DashCache - a dashboard focused caching tool for prometheus

DashCache is a proxy for prometheus which can accelerate the common case of graphing a time range, and then asking for almost the same data 1m later but shifted 1m.

The idea is that you can serve the same content 1m later, but chop off the previous minute and query prom for what’s happened in the last minute.

Ancedotally it makes many of my production workloads 4x - 50x faster.

It leverages the postgres range datatype and GIST indexes to make range overlap queries pretty efficient.

Warning: this is pre-alpha. I literally don’t even know if it will continue to work after a couple days, or how aggressively old cache entries need to be purged, or how the efficient-looking postgres query plans will scale as the dataset gets larger.

Below: query plan for cache fetch query, which essentially says: “Look for entries matching my query and step values that are to the left of the current range and also intersect with the current range. Order by tsrange and give me the first match.

 Limit  (cost=9.53..9.54 rows=1 width=64) (actual time=0.044..0.044 rows=0 loops=1)
   ->  Sort  (cost=9.53..9.54 rows=1 width=64) (actual time=0.043..0.043 rows=0 loops=1)
         Sort Key: tsrange DESC
         Sort Method: quicksort  Memory: 25kB
         ->  Bitmap Heap Scan on query_cache  (cost=4.16..9.52 rows=1 width=64) (actual time=0.010..0.010 rows=0 loops=1)
               Recheck Cond: ((tsrange &< '[1511834338,1511841538)'::int8range) AND (tsrange && '[1511834338,1511841538)'::int8range))
               Filter: ((tsrange <> '[1511834338,1511841538)'::int8range) AND (query = 'sum(odnd_fetchCount{origin="www_walmart_com"})'::text) AND (step = 20))
               ->  Bitmap Index Scan on query_cache_tsrange  (cost=0.00..4.16 rows=2 width=0) (actual time=0.006..0.006 rows=0 loops=1)
                     Index Cond: ((tsrange &< '[1511834338,1511841538)'::int8range) AND (tsrange && '[1511834338,1511841538)'::int8range))
 Planning time: 0.629 ms
 Execution time: 0.109 ms
(11 rows)

Building

The daemon is in the root of the repo: go get github.com/shanemhansen/dashcache

Running

Dashcache requires a postgres database. It does not auto-create the schema or anything. The schema file has a schema with the proper index setup.