Enable caching on intermediate realtime persists

[xvrl]

No description provided.

[nishantmonu51]

unused import ?

[nishantmonu51]

do we need to iterate for the close to be called ? or may be simply replace it with a for loop.
Also, can we abstract the the namespace creation method.

[gianm]

The spills are not consistent across nodes so this would only work if the cache is local. That should probably be called out or enforced or something.

[gianm]

@xvrl is it possible to get real world performance numbers?

[xvrl]

@gianm addressed comments, will look into getting some real-word numbers.

[xvrl]

@gianm roughly 20% speedup on a typical top-n query 10-15min into the hour of a 1-hour realtime index task, with 10 persists completed (~ 1.4M rows to scan)

Unit: milliseconds
           expr      min       lq     mean   median       uq      max neval
  cache("true") 226.2689 269.5347 290.6316 279.4486 302.8039 542.1334   100
 cache("false") 285.9610 343.3580 395.8290 396.8076 419.2823 708.2887   100

[himanshug]

👍 since old behavior is preserved by default and this gets into action only if user explicitly provides the configuration.
not sure if documentation update is left out intentionally , if not then pls update same.

[xvrl]

@himanshug yes, will add docs, was waiting for comments before doing so