Query acceleration

Simple Log Service accelerates PromQL queries through global cache and concurrent computing.

How it works

Global cache and concurrent computing work as follows.

Global cache

The Prometheus Query engine does not cache results by default — every query recomputes all data, which degrades performance over large datasets and long time ranges. Global cache reuses partial results from previous identical queries (same PromQL statement and step value). Cached results cover the overlapping time range; only uncached intervals are recomputed.

Important

Global cache aligns the query time range (start/end) to integer multiples of the step parameter, which increases the cache hit ratio. Query results are updated in the cache after each query completes.
Incomplete query results and results outside the cached time range are not cached, which preserves data integrity.

Concurrent computing

Standard Prometheus queries run in a single coroutine on one server, which is slow for large numbers of time series, long time ranges, or complex logic. Simple Log Service extends the Prometheus engine to split PromQL queries by time interval or time series and distribute them across multiple servers, improving performance by 2x to 10x.

Splitting by time interval

Example: a 12-hour query is split into 6 subqueries at 2-hour intervals. All 6 subqueries run concurrently and results are merged.
```
query:    sum(metric)
interval: 12h
step:     2m
```
Splitting by time series

Example: a metric with 500,000 time series and 10 concurrent tasks runs 10 parallel tasks of 50,000 time series each, then merges results.

Important

The SLS time series engine automatically identifies PromQL queries eligible for concurrent computing and distributes them across nodes.
Concurrent computing benefits queries with long time ranges or large numbers of time series. For small data volumes, it may degrade performance. Use HTTP FormValue mode to configure concurrent computing per query.
Aggregation without() (equivalent to by on all labels) yields minimal performance gains because intermediate results are barely reduced. With many time series, performance may even degrade. Aggregation by on multiple labels also shows limited improvement.

Parameters

Configure query acceleration parameters in MetricsConfig or HTTP FormValue mode.

Category	Parameter	Description	MetricsConfig	FormValue	Description
parallel_config (parameters of concurrent computing)	enable	Enables concurrent computing. Disabled by default.	Supported	Supported	Splits a query into subqueries distributed across child compute nodes. Results are aggregated on the primary node.
	mode	Concurrent computing mode. Valid values: auto: The system selects concurrency based on recent query results. static: Manually configure time splitting and concurrency.	Supported	Unsupported	auto or static mode. For static mode, consult Simple Log Service technical support.
	timePieceInterval	Time-split interval in seconds. Valid values: 3600 to 2592000. Default: 21600 (6 hours).	Supported	Supported	Time-split interval in seconds. In the console, specify a value in whole hours.
	timePieceCount	Number of subqueries after time-based splitting. Valid values: 1 to 16. Default: 8.	Supported	Supported	Number of subqueries after time-based splitting.
	totalParallelCount	Total concurrent tasks for time-series splitting. Valid values: 2 to 64. Default: 8.	Supported	Supported	Total tasks for time-series splitting. Example: a metric with 5 million time series and totalParallelCount=10 runs 10 tasks of 500,000 time series each.
	parallelCountPerHost	Concurrent tasks per server. Valid values: 1 to 8. Default: 2.	Supported	Supported	Number of time-series split tasks assigned to each server.
query_cache_config (parameters of global cache)	enable	Enables global cache. Disabled by default.	Supported	Supported	Reuses partial results from previous identical queries.

MetricsConfig mode

MetricsConfig is a per-Metricstore parameter configurable through the Simple Log Service console or SDKs. Changes take effect after 3 minutes.

Console

On the Metricstore Attribute page, perform the following operations. For more information about how to access the Metricstore Attribute page, see Modify a metricstore.

Concurrent computing modes:

In auto mode, Simple Log Service estimates the concurrency level based on data volume from recent identical queries.
In static mode, you manually configure concurrency. Consult Simple Log Service technical support before using static mode.

SDK

Use the Simple Log Service SDK for Java to modify MetricsConfig in JSON format. Example:

{
  "parallel_config": {
    "enable": true,
    "mode": "static",
    "parallel_count_per_host": 2,
    "time_piece_count": 8,
    "time_piece_interval": 21600,
    "total_parallel_count": 8
  },
  "query_cache_config": {
    "enable": true
  }
}

The following table maps query acceleration parameters to MetricsConfig JSON fields and FormValue keys.

Category	Parameter	MetricsConfig	FormValue
parallel_config	enable	enable	x-sls-parallel-enable
	mode	mode	None
	timePieceInterval	time_piece_interval	x-sls-parallel-time-piece-interval
	timePieceCount	time_piece_count	x-sls-parallel-time-piece-count
	totalParallelCount	total_parallel_count	x-sls-parallel-count
	parallelCountPerHost	parallel_count_per_host	x-sls-parallel-count-per-host
query_cache_config	enable	enable	x-sls-global-cache-enable

HTTP FormValue mode

In HTTP FormValue mode, query acceleration settings apply only to the current request. For more information about Metricstore-related HTTP API, see Time series metric query APIs.

Examples:

Global cache

Add x-sls-global-cache-enable=true to enable global cache.

https://{project}.{sls-endpoint}/prometheus/{project}/{metricstore}/api/v1/query_range?query=sum(up)&start=1690876800&end=1690877800&step=10&x-sls-global-cache-enable=true

Concurrent computing

Add x-sls-parallel-enable=true&x-sls-parallel-count=16 to enable concurrent computing and set the degree of concurrency to 16.

https://{project}.{sls-endpoint}/prometheus/{project}/{metricstore}/api/v1/query_range?query=sum(up)&start=1690876800&end=1690877800&step=10&x-sls-parallel-enable=true&x-sls-parallel-count=16