In Alibaba Cloud Elasticsearch Agent Engine Edition, Slice cache warming loads the Slice data that is about to be accessed into the local cache of the query nodes in advance. Warming reduces the query latency of the first access to cold data, so users get search results faster after they open an infrequently accessed knowledge base or code repository. Subsequent queries reuse the cached data instead of waiting for reads from object storage.
Use cases
Business timing | Usage and benefit |
You select a knowledge base or a code repository | Warm up the corresponding Slices asynchronously in advance to reduce the wait for the first search that follows. |
You are about to switch traffic after a bulk import | After the data becomes searchable, warm up the Slices that are about to be accessed to reduce cold-cache latency for the first batch of queries. |
You expect an access peak | Divide the data that you expect to access into batches, and prepare each batch before the peak to reduce resource contention between queries and cold data loading. |
You can run cache warming by business space so that the cache is filled first with the data that is about to be accessed. After a batch of data is cached, you do not need to warm it up again for every query.
Considerations
For data that is already active, or for workloads whose access scope cannot be predicted, let queries load data on demand. A large number of concurrent warming requests consumes network and cache resources and may evict existing hot data, so control the warming scope and the concurrency.
Cache warming is an optional performance optimization. To use it, first wait for the background automatic refresh to make newly written data searchable. Cache warming does not make data searchable earlier, and it does not guarantee that data stays in the cache. Writable data branch: Branch Slice (Beta) does not support cache warming until it is converted into a regular Slice.
What to do next
For warming requests, synchronous and asynchronous calls, task result evaluation, and query affinity configuration, see Warm Up the Slice Cache.
To organize data by tenant or by knowledge base, see Getting Started.