apache / parquet-cpp / 802 / 1

Ran 03 Sep 2016 03:18PM UTC

Files 111

Run time 4s

Badge

Embed ▾

Committed 03 Sep 2016 03:10PM UTC coverage: 97.144% (+0.003%) from 97.141%

Job # 802.1

Build Type

push

travis-ci

Committed by

wesm

Commit Message

PARQUET-676: Fix incorrect MaxBufferSize for small bit widths

This function was implemented incorrectly in the original source code from Impala. The bug never presented itself because Impala allocates a much larger buffer than is required for the data page rather than using the `MaxBufferSize` value.

In a worst case scenario, the RleEncoder can result in a concatenation of short RLE runs of 8 values (the minimum for RLE). For example:

0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 ...

In encoded form, each RLE run occupies 2 bytes (for bit width 1). Thus 1024 values with this structure and bit width 1, at least 256 bytes are required to fully encode. If these were encoded as all literal runs, you can encode up to 504 values in a literal run (plus the indicator byte). Thus, in the same case (bit width 1) only

(504 + 504 + 16)
64 + 64 + 3 = 131

bytes are required. This should also fix PARQUET-698.

Author: Wes McKinney <wes.mckinney@twosigma.com>

Closes #150 from wesm/PARQUET-676 and squashes the following commits:

6ca44a8 [Wes McKinney] clang-format
e48b0bf [Wes McKinney] Fix maximum buffer sizing for hybrid RLE encoding case where short RLE runs use more space than mostly literal runs

Run Details

8402 of 8649 relevant lines covered (97.14%)

11159.2 hits per line

apache / parquet-cpp / 802 / 1

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 802.1

apache / parquet-cpp / 802 / 1

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 802.1

README BADGES
x