ES|QL TS command

The TS source command is similar to the FROM source command, with the following key differences:

• Targets only time series indices
• Enables the use of time series aggregation functions inside the STATS command

TS index_pattern [METADATA fields]
		

index_pattern

A list of indices, data streams or aliases. Supports wildcards and date math.

fields

A comma-separated list of metadata fields to retrieve.

The TS source command enables time series semantics and adds support for time series aggregation functions to the STATS command, such as AVG_OVER_TIME(), or RATE. These functions are implicitly evaluated per time series, then aggregated by group using a secondary aggregation function. For an example, refer to Calculate the rate of search requests per host.

This paradigm (a pair of aggregation functions) is standard for time series querying. For supported inner (time series) functions per metric type, refer to ES|QL time series aggregation functions. These functions also apply to downsampled data, with the same semantics as for raw data.

You can use time series aggregation functions directly in the STATS command ( ). The output will contain one aggregate value per time series and time bucket (if specified). For an example, refer to Use time series aggregation functions directly.

You can also combine time series aggregation functions with regular aggregation functions such as SUM(), as outer aggregation functions. For examples, refer to Combine SUM and RATE and Combine SUM and AVG_OVER_TIME.

However, using a time series aggregation function in combination with an inner time series function causes an error. For an example, refer to Invalid query: nested time series functions.

• Avoid aggregating multiple metrics in the same query when those metrics have different dimensional cardinalities. For example, in STATS max(rate(foo)) + rate(bar)), if foo and bar don't share the same dimension values, the rate for one metric will be null for some dimension combinations. Because the + operator returns null when either input is null, the entire result becomes null for those dimensions. Additionally, queries that aggregate a single metric can filter out null values more efficiently.
• Use the TS command for aggregations on time series data, rather than FROM. The FROM command is still available (for example, for listing document contents), but it's not optimized for processing time series data and may produce unexpected results.
• The TS command can't be combined with certain operations (such as FORK) before the STATS command is applied. Once STATS is applied, you can process the tabular output with any applicable ES|QL operations.
• Add a time range filter on @timestamp to limit the data volume scanned and improve query performance.

The following examples demonstrate common time series query patterns using TS.

Calculate the total rate of search requests (tracked by the search_requests counter) per host and hour. The RATE() function is applied per time series in hourly buckets. These rates are summed for each host and hourly bucket (since each host can map to multiple time series):

TS metrics
  | WHERE @timestamp >= now() - 1 hour
  | STATS SUM(RATE(search_requests)) BY TBUCKET(1 hour), host
		

The following two queries are equivalent, returning the average of the last memory usage values per time series. If a query is missing an inner (time series) aggregation function, LAST_OVER_TIME() is assumed and used implicitly:

TS metrics | STATS AVG(memory_usage)

TS metrics | STATS AVG(LAST_OVER_TIME(memory_usage))
		

This query calculates the average memory usage across per-time-series averages, rather than the average of all raw values:

TS metrics | STATS AVG(AVG_OVER_TIME(memory_usage))
		

You can use a time series aggregation function directly in STATS ( ):

TS metrics
| WHERE TRANGE(1 day)
| STATS RATE(search_requests) BY TBUCKET(1 hour)
		

Use SUM as the outer aggregation to sum counter rates across groups:

TS metrics | STATS SUM(RATE(search_requests)) BY host
		

Use AVG_OVER_TIME to compute per-time-series averages, then group the results by host and time bucket:

TS metrics
| WHERE @timestamp >= now() - 1 day
| STATS SUM(AVG_OVER_TIME(memory_usage)) BY host, TBUCKET(1 hour)
		

Using a time series aggregation function in combination with an inner time series function causes an error:

TS metrics | STATS AVG_OVER_TIME(RATE(memory_usage))