RC Binary 2-4x Gain in CPU Efficiency

Presto uses custom fast-path decoding logic for specific Hive file formats. In this release we have added a fast path for for RCFile when using the Binary SerDe (LazyBinaryColumnarSerDe). In our micro benchmarks, we see a gain between 2x and 4x in CPU efficiency compared to the generic (slow) path. Since Hive data decoding accounts for a significant portion of the CPU time, this should result in measurable gains for most queries over RC Binary encoded data. Note that this optimization may not result in a reduction in latency if your cluster is network or disk I/O bound.

Hash Distributed Aggregations

GROUP BY aggregations are now distributed across a fixed number of machines. This is controlled by the property query.initial-hash-partitions set in etc/config.properties of the coordinator and workers. If the value is larger than the number of machines available during query scheduling, Presto will use all available machines. The default value is 8.

The maximum memory size of an aggregation is now query.initial-hash-partitions times task.max-memory.

Simple Distinct Aggregations

We have added support for the DISTINCT argument qualifier for aggregation functions. This is currently limited to queries without a GROUP BY clause and where all the aggregation functions have the same input expression. For example:

SELECT count(DISTINCT country)
FROM users

Support for complete DISTINCT functionality is in our roadmap.

Range Predicate Pushdown

We’ve modified the connector API to support range predicates in addition to simple equality predicates. This lays the ground work for adding connectors to systems that support range scans (e.g., HBase, Cassandra, JDBC, etc).

In addition to receiving range predicates, the connector can also communicate back the ranges of each partition for use in the query optimizer. This can be a major performance gain for JOIN queries where one side of the join has only a few partitions. For example:

SELECT * FROM data_1_year JOIN data_1_week USING (ds)

If data_1_year and data_1_week are both partitioned on ds, the connector will report back that one table has partitions for 365 days and the other table has partitions for only 7 days. Then the optimizer will limit the scan of the data_1_year table to only the 7 days that could possible match. These constraints are combined with other predicates in the query to further limit the data scanned.

json_array_get Function

The json_array_get() function makes it simple to fetch a single element from a scalar json array.

Non-reserved Keywords

The keywords DATE, TIME, TIMESTAMP, and INTERVAL are no longer reserved keywords in the grammar. This means that you can access a column named date without quoting the identifier.

CLI source Option

The Presto CLI now has an option to set the query source. The source value is shown in the UI and is recorded in events. When using the CLI in shell scripts it is useful to set the --source option to distinguish shell scripts from normal users.

SHOW SCHEMAS FROM

Although the documentation included the syntax SHOW SCHEMAS [FROM catalog], it was not implemented. This release now implements this statement correctly.

Hive Bucketed Table Fixes

For queries over Hive bucketed tables, Presto will attempt to limit scans to the buckets that could possible contain rows that match the WHERE clause. Unfortunately, the algorithm we were using to select the buckets was not correct, and sometimes we would either select the wrong files or fail to select any files. We have aligned the algorithm with Hive and now the optimization works as expected.

We have also improved the algorithm for detecting tables that are not properly bucketed. It is common for tables to declare bucketing in the Hive metadata, but not actually be bucketed in HDFS. When Presto detects this case, it fallback to a full scan of the partition. Not only does this change make bucketing safer, but it makes it easier to migrate a table to use bucketing without rewriting all of the data.