hello, im just curious how queries run faster when you call the exact same query the second time round. how does caching work in pinot? if it matters, i’m using the inverted index, with no partition pruning (only 1 broker running). broker looks something like this:

Processed requestId=34,table=sorted_events_OFFLINE,segments(queried/processed/matched/consuming)=198/198/198/-1,schedulerWaitMs=0,reqDeserMs=0,totalExecMs=426,resSerMs=0,totalTimeMs=426,minConsumingFreshnessMs=-1,broker=Broker_172.26.0.4_8099,numDocsScanned=259467,scanInFilter=619119085,scanPostFilter=259467,sched=fcfs

Slow query: request handler processing time: 427, send response latency: 3, total time to handle request: 430
Processed requestId=35,table=events_OFFLINE,segments(queried/processed/matched/consuming)=198/198/118/-1,schedulerWaitMs=0,reqDeserMs=5,totalExecMs=221,resSerMs=0,totalTimeMs=226,minConsumingFreshnessMs=-1,broker=Broker_172.26.0.4_8099,numDocsScanned=657,scanInFilter=346815,scanPostFilter=657,sched=fcfs

also, i’m wondering what is considered prompts

"Slow query: …"

to show up in logs? does this mean pinot is suggesting that some optimization is possible to speed up my queries?

It is not caching, but cold start. Pinot mmaps index files, the OS needs to pull in pages the very first time.

understood. thanks!

Right now the default value to print

slow query

is when latency is

> 100ms

so its a hardcoded value. got it! also kind of mindblowing that you guys chose 100ms as the hardcoded value 😂

yeah, 100ms is

slow

for Pinot 😉