES statement creates a separate tiny data file. If you have thousands of files all in the same directory, but each one is megabytes in size, consider using a partitioned table so that each partition contains a smaller number of files. See the following point for more on partitioning.
• If your data is easy to group according to time or geographic region, have you partitioned your table based on the corresponding columns such as YEAR, MONTH, and/or DAY? Partitioning a table based on certain columns allows queries that filter based on those same columns to avoid reading the data files for irrelevant years, postal codes, and so on. (Do not partition down to too fine a level; try to structure the partitions so that there is still sufficient data in each one to take advantage of the multi-megabyte HDFS block size.) See Partitioning for Impala Tables for details.

When a SELECT statement fails, the cause usually falls into one of the following categories:

• A timeout because of a performance, capacity, or network issue affecting one particular node.
• Excessive memory use for a join query, resulting in automatic cancellation of the query.
• A low-level issue affecting how native code is generated on each node to handle particular WHERE clauses in the query. For example, a machine instruction could be generated that is not supported by the processor of a certain node. If the error message in the log suggests the cause was an illegal instruction, consider turning off native code generation temporarily, and trying the query again.
• Malformed input data, such as a text data file with an enormously long line, or with a delimiter that does not match the character specified in the FIELDS TERMINATED BY clause of the CREATE TABLE statement.

When an INSERT statement fails, it is usually the result of exceeding some limit within a Hadoop component, typically HDFS.

• An INSERT into a partitioned table can be a strenuous operation due to the possibility of opening many files and associated threads simultaneously in HDFS. Impala 1.1.1 includes some improvements to distribute the work more efficiently, so that the values for each partition are written by a single node, rather than as a separate data file from each node.
• Certain expressions in the SELECT part of the INSERT statement can complicate the execution planning and result in an inefficient INSERT operation. Try to make the column data types of the source and destination tables match up, for example by doing ALTER TABLE ... REPLACE COLUMNS on the source table if necessary. Try to avoid CASE expressions in the SELECT portion, because they make the result values harder to predict than transferring a column unchanged or passing the column through a built-in function.
• Be prepared to raise some limits in the HDFS configuration settings, either temporarily during the INSERT or permanently if you frequently run such INSERT statements as part of your ETL pipeline.
• The resource usage of an INSERT statement can vary depending on the file format of the destination table. Inserting into a Parquet table is memory-intensive, because the data for each partition is buffered in memory until it reaches 1 gigabyte, at which point the data file is written to disk. Impala can distribute the work for an INSERT more efficiently when statistics are available for the source table that is queried during the INSERT statement. See Table and Column Statistics for details about gathering statistics.

Yes. Impala scales with the number of hosts. It is important to install Impala on all the DataNodes in the cluster, because otherwise some of the nodes must do remote reads to retrieve data not available for local reads. Data locality is an important architectural aspect for Impala performance. See this Impala performance blog post for background. Note that this blog post refers to benchmarks with Impala 1.1.1; Impala has added even more performance features in the 1.2.x series.

No. Impala does not make any changes to the HDFS or HBase data sets.

The default Parquet block size is relatively large (256 MB in Impala 2.0 and later; 1 GB in earlier releases). You can control the block size when creating Parquet files using the PARQUET_FILE_SIZE query option.

Impala does not cache table data. It does cache some table and file metadata. Although queries might run faster on subsequent iterations because the data set was cached in the OS buffer cache, Impala does not explicitly control this.

Impala takes advantage of the HDFS caching feature in CDH 5. You can designate which tables or partitions are cached through the CACHED and UNCACHED clauses of the CREATE TABLE and ALTER TABLE statements. Impala can also take advantage of data that is pinned in the HDFS cache through the hdfscacheadmin command. See Using HDFS Caching with Impala (CDH 5.1 or higher only) for details.

Impala is well-suited to executing SQL queries for interactive exploratory analytics on large data sets. Hive and MapReduce are appropriate for very long running, batch-oriented tasks such as ETL.

Impala does not use MapReduce at all.

For example, in an industrial environment, many agents may generate large amounts of data. Can Impala be used to analyze this data, checking for notable changes in the environment?

Complex Event Processing (CEP) is usually performed by dedicated stream-processing systems. Impala is not a stream-processing system, as it most closely resembles a relational database.

Ad-hoc queries are the primary use case for Impala. We anticipate it being used in many other situations where low-latency is required. Whether Impala is appropriate for any particular use-case depends on the workload, data size and query volume. See Impala Benefits for the primary benefits you can expect when using Impala.

Impala is different from Hive and Pig because it uses its own daemons that are spread across the cluster for queries. Because Impala does not rely on MapReduce, it avoids the startup overhead of MapReduce jobs, allowing Impala to return results in real time.

Impala adds support for UDFs in Impala 1.2. You can write your own functions in C++, or reuse existing Java-based Hive UDFs. The UDF support includes scalar functions and user-defined aggregate functions (UDAs). User-defined table functions (UDTFs) are not currently supported.

