Tim Armstrong has posted comments on this change.

Change subject: IMPALA-3200: Implement suballocator for splitting buffers
......................................................................


Patch Set 4:

(3 comments)

Sorry for the slow turnaround, got caught up in things.

http://gerrit.cloudera.org:8080/#/c/4715/4//COMMIT_MSG
Commit Message:

Line 14: larger chunks. This helps avoid fragmentation and is quite effective
> I think there are a few ways to get both what I am concerned about (buddy a
Gotcha.

I added the static_assert - that is a good idea.

I think the Atom idea is nice but I think we end up with the same problem with 
large buffers, since the buffer pool only supports powers-of-two.


http://gerrit.cloudera.org:8080/#/c/4715/11/be/src/bufferpool/suballocator-test.cc
File be/src/bufferpool/suballocator-test.cc:

Line 107:   // Make allocations smaller than the buffer size.
> What happened to the const?
The const was incorrect - I pushed a patchset without the fix. The problem is 
that GetUsedReservation() is a non-const method since it acquires the mutex.


http://gerrit.cloudera.org:8080/#/c/4715/4/be/src/bufferpool/suballocator.h
File be/src/bufferpool/suballocator.h:

Line 171:   Suballocation* prev_free_;
> Just so I'm clear, in the current system, a Suballocation may have several 
Yep, that's true. I realise that it's really not obvious why the lifetime of 
'buddy_' and 'prev_free_' is shorter than the unique_ptr. I documented that and 
then described the ownership cases in the class header.


-- 
To view, visit http://gerrit.cloudera.org:8080/4715
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I8bfe0e429f67ad273f7c7d0816703a9e6c3da788
Gerrit-PatchSet: 4
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Tim Armstrong <[email protected]>
Gerrit-Reviewer: Jim Apple <[email protected]>
Gerrit-Reviewer: Michael Ho
Gerrit-Reviewer: Michael Ho <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-HasComments: Yes

Reply via email to