Edit

SQL Performance Tuning

This article outlines basic and advanced optimization techniques for Ignite SQL queries. Some of the sections are also useful for debugging and troubleshooting.

Basic Considerations: Ignite vs RDBMS

Ignite is frequently compared to relational databases for their SQL capabilities with an expectation that existing SQL queries, created for an RDBMS, will work out of the box and perform faster in Ignite without any changes. Usually, such an assumption is based on the fact that Ignite stores and processes data in-memory. However, it’s not enough just to put data in RAM and expect an order of magnitude increase in performance. Generally, extra tuning is required. Below you can see a standard checklist of best practices to consider before you benchmark Ignite against an RDBMS or do any performance testing:

Ignite is optimized for multi-nodes deployments with RAM as a primary storage. Don’t try to compare a single-node Ignite cluster to a relational database. You should deploy a multi-node Ignite cluster with the whole copy of data in RAM.
Be ready to adjust your data model and existing SQL queries. Use the affinity colocation concept during the data modelling phase for proper data distribution. Remember, it’s not enough just to put data in RAM. If your data is properly colocated, you can run SQL queries with JOINs at massive scale and expect significant performance benefits.
Define secondary indexes and use other standard, and Ignite-specific, tuning techniques described below.
Keep in mind that relational databases leverage local caching techniques and, depending on the total data size, an RDBMS can complete some queries even faster than Ignite even in a multi-node configuration. If your data set is around 10-100GB and an RDBMS has enough RAM for caching data locally than it, for instance, can outperform a multi-node Ignite cluster because the latter will be utilizing the network. Store much more data in Ignite to see the difference.

Using the EXPLAIN Statement

Ignite supports the EXPLAIN statement which could be used to read the execution plan of a query. Use this command to analyse your queries for possible optimization. Note that the plan will contain multiple rows: the last one will contain a query for the reducing side (usually your application), others are for map nodes (usually server nodes). Read the Distributed Queries section to learn how queries are executed in Ignite.

EXPLAIN SELECT name FROM Person WHERE age = 26;

The execution plan is generated by H2 as described here.

OR Operator and Selectivity

If a query contains an OR operator, then indexes may not be used as expected depending on the complexity of the query. For example, for the query select name from Person where gender='M' and (age = 20 or age = 30), an index on the gender field will be used instead of an index on the age field, although the latter is a more selective index. As a workaround for this issue, you can rewrite the query with UNION ALL (notice that UNION without ALL will return DISTINCT rows, which will change the query semantics and will further penalize your query performance):

SELECT name FROM Person WHERE gender='M' and age = 20
UNION ALL
SELECT name FROM Person WHERE gender='M' and age = 30

Avoid Having Too Many Columns

Avoid having too many columns in the result set of a SELECT query. Due to limitations of the H2 query parser, queries with 100+ columns may perform worse than expected.

Lazy Loading

By default, Ignite will set Lazy Loading enabled, this will minimize memory consumption at the cost of moderate performance degradation.

Otherwise, Ignite attempts to load the whole result set to memory and send it back to the query initiator (which is usually your application). This approach provides optimal performance for queries of small or medium result sets, and minimizes the duration of internal database locks, thus increasing concurrency.

WARNING, if the result set is too big to fit in the available memory, it can lead to prolonged GC pauses and even OutOfMemoryError exceptions. This property is deprecated since Ignite 2.15.0.

To change the default behavior:

SqlFieldsQuery query = new SqlFieldsQuery("SELECT * FROM Person WHERE id > 10");

// Result set will be loaded lazily.
query.setLazy(false);

jdbc:ignite:thin://192.168.0.15?lazy=false

var query = new SqlFieldsQuery("SELECT * FROM Person WHERE id > 10")
{
    // Result set will be loaded lazily.
    Lazy = false
};

Querying Colocated Data

When Ignite executes a distributed query, it sends sub-queries to individual cluster nodes to fetch the data and groups the results on the reducer node (usually your application). If you know in advance that the data you are querying is colocated by the GROUP BY condition, you can use SqlFieldsQuery.collocated = true to tell the SQL engine to do the grouping on the remote nodes. This will reduce network traffic between the nodes and query execution time. When this flag is set to true, the query is executed on individual nodes first and the results are sent to the reducer node for final calculation.

Consider the following example, in which we assume that the data is colocated by department_id (in other words, the department_id field is configured as the affinity key).

SELECT SUM(salary) FROM Employee GROUP BY department_id

Because of the nature of the SUM operation, Ignite will sum the salaries across the elements stored on individual nodes, and then send these sums to the reducer node where the final result will be calculated. This operation is already distributed, and enabling the collocated flag will only slightly improve performance.

Let’s take a slightly different example:

SELECT AVG(salary) FROM Employee GROUP BY department_id

In this example, Ignite has to fetch all (salary, department_id) pairs to the reducer node and calculate the results there. However, if employees are colocated by the department_id field, i.e. employee data for the same department is stored on the same node, setting SqlFieldsQuery.collocated = true will reduce query execution time because Ignite will calculate the averages for each department on the individual nodes and send the results to the reducer node for final calculation.

