kylebarron / parquet-wasm

Rust-based WebAssembly bindings to read and write Apache Parquet data
https://kylebarron.dev/parquet-wasm/
Apache License 2.0
481 stars 19 forks source link

get wrong result when use `columns` in option #523

Open simline opened 2 months ago

simline commented 2 months ago

v0.6, test with await ParquetFile.from[File/Url]

such as tpch lintitem sf1: without columns in option: l_extendedprice returns with 3 of 0 between two number with 4x typedarray length

pick l_extendedprice l_returnflag, will get worng values: l_extendedprice returns with 0 between two number with same typedarray length l_returnflag returns values as index

kylebarron commented 2 months ago

This is not a reproducible example. A reproducible example with code is required to debug bugs.

simline commented 2 months ago

With columns:

> function meminfo() {
...     global.gc()
...     const memoryUsage = process.memoryUsage();
...     console.log('RSS:', memoryUsage.rss / (1024 * 1024), 'MB');
...     console.log('Heap Total:', memoryUsage.heapTotal / (1024 * 1024), 'MB');
...     console.log('Heap Used:', memoryUsage.heapUsed / (1024 * 1024), 'MB');
... }
undefined
>     
undefined
> const {ParquetFile, readParquet, readParquetStream, wasmMemory}       = require("parquet-wasm")
undefined
> const {parseTable, parseRecordBatch}                                                  = require("arrow-js-ffi")
undefined
> var           WASM_MEMORY                                                                                             = wasmMemory();
undefined
> 
> meminfo()
RSS: 70.890625 MB
Heap Total: 17.921875 MB
Heap Used: 7.547981262207031 MB
undefined
> 
> console.time('pq')
undefined
> var table             = await ParquetFile.fromUrl('http://localhost:8000/lineitem.parquet')
undefined
> var numRows           = table.metadata().fileMetadata().numRows()
undefined
> var option            = {
...     columns     : ['l_quantity', 'l_extendedprice', 'l_discount', 'l_tax', 'l_returnflag', 'l_linestatus', 'l_shipdate'],
...     batchSize   : 122_880,
... }
undefined
> meminfo()
RSS: 95.42578125 MB
Heap Total: 14.0703125 MB
Heap Used: 10.886177062988281 MB
undefined
> 
> var ffiTable  = (await table.read(option)).intoFFI();
undefined
> meminfo()
RSS: 641.65234375 MB
Heap Total: 14.0703125 MB
Heap Used: 10.615234375 MB
undefined
> 
> var arrowTable        = parseTable(
...     WASM_MEMORY.buffer,
...     ffiTable.arrayAddrs(),
...     ffiTable.schemaAddr()
... )
undefined
> console.timeEnd('pq')
pq: 2.231s
undefined
> meminfo()
RSS: 1097.15234375 MB
Heap Total: 17.3203125 MB
Heap Used: 11.910774230957031 MB
undefined
> 
> console.log(numRows, table.metadata().fileMetadata().createdBy());
6001215 DuckDB
undefined
> arrowTable.batches[0].data.children[4]
Data {
  type: Decimal { typeId: 7, scale: 2, precision: 15, bitWidth: 128 },
  children: [],
  dictionary: undefined,
  offset: 0,
  length: 122880,
  _nullCount: 0,
  stride: 4,
  values: Uint32Array(491520) [
     0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11,
    12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
    24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
    36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47,
    48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59,
    60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71,
    72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83,
    84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95,
    96, 97, 98, 99,
    ... 491420 more items
  ],
  nullBitmap: Uint8Array(0) []
}

Without Columns:

> console.time('pq')
undefined
> var table             = await ParquetFile.fromUrl('http://localhost:8000/lineitem.parquet')
undefined
> var numRows           = table.metadata().fileMetadata().numRows()
undefined
> var option            = {
...     // columns     : ['l_quantity', 'l_extendedprice', 'l_discount', 'l_tax', 'l_returnflag', 'l_linestatus', 'l_shipdate'],
...     batchSize   : 122_880,
... }
undefined
> meminfo()
RSS: 1099.27734375 MB
Heap Total: 14.3203125 MB
Heap Used: 11.8231201171875 MB
undefined
> 
> var ffiTable  = (await table.read(option)).intoFFI();
undefined
> meminfo()
RSS: 2150.1640625 MB
Heap Total: 15.0703125 MB
Heap Used: 11.835853576660156 MB
undefined
> 
> var arrowTable        = parseTable(
...     WASM_MEMORY.buffer,
...     ffiTable.arrayAddrs(),
...     ffiTable.schemaAddr()
... )
undefined
> console.timeEnd('pq')
pq: 5.992s
undefined
> meminfo()
RSS: 3131.0390625 MB
Heap Total: 17.8203125 MB
Heap Used: 11.851348876953125 MB
undefined
> 
> console.log(numRows, table.metadata().fileMetadata().createdBy());
6001215 DuckDB
undefined