impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dan Hecht (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-3202: DiskIoMgr improvements for new buffer pool
Date Tue, 25 Oct 2016 20:53:10 GMT
Dan Hecht has posted comments on this change.

Change subject: IMPALA-3202: DiskIoMgr improvements for new buffer pool

Patch Set 1:

File be/src/exec/hdfs-scan-node-base.h:

PS1, Line 193: BufferOpts
const& ?
File be/src/runtime/

PS1, Line 642: without the client providing a "
             :           "buffer"
is this going to make sense to a user?

Line 708:   DCHECK_LE(buffer_size, max_buffer_size_);
how do we ensure this is true? is it the check in DiskIoMgr::Read() that will always enforce?
File be/src/runtime/disk-io-mgr.h:

PS1, Line 147: these buffers are returned to the caller wrapped
             : /// in BufferDescriptors.
took a few seconds to parse and understand. how about:

... these buffers are wrapped in BufferDescriptors and return to the caller.

Line 150: /// another read. In error cases, the IoMgr will recycle the buffers more promptly
is this relevant for all cases, or only cases (a) and (b)?

PS1, Line 392: bool
maybe make all of these members 'const' since they are "options" that don't change, right?

PS1, Line 422: BufferOpts
const BufferOpts& ?

Line 506:     int64_t client_buffer_len_;
maybe move this right before cached_buffer_ since it's related in the sense that it's another
possible source of the buffer.  Alternatively, consider even making a union to make it clear
that they are mutually exclusive, and the union tag indicates the buffer "mode" (Managed,
HdfsCached, Client - or something like that).  Though it could be that doesn't work out cleanly
given that HdfsCached can fallback to Managed mode.

To view, visit
To unsubscribe, visit

Gerrit-MessageType: comment
Gerrit-Change-Id: I913fbb8ae6c242f1585992eb2a622509344dccec
Gerrit-PatchSet: 1
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Tim Armstrong <>
Gerrit-Reviewer: Dan Hecht <>
Gerrit-Reviewer: Michael Ho
Gerrit-HasComments: Yes

View raw message