Enforcing Join Order

When this flag is set, the query optimizer will not reorder tables in joins. In other words, the order in which joins are applied during query execution will be the same as specified in the query. Without this flag, the query optimizer can reorder joins to improve performance. However, sometimes it might make an incorrect decision. This flag helps to control and explicitly specify the order of joins instead of relying on the optimizer.

Consider the following example:

SELECT * FROM Person p
JOIN Company c ON p.company = c.name where p.name = 'John Doe'
AND p.age > 20
AND p.id > 5000
AND p.id < 100000
AND c.name NOT LIKE 'O%';

This query contains a join between two tables: Person and Company. To get the best performance, we should understand which join will return the smallest result set. The table with the smaller result set size should be given first in the join pair. To get the size of each result set, let’s test each part.

Q1:

SELECT count(*)
FROM Person p
where
p.name = 'John Doe'
AND p.age > 20
AND p.id > 5000
AND p.id < 100000;

Q2:

SELECT count(*)
FROM Company c
where
c.name NOT LIKE 'O%';

After running Q1 and Q2, we can get two different outcomes:

Case 1:

30000

100000

Q2 returns more entries than Q1. In this case, we don’t need to modify the original query, because smaller subset has already been located on the left side of the join.

Case 2:

50000

10000

Q1 returns more entries than Q2. So we need to change the initial query as follows:

SELECT *
FROM Company c
JOIN Person p
ON p.company = c.name
where
p.name = 'John Doe'
AND p.age > 20
AND p.id > 5000
AND p.id < 100000
AND c.name NOT LIKE 'O%';

The force join order hint can be specified as follows:

JDBC driver connection parameter
ODBC driver connection attribute
If you use SqlFieldsQuery to execute SQL queries, you can set the enforce join order hint by calling the SqlFieldsQuery.setEnforceJoinOrder(true) method.

Increasing Index Inline Size

Every entry in the index has a constant size which is calculated during index creation. This size is called index inline size. Ideally this size should be enough to store full indexed entry in serialized form. When values are not fully included in the index, Ignite may need to perform additional data page reads during index lookup, which can impair performance if persistence is enabled.

Here is how values are stored in the index:

int
0     1       5
| tag | value |
Total: 5 bytes

long
0     1       9
| tag | value |
Total: 9 bytes

String
0     1      3             N
| tag | size | UTF-8 value |
Total: 3 + string length

POJO (BinaryObject)
0     1         5
| tag | BO hash |
Total: 5

For primitive data types (bool, byte, short, int, etc.), Ignite automatically calculates the index inline size so that the values are included in full. For example, for int fields, the inline size is 5 (1 byte for the tag and 4 bytes for the value itself). For long fields, the inline size is 9 (1 byte for the tag + 8 bytes for the value).

For binary objects, the index includes the hash of each object, which is enough to avoid collisions. The inline size is 5.

For variable length data, indexes include only first several bytes of the value. Therefore, when indexing fields with variable-length data, we recommend that you estimate the length of your field values and set the inline size to a value that includes most (about 95%) or all values. For example, if you have a String field with 95% of the values containing 10 characters or fewer, you can set the inline size for the index on that field to 13.

The inline sizes explained above apply to single field indexes. However, when you define an index on a field in the value object or on a non-primary key column, Ignite creates a composite index by appending the primary key to the indexed value. Therefore, when calculating the inline size for composite indexes, add up the inline size of the primary key.

Below is an example of index inline size calculation for a cache where both key and value are complex objects.

public class Key {
    @QuerySqlField
    private long id;

    @QuerySqlField
    @AffinityKeyMapped
    private long affinityKey;
}

public class Value {
    @QuerySqlField(index = true)
    private long longField;

    @QuerySqlField(index = true)
    private int intField;

    @QuerySqlField(index = true)
    private String stringField; // we suppose that 95% of the values are 10 symbols
}

The following table summarizes the inline index sizes for the indexes defined in the example above.

Index	Kind	Recommended Inline Size	Comment
(_key)	Primary key index	5	Inlined hash of a binary object (5)
(affinityKey, _key)	Affinity key index	14	Inlined long (9) + binary object’s hash (5)
(longField, _key)	Secondary index	14	Inlined long (9) + binary object’s hash (5)
(intField, _key)	Secondary index	10	Inlined int (5) + binary object up to hash (5)
(stringField, _key)	Secondary index	18	Inlined string (13) + binary object’s hash (5) (assuming that the string is ~10 symbols)

Index

Kind

Recommended Inline Size

Comment

(_key)

Primary key index

Inlined hash of a binary object (5)

(affinityKey, _key)

Affinity key index

Inlined long (9) + binary object’s hash (5)

(longField, _key)

Secondary index

Inlined long (9) + binary object’s hash (5)

(intField, _key)

Secondary index

Inlined int (5) + binary object up to hash (5)

(stringField, _key)

Secondary index

Inlined string (13) + binary object’s hash (5) (assuming that the string is ~10 symbols)

Note that you will only have to set the inline size for the index on stringField. For other indexes, Ignite will calculate the inline size automatically.

