Optimize performance using FetchXml

This article describes ways you can optimize performance when retrieving data using FetchXml.

Patterns to avoid

Composing optimized queries for Dataverse is vital to ensure applications provide a fast, responsive, and reliable experience. This section describes patterns to avoid and concepts to understand when composing queries for standard tables using the RetrieveMultiple message, or messages that have a parameter that inherits from the QueryBase class. This guidance also applies when sending a GET request against a collection of records using OData. The guidance here might not apply for Elastic tables or when using Dataverse Search.

Minimize the number of selected columns

Don't include columns you don't need in your query. Queries that return all columns or include a large number of columns can encounter performance issues due to the size of the dataset or complexity of the query.

This practice is especially true for logical columns. A logical column contains values that are stored in different database tables. The AttributeMetadata.IsLogical property tells you whether a column is a logical column. Queries that contain many logical columns are slower because Dataverse needs to combine the data from other database tables.

Avoid leading wild cards in filter conditions

Queries that use conditions with leading wild cards (either explicitly, or implicitly with an operator like ends-with) can lead to bad performance. Dataverse can't take advantage of database indexes when a query using leading wild cards, which forces SQL to scan the entire table. Table scans can happen even if there are other nonleading wild card queries that limit the result set.

The following example is a FetchXml condition element that uses a leading wild card:

<condition attribute='accountnumber'
   operator='like'
   value='%234' />

The following example is a QueryExpression ConditionExpression that uses a leading wild card:

new ConditionExpression("accountnumber", ConditionOperator.Like, "%234")

The following example is a OData query that uses a leading wild card:

$filter=startswith(accountnumber,'%234')

When queries time out and this pattern is detected, Dataverse returns a unique error to help identify which queries are using this pattern:

Name: LeadingWildcardCauseTimeout
Code: 0x80048573
Number: -2147187341
Message: The database operation timed out; this may be due to a leading wildcard value being used in a filter condition. Please consider removing filter conditions on leading wildcard values, as these filter conditions are expensive and may cause timeouts.

Dataverse heavily throttles leading wild card queries that are identified as a risk to the health of the org to help prevent outages. Learn more about query throttling

If you find yourself using leading wild card queries, look into these options:

• Use Dataverse search instead.
• Change your data model to help people avoid needing leading wild cards.

Other wildcard characters

As described in Use wildcard characters in conditions for string values, other characters beyond the percent sign ('%') character can act like a wildcard. Following are two example query strings that also behave like leading wildcards:

• _234%
• [^a]234%

Dataverse heavily throttles queries with search strings that start with these other leading wildcard special characters.

Hyphen character

Database collation unicode sorting rules make some search strings that start with a hyphen ('-') perform like leading wildcard searches. Search strings that start with a hyphen can't take advantage of database indexes if the search string doesn't contain a nonwildcard character before the occurrence of the '%' character in the string. For example, -% and -%234 can't efficiently use database indexes, while -234% can. Dataverse heavily throttles inefficient search strings that start with hyphens. To understand more about the database collation unicode sorting rules for hyphens, see SQL Server collations.

Avoid using formula or calculated columns in filter conditions

Formula and calculated column values are calculated in real-time when they're retrieved. Queries that use filters on these columns force Dataverse to calculate the value for each possible record that can be returned so the filter can be applied. Queries are slower because Dataverse can't improve the performance of these queries using SQL.

When queries time out and this pattern is detected, Dataverse returns a unique error to help identify which queries are using this pattern:

Name: ComputedColumnCauseTimeout
Code: 0x80048574
Number: -2147187340
Message: The database operation timed out; this may be due to a computed column being used in a filter condition. Please consider removing filter conditions on computed columns, as these filter conditions are expensive and may cause timeouts.

To help prevent outages, Dataverse applies throttles on queries that have filters on calculated columns that are identified as a risk to the health of the environment. Learn more about query throttling

Avoid ordering by choice columns

When you use FetchXml or QueryExpression, when you order query results using a choice column, the results are sorted using the localized label for each choice option. Ordering by the number value stored in the database wouldn't provide a good experience in your application. You should know that ordering on choice columns requires more compute resources to join and sort the rows by the localized label value. This extra work makes the query slower. If possible, try to avoid ordering results by choice column values.

