[native] Add support for cache periodic full persistence

[zacw7]

Description

Currently, maxWriteRatio(AsyncDataCache::Options), the maximum ratio of the number of in-memory cache entries being written to the SSD cache over the total number of cache entries, is set to the default value of 0.7, meaning that we only persist 70% of cache entries to the SSD and upon server restart 30% of cache will be lost.

This change adds an optional periodic daemon to flush cache data to ssd. This can help to make sure all the in-memory data is flushed to ssd.

== RELEASE NOTES ==

General Changes
* Add support for persisting full memory cache to SSD periodically on Presto C++ worker. This can be enabled by setting `cache.velox.full-persistence-interval` to a non-zero value. :pr:`23192`

[aditi-pandit]

Thanks @zacw7.

Please can you add a release note for this functionality. Would greatly help users.

[zacw7]

Thanks @zacw7.

Please can you add a release note for this functionality. Would greatly help users.

release note added. thanks!

[steveburnett]

Nit, please add the PR number to the release note entry following the Release Notes Guidelines:

== RELEASE NOTES ==

General Changes
* Add support for persisting full memory cache to SSD periodically on Presto C++ worker. This can be enabled by setting `cache.velox.full-persistence-interval` to a non-zero value. :pr:`23192`

[tanjialiang]

is it possible to get ssd cache outside of the periodic func? Also if isCachePeriodicFullPersistenceEnabled() is true shouldn't cache and ssdcache both be non-null? Can we change these conditions to VELOX_CHECKs?

[zacw7]

If cache and ssdcache both are non-null, shall we just ignore it instead of failing the server hard? Similarly, we still allow the server operation api to be called even when there is no cache:

presto/presto-native-execution/presto_cpp/main/PrestoServerOperations.cpp

Line 351 in b73ab7d

std::string PrestoServerOperations::serverOperationWriteSsd(

[tanjialiang]

What I meant is something like this

if (isCachePeriodicFullPersistenceEnabled()) {
  VELOX_CHECK_NOT_NULL(cache);
  VELOX_CHECK_NOT_NULL(ssdCache);
}

Or are there any scenarios that they can actually be NULL when the if condition is true?

[tanjialiang]

Thanks @zacw7 for the revision. Some minors

[tanjialiang]

I think you need to capture this pointer by copy in the below lambda. Or you can simply put the SystemConfig::instance() call inside lambda. Yeah let's just move it inside lambda. There is not a strong need of having it outside.

[tanjialiang]

What I meant is something like this

if (isCachePeriodicFullPersistenceEnabled()) {
  VELOX_CHECK_NOT_NULL(cache);
  VELOX_CHECK_NOT_NULL(ssdCache);
}

Or are there any scenarios that they can actually be NULL when the if condition is true?

[xiaoxmeng]

saveToSsd(false)