Refer to the Configuring Index Inline Size section for the information on how to change the inline size.

You can check the inline size of an existing index in the INDEXES system view.

Warning

Note that since Ignite encodes strings to UTF-8, some characters use more than 1 byte.

Query Parallelism

By default, a SQL query is executed in a single thread on each participating Ignite node. This approach is optimal for queries returning small result sets involving index search. For example:

SELECT * FROM Person WHERE p.id = ?;

Certain queries might benefit from being executed in multiple threads. This relates to queries with table scans and aggregations, which is often the case for HTAP and OLAP workloads. For example:

SELECT SUM(salary) FROM Person;

The number of threads created on a single node for query execution is configured per cache and by default equals 1. You can change the value by setting the CacheConfiguration.queryParallelism parameter. If you create SQL tables using the CREATE TABLE command, you can use a cache template to set this parameter.

If a query contains JOINs, then all the participating caches must have the same degree of parallelism.

Index Hints

Index hints are useful in scenarios when you know that one index is more suitable for certain queries than another. You can use them to instruct the query optimizer to choose a more efficient execution plan. To do this, you can use USE INDEX(indexA,…,indexN) statement as shown in the following example.

SELECT * FROM Person USE INDEX(index_age)
WHERE salary > 150000 AND age < 35;

Partition Pruning

Partition pruning is a technique that optimizes queries that use affinity keys in the WHERE condition. When executing such a query, Ignite will scan only those partitions where the requested data is stored. This will reduce query time because the query will be sent only to the nodes that store the requested partitions.

In the following example, the employee objects are colocated by the id field (if an affinity key is not set explicitly then the primary key is used as the affinity key):

CREATE TABLE employee (id BIGINT PRIMARY KEY, department_id INT, name VARCHAR)

/* This query is sent to the node where the requested key is stored */
SELECT * FROM employee WHERE id=10;

/* This query is sent to all nodes */
SELECT * FROM employee WHERE department_id=10;

In the next example, the affinity key is set explicitly and, therefore, will be used to colocate data and direct queries to the nodes that keep primary copies of the data:

CREATE TABLE employee (id BIGINT PRIMARY KEY, department_id INT, name VARCHAR) WITH "AFFINITY_KEY=department_id"

/* This query is sent to all nodes */
SELECT * FROM employee WHERE id=10;

/* This query is sent to the node where the requested key is stored */
SELECT * FROM employee WHERE department_id=10;

Note	Refer to affinity colocation page for more details on how data gets colocated and how it helps boost performance in distributed storages like Ignite.

Skip Reducer on Update

When Ignite executes a DML operation, it first fetches all the affected intermediate rows for analysis to the reducer node (usually your application), and only then prepares batches of updated values that will be sent to remote nodes.

This approach might affect performance and saturate the network if a DML operation has to move many entries.

Use this flag as a hint for the SQL engine to do all intermediate rows analysis and updates “in-place” on the server nodes. The hint is supported for JDBC and ODBC connections.

//jdbc connection string
jdbc:ignite:thin://192.168.0.15/skipReducerOnUpdate=true

SQL On-heap Row Cache

Ignite stores data and indexes in its own memory space outside of Java heap. This means that with every data access, a part of the data will be copied from the off-heap space to Java heap, potentially deserialized, and kept in the heap as long as your application or server node references it.

The SQL on-heap row cache is intended to store hot rows (key-value objects) in Java heap, minimizing resources spent for data copying and deserialization. Each cached row refers to an entry in the off-heap region and can be invalidated when one of the following happens:

The master entry stored in the off-heap region is updated or removed.
The data page that stores the master entry is evicted from RAM.

The on-heap row cache can be enabled for a specific cache/table (if you use CREATE TABLE to create SQL tables and caches, then the parameter can be passed via a cache template):

<bean class="org.apache.ignite.configuration.IgniteConfiguration">
    <property name="cacheConfiguration">
        <bean class="org.apache.ignite.configuration.CacheConfiguration">
            <property name="name" value="myCache"/>
            <property name="sqlOnheapCacheEnabled" value="true"/>
        </bean>
    </property>
</bean>

If the row cache is enabled, you might be able to trade RAM for performance. You might get up to a 2x performance increase for some SQL queries and use cases by allocating more RAM for rows caching purposes.

Warning

SQL On-Heap Row Cache Size

Presently, the cache is unlimited and can occupy as much RAM as allocated to your memory data regions. Make sure to:

Set the JVM max heap size equal to the total size of all the data regions that store caches for which this on-heap row cache is enabled.
Tune JVM garbage collection accordingly.

Using TIMESTAMP instead of DATE

Use the TIMESTAMP type instead of DATE whenever possible. Presently, the DATE type is serialized/deserialized very inefficiently resulting in performance degradation.

© 2024 The Apache Software Foundation.
Apache, Apache Ignite, the Apache feather and the Apache Ignite logo are either registered trademarks or trademarks of The Apache Software Foundation.