Note

OData is different. With the Dataverse Web API, $orderby sorts rows using the integer value of the choice column rather than the localized label.

Avoid ordering by columns in related tables

Ordering by columns on related tables makes the query slower because of the added complexity.

Ordering by related tables should only be done when needed to as described here:

• Order rows using FetchXml
• Order rows using QueryExpression
• Order rows using OData

Avoid using conditions on large text columns

Dataverse has two types of columns that can store large strings of text:

• StringAttributeMetadata can store up to 4,000 characters.
• MemoAttributeMetadata can store a higher number.

The limit for both of these columns is specified using the MaxLength property.

You can use conditions on string columns that have a MaxLength configured for fewer than 850 characters.

All memo columns or string columns with a MaxLength greater than 850 are defined in Dataverse as large text columns. Large text columns are too large to effectively index, which leads to bad performance when included in a filter condition.

Dataverse search is a better choice to query data in these kinds of columns.

Late Materialize query

If you select many lookup and computed columns, and you're experiencing performance issues, you can try setting the fetch element boolean latematerialize attribute. Behind the scenes, this setting breaks the query into smaller parts and reassembles the results before returning them to you.

Using the latematerialize attribute might not always provide a performance benefit. It might make simple queries run more slowly. It's most beneficial when your query:

• Has many joins
• Contains many lookup or computed columns

Query Hints

Important

Only apply these options when recommended by Microsoft technical support. Incorrect use of these options can damage the performance of a query.

Microsoft SQL Server supports many query hints to optimize queries. FetchXML supports query hints and can pass these query options to SQL Server using the fetch element options attribute.

Query option	SQL Server hint
ForceOrder	Force Order
DisableRowGoal	Hint: DISABLE_OPTIMIZER_ROWGOAL
EnableOptimizerHotfixes	Hint: ENABLE_QUERY_OPTIMIZER_HOTFIXES
LoopJoin	Loop Join
MergeJoin	Merge Join
HashJoin	Hash Join
NO_PERFORMANCE_SPOOL	NO_PERFORMANCE_SPOOL
ENABLE_HIST_AMENDMENT_FOR_ASC_KEYS	Hint: ENABLE_HIST_AMENDMENT_FOR_ASC_KEYS

More information: Hints (Transact-SQL) - Query

No lock

In earlier versions, the no-lock attribute used to prevent shared locks on records. It's no longer necessary to include this attribute.

Union Hint

You can improve performance when adding a filter element that sets the condition for columns in different tables by setting the hint attribute to union. But there are some restrictions:

• The filter must use the or filter type.
• Each query can contain only one union hint.
• If a filter with union hint isn't at top level filter, Dataverse transforms the query and move the filter with a union hint to root filter.
• If a union hint is more than three levels deep, it's ignored.

The following example sets a filter with the union hint on the telephone1 column for both the account and contact tables.

<fetch>
   <entity name="email">
      <attribute name="activityid" />
      <attribute name="subject" />
      <filter type="and">
         <condition attribute="subject"
            operator="like"
            value="Alert:%" />
         <condition attribute="statecode"
            operator="eq"
            value="0" />
         <filter type="or"
            hint="union">
            <condition attribute="telephone1"
               operator="eq"
               value="555-123-4567"
               entityname="ac" />
            <condition attribute="telephone1"
               operator="eq"
               value="555-123-4567"
               entityname="co" />
         </filter>
      </filter>
      <link-entity name="account"
         from="accountid"
         to="regardingobjectid"
         link-type="outer"
         alias="ac" />
      <link-entity name="contact"
         from="contactid"
         to="regardingobjectid"
         link-type="outer"
         alias="co" />
   </entity>
</fetch>

See also

Query data using FetchXml
Use FetchXml to retrieve data
Select columns using FetchXml
Join tables using FetchXml
Order rows using FetchXml
Filter rows using FetchXml
Page results using FetchXml
Aggregate data using FetchXml
Count rows using FetchXml
FetchXml reference
FetchXml sample code