[GitHub] [incubator-druid] Eshcar commented on issue #8126: [Proposal] BufferAggregator support for growable sketches.

[GitBox <gi...@apache.org>]

Eshcar commented on issue #8126: [Proposal] BufferAggregator support for growable sketches.
URL: https://github.com/apache/incubator-druid/issues/8126#issuecomment-516411997
 
 
   As part of the work we are doing towards integrating Oak (off-heap based incremental index) integration into druid #5698, we invested some thought on how to bridge the gap between Oak--with its internal memory management, off-heap sketches--based on WritableMemory, and druid aggregators.
   I can share our thoughts, they aim to handle the same problems raised in this issue however the solution is different, hence it might be better to introduce it in a different issue.
   In a nutshell, 
   (1) Oak manages its memory and needs all allocations of buffers to go through the internal memory manager and only be exposed through Oak's API.
   (2) Off-heap sketches are based on WritableMemory, and can work with external memory manager that can allocate new WritableMemory when the sketch needs to grow
   (3) Druid aggregators access sketches through the aggregator API (init, aggregate, get) and a mapping from bytebuffer,position -> sketch
   
   What we suggest is to have oak manage the memory, including re-allocation of space when needed. Oak will implement its own WritebleMemory and MemoryRequestServer that are needed for a correct behaviour of the sketches wrt Oak index. Finally, we suggest to have a new Aggregator type - WritableMemoryAggregator that maps WritableMemory to sketch and can work the same way as buffer aggregators are working, and it does not need to worry about growing size of sketches. 
   
   There might be other alternatives for closing this loop; let's discuss them.
   
   Does all this make sense @leventov @himanshug ?

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@druid.apache.org
For additional commands, e-mail: commits-help@druid.apache.org