Impala does not currently support an extensible serialization-deserialization framework (SerDes), and so adding extra functionality to Impala is not as straightforward as for Hive or Pig.

Yes. There are some minor differences in how some queries are handled, but Impala queries can also be completed in Hive. Impala SQL is a subset of HiveQL, with some functional limitations such as transforms. For details of the Impala SQL dialect, see Impala SQL Statements. For the Impala built-in functions, see Impala Built-In Functions. For the detailed list of differences between Impala and HiveQL, see SQL Differences Between Impala and Hive.

There are no additional steps to allow Impala to query tables managed by Hive, whether they are stored in HDFS or HBase. Make sure that Impala is configured to access the Hive metastore correctly and you should be ready to go. Keep in mind that impalad, by default, runs as the impala user, so you might need to adjust some file permissions depending on how strict your permissions are currently.

See Using Impala to Query HBase Tables for details about querying data in HBase.

The Hive metastore service is a requirement. Impala shares the same metastore database as Hive, allowing Impala and Hive to access the same tables transparently.

Hive itself is optional, and does not need to be installed on the same nodes as Impala. Currently, Impala supports a wider variety of read (query) operations than write (insert) operations; you use Hive to insert data into tables that use certain file formats. See How Impala Works with Hadoop File Formats for details.

Impala has finished its beta release cycle, and the 1.0, 1.1, and 1.2 GA releases are production ready. The 1.1.x series includes additional security features for authorization, an important requirement for production use in many organizations. The 1.2.x series includes important performance features, particularly for large join queries. Some Cloudera customers are already using Impala for large workloads.

The Impala 1.3.0 and higher releases are bundled with corresponding levels of CDH 5. The number of new features grows with each release. See What's New in Apache Impala for a full list.

You can set up a proxy server to relay requests back and forth to the Impala servers, for load balancing and high availability. See Using Impala through a Proxy for High Availability for details.

You can enable HDFS HA for the Hive metastore. See the CDH5 High Availability Guide for details.

There is not a single point of failure in Impala. All Impala daemons are fully able to handle incoming queries. If a machine fails however, all queries with fragments running on that machine will fail. Because queries are expected to return quickly, you can just rerun the query if there is a failure. See Impala Concepts and Architecture for details about the Impala architecture.

The longer answer: Impala must be able to connect to the Hive metastore. Impala aggressively caches metadata so the metastore host should have minimal load. Impala relies on the HDFS NameNode, and, in CDH4, you can configure HA for HDFS. Impala also has centralized services, known as the statestore and catalog services, that run on one host only. Impala continues to execute queries if the statestore host is down, but it will not get state updates. For example, if a host is added to the cluster while the statestore host is down, the existing instances of impalad running on the other hosts will not find out about this new host. Once the statestore process is restarted, all the information it serves is automatically reconstructed from all running Impala daemons.

There is no defined maximum. Some customers have used Impala to query a table with over a trillion rows.

Yes. See Controlling Impala Resource Usage for how to control Impala resource usage using the Linux cgroup mechanism, and Integrated Resource Management with YARN for how to use Impala with the YARN resource management framework. Impala is designed to run on the DataNode hosts. Any contention depends mostly on the cluster setup and workload.

For a detailed example of configuring a cluster to share resources between Impala queries and MapReduce jobs, see Setting up a Multi-tenant Cluster for Impala and MapReduce

Cloudera strongly recommends running the impalad daemon on each DataNode for good performance. Although this topology is not a hard requirement, if there are data blocks with no Impala daemons running on any of the hosts containing replicas of those blocks, queries involving that data could be very inefficient. In that case, the data must be transmitted from one host to another for processing by "remote reads", a condition Impala normally tries to avoid. See Impala Concepts and Architecture for details about the Impala architecture. Impala schedules query fragments on all hosts holding data relevant to the query, if possible.

By default, Impala automatically determines the most efficient order in which to join tables using a cost-based method, based on their overall size and number of rows. (This is a new feature in Impala 1.2.2 and higher.) The COMPUTE STATS statement gathers information about each table that is crucial for efficient join performance. Impala chooses between two techniques for join queries, known as "broadcast joins" and "partitioned joins". See Joins in Impala SELECT Statements for syntax details and Performance Considerations for Join Queries for performance considerations.

Impala utilizes multiple strategies to allow joins between tables and result sets of various sizes. When joining a large table with a small one, the data from the small table is transmitted to each node for intermediate processing. When joining two large tables, the data from one of the tables is divided into pieces, and each node processes only selected pieces. See Joins in Impala SELECT Statements for details about join processing, Performance Considerations for Join Queries for performance considerations, and Query Hints in Impala SELECT Statements for how to fine-tune the join strategy.

Impala currently only supports in-memory hash aggregation. In Impala 2.0 and higher, if the memory requirements for a join or aggregation operation exceed the memory limit for a particular host, Impala uses a temporary work area on disk to help the query complete successfully.

Impala uses two pieces of metadata: the catalog information from the Hive metastore and the file metadata from the NameNode. Currently, this metadata is lazily populated and cached when an impalad needs it to plan a query.

The REFRESH statement updates the metadata for a particular table after loading new data through Hive. The INVALIDATE METADATA Statement statement refreshes all metadata, so that Impala recognizes new tables or other DDL and DML changes performed through Hive.

In Impala 1.2 and higher, a dedicated catalogd daemon broadcasts metadata changes due to Impala DDL or DML statements to all nodes, reducing or eliminating the need to use the REFRESH and INVALIDATE METADATA statements.

The load Impala generates is very similar to MapReduce. Impala contacts the NameNode during the planning phase to get the file metadata (this is only run on the host the query was sent to). Every impalad will read files as part of normal processing of the query.

These are the main factors in the performance of Impala versus that of other Hadoop components and related technologies.

Impala avoids MapReduce. While MapReduce is a great general parallel processing model with many benefits, it is not designed to execute SQL. Impala avoids the inefficiencies of MapReduce in these ways:

• Impala does not materialize intermediate results to disk. SQL queries often map to multiple MapReduce jobs with all intermediate data sets written to disk.
• Impala avoids MapReduce start-up time. For interactive queries, the MapReduce start-up time becomes very noticeable. Impala runs as a service and essentially has no start-up time.
• Impala can more naturally disperse query plans instead of having to fit them into a pipeline of map and reduce jobs. This enables Impala to parallelize multiple stages of a query and avoid overheads such as sort and shuffle when unnecessary.

Impala uses a more efficient execution engine by taking advantage of modern hardware and technologies:

• Impala generates runtime code. Impala uses LLVM to generate assembly code for the query that is being run. Individual queries do not have to pay the overhead of running on a system that needs to be able to execute arbitrary queries.
• Impala uses available hardware instructions when possible. Impala uses the supplemental SSE3 (SSSE3) instructions which can offer tremendous speedups in some cases. (Impala 2.0 and 2.1 required the SSE4.1 instruction set; Impala 2.2 and higher relax the restriction again so only SSSE3 is required.)
• Impala uses better I/O scheduling. Impala is aware of the disk location of blocks and is able to schedule the order to process blocks to keep all disks busy.
• Impala is designed for performance. A lot of time has been spent in designing Impala with sound performance-oriented fundamentals, such as tight inner loops, inlined function calls, minimal branching, better use of cache, and minimal memory usage.

Currently, if the memory required to process intermediate results on a node exceeds the amount available to Impala on that node, the query is cancelled. You can adjust the memory available to Impala on each node, and you can fine-tune the join strategy to reduce the memory required for the biggest queries. We do plan on supporting external joins and sorting in the future.

Keep in mind though that the memory usage is not directly based on the input data set size. For aggregations, the memory usage is the number of rows after grouping. For joins, the memory usage is the combined size of the tables excluding the biggest table, and Impala can use join strategies that divide up large joined tables among the various nodes rather than transmitting the entire table to each node.

If a query fails with an error indicating "memory limit exceeded", you might suspect a memory leak. The problem could actually be a query that is structured in a way that causes Impala to allocate more memory than you expect, exceeded the memory allocated for Impala on a particular node. Some examples of query or table structures that are especially memory-intensive are:

• INSERT statements using dynamic partitioning, into a table with many different partitions. (Particularly for tables using Parquet format, where the data for each partition is held in memory until it reaches the full block size in size before it is written to disk.) Consider breaking up such operations into several different INSERT statements, for example to load data one year at a time rather than for all years at once.
• GROUP BY on a unique or high-cardinality column. Impala allocates some handler structures for each different value in a GROUP BY query. Having millions of different GROUP BY values could exceed the memory limit.
• Queries involving very wide tables, with thousands of columns, particularly with many STRING columns. Because Impala allows a STRING value to be up to 32 KB, the intermediate results during such queries could require substantial memory allocation.

Impala does not currently have an UPDATE statement, which would typically be used to change a single row, a small group of rows, or a specific column. The HDFS-based files used by typical Impala queries are optimized for bulk operations across many megabytes of data at a time, making traditional UPDATE operations inefficient or impractical.

You can use the following techniques to achieve the same goals as the familiar UPDATE statement, in a way that preserves efficient file layouts for subsequent queries:

• Replace the entire contents of a table or partition with updated data that you have already staged in a different location, either using INSERT OVERWRITE, LOAD DATA, or manual HDFS file operations followed by a REFRESH statement for the table. Optionally, you can use built-in functions and expressions in the INSERT statement to transform the copied data in the same way you would normally do in an UPDATE statement, for example to turn a mixed-case string into all uppercase or all lowercase.
• To update a single row, use an HBase table, and issue an INSERT ... VALUES statement using the same key as the original row. Because HBase handles duplicate keys by only returning the latest row with a particular key value, the newly inserted row effectively hides the previous one.