|
8
|
1
|
* DAT/PARQUET |
|
9
|
2
|
:PROPERTIES: |
|
|
3
|
:ID: 657a645b-0fad-4f95-a022-cd837ce188d6 |
|
|
4
|
:END: |
|
8
|
5
|
https://github.com/apache/parquet-format |
|
|
6
|
https://github.com/apache/parquet-format/blob/master/src/main/thrift/parquet.thrift |
|
|
7
|
https://github.com/apache/parquet-testing |
|
|
8
|
https://github.com/apache/parquet-java |
|
|
9
|
** glossary |
|
9
|
10
|
:PROPERTIES: |
|
|
11
|
:ID: e71f388c-9ed1-4862-8890-7f74271e8df0 |
|
|
12
|
:END: |
|
8
|
13
|
- block :: same as HDFS block |
|
|
14
|
- file :: file metadata is required, data is not |
|
|
15
|
- row-group :: a logical horizontal partitioning of the data into |
|
|
16
|
rows. no physical rep is guaranteed for row-group |
|
|
17
|
- column-chunk :: a chunk of the data for a particular column |
|
|
18
|
- page :: column chunks are divided into pages. a page is conceptually |
|
|
19
|
indivisible in terms of compression/encoding. multiple page types |
|
|
20
|
can be interleaved in a column chunk. |
|
|
21
|
|
|
|
22
|
Files consists of 1+ row-groups. A row-group contains exactly one |
|
|
23
|
column chunk per column. Column chunks contain one or more pages. |
|
|
24
|
|
|
|
25
|
** format summary |
|
9
|
26
|
:PROPERTIES: |
|
|
27
|
:ID: ae54516c-c8a8-49f8-aac6-a95c18f5de8e |
|
|
28
|
:END: |
|
8
|
29
|
#+begin_example |
|
|
30
|
4-byte magic number "PAR1" |
|
|
31
|
<Column 1 Chunk 1> |
|
|
32
|
<Column 2 Chunk 1> |
|
|
33
|
... |
|
|
34
|
<Column N Chunk 1> |
|
|
35
|
<Column 1 Chunk 2> |
|
|
36
|
<Column 2 Chunk 2> |
|
|
37
|
... |
|
|
38
|
<Column N Chunk 2> |
|
|
39
|
... |
|
|
40
|
<Column 1 Chunk M> |
|
|
41
|
<Column 2 Chunk M> |
|
|
42
|
... |
|
|
43
|
<Column N Chunk M> |
|
|
44
|
File Metadata |
|
|
45
|
4-byte length in bytes of file metadata (little endian) |
|
|
46
|
4-byte magic number "PAR1" |
|
|
47
|
#+end_example |
