Search index limits - Tablestore - Alibaba Cloud Documentation Center

This topic describes the limits on search indexes.

Index limits

The following table describes the limits of a search index. These limits include the number of rows, total data size, number of fields, read and write rate, synchronization latency, and nested fields. The table also describes the limits on the fields in the data table from which the search index is created.

Restrictions	Limit	Description
Number of search indexes	None	We recommend that you create only one search index that contains all fields that you want to retrieve.
Number of rows	1 trillion	If your business requirements exceed the limit, please submit a ticket to contact us.
Total size	1 PB	If your business requirements exceed the limit, please submit a ticket to contact us.
Number of index fields	1000	If your business requirements exceed the limit, please submit a ticket to contact us.
Read and write rate	100,000 rows/s	The first time data is written or when a large volume of data is written in a short period of time, Tablestore balances loads within a few minutes. The maximum throughput for index creation strongly depends on the field type and size. The throughput capacity decreases for hot spots or when indexing large text fields, nested fields, or vector fields. If your business requirements exceed the limits, please submit a ticket to contact us.
Synchronization latency	3 seconds	In most cases, the synchronization latency is within 3 seconds. If you have higher business requirements, please submit a ticket to contact us. A new index requires up to 1 minute for initialization.
Array length	256	The maximum number of elements in an array.
Nested field	Must be unique within the search index.	A Nested field supports up to five levels of nesting. A Nested field supports a maximum of 256 child rows. A Nested field supports a maximum of 100 subfields.
Vector field	5	A search index can contain up to 5 vector type fields. If your business requirements exceed this limit, please submit a ticket. A vector field supports a maximum of 4,096 dimensions.
Total length of values in all primary key columns	1,000 bytes	The total length of values in all primary key columns cannot exceed 1,000 bytes.
Length of a String value in a primary key column	1,000 bytes	When you create an index for a primary key column of the String type, the column value cannot exceed 1,000 bytes in length.
Length of a String value in an attribute column (indexed as Keyword)	4 KB	None.
Length of a String value in an attribute column (indexed as FuzzyKeyword)	2 KB	None.
Length of a String value in an attribute column (indexed as Text)	2 MB	In most cases, this limit is the same as the length limit for an attribute column in a data table. When you perform a fuzzy query on a Text field using fuzzy tokenization, the original content is limited to a maximum of 1,024 characters or Chinese characters. Content that exceeds this limit is truncated, and only the first 1,024 characters or Chinese characters are indexed.

Query limits

Search API

The following table describes the limits of the Search API for querying, aggregating, and analyzing data.

Category	Limitations	Limitations	Description
General limits	offset+limit	100000	The default value is 10 and the maximum value is 100 for the limit parameter.If your business requirements exceed the limit, please submit a ticket to contact us. When using the Search API to query data from specific columns, if the search index contains the data for those columns, the limit parameter can be set up to 1000. This allows a single request to return up to 1,000 results. To retrieve more results, use the next_token parameter to paginate through the data.
	timeout	10 seconds	None.
	CU	None	None.
	QPS	100,000	The upper limit for lightweight transaction processing (TP) is 100,000 queries per second (QPS).If your business requirements exceed the limit, please submit a ticket to contact us.
	Number of queries in a single Search request	1,024	Complex nested queries in a Search request can affect query performance. Simplify the query for better efficiency.
Terms query	Number of search terms	1,024	A terms query supports a maximum of 1,024 search terms.
Wildcard query	Search query length	32 characters	The search query for a WildcardQuery cannot exceed 32 characters.
Prefix query	Search query length	1,000 characters	The search query for a PrefixQuery cannot exceed 1,000 characters.
Full-text search	Total number of highlighted fragments	500	The total number of highlighted fragments that can be requested and returned within inner_hits and highlight.
Full-text search	Number of highlighted fields	10	You can specify highlight parameters for a maximum of 10 fields.
Vector search	Number of top K rows returned	1,000	A KnnVectorQuery can return a maximum of the top 1,000 rows.
Aggregation	Number of Aggregations at the same level	32	When an Aggregation is added to a SubGroupBy, the count starts again from 0.
	Number of GroupBys at the same level	5	When a GroupBy is added to a SubGroupBy, the count starts again from 0.
	GroupBy nesting levels	Layer 3	The level count includes the GroupBy itself.
	Number of Filters in GroupByFilter	10	None.
	Number of groups returned by GroupByField	2,000	None.
	Number of internal sources in GroupByComposite	32	The maximum number of columns supported for grouping by multiple columns.
	Number of Ranges in GroupByRange	100	None.
	Number of Ranges in GroupByGeodistance	10	None.

ParallelScan API

The following table describes the limits of the ParallelScan API for querying data and performing parallel scans.

Category	Restrictions	Limitations	Description
General limits	offset+limit	Not supported	The offset and limit parameters are not supported. You can only traverse data from the beginning. The maximum value of the limit parameter is 2000.
	CU	None	None.
	QPS	None	None.
Parallel scan	Maximum concurrency	Equal to the MaxParallel value	The MaxParallel value returned by the ComputeSplits API.
	Task validity period (aliveTime)	10 minutes	None.
	Limit on the number of concurrent ParallelScan tasks	10	Concurrent tasks that have the same session ID and ScanQuery are considered a single task. For more information, see parallel scan.