Segment query cache

This feature caches Query Results on a per Segment basis. Use the spark.sparklinedata.spmd.segment.query.cache setting to control Query Cache behavior. For example, here is a sample setting:

spark.sparklinedata.spmd.segment.query.cache={"useCache" : true, "sizeInMBytes" : 1024,"expireAfterSeconds" : 60,"resultSizeMax" : 20000}

Parameter	Description
useCache	turn on or off caching
sizeInMBytes	total size of the Query Cache in MB
expireAfterSeconds	the duration after which a query result is evictable
resultSizeMax	for GroupBy and Search Queries only if the number of rows in the result is below this size are they cached. The result size is estimated in a very conservative way: as the product of the cardinalities of the dimensions involved.

• under the covers the Caffeine Library is used for caching
  • Caffeine is configured to evict based on query usage and the Query Result memory footprint.
• Spark's CachedBatch mechanism is used to represent Query results in the cache.
• Currently only GroupBy, Search and Timeseries Queries are considered for caching.

Metrics

The following metrics are exposed:

Metric	Description
Hit-Count	number of times Result found in Cache
Miss-Count	number of times Result not found in Cache
Hit-Rate	Percentage of Hits
Eviction-Count	number of times a Result eviction happened because of memory pressure
Eviction-Weight	total memory size of Results evicted.
Result Serializes	Time to serialize Query Results, the histogram of the time taken, and the Histogram of the Query Result Size
Result Deserializes	Time to deserialize Query Results, the histogram of the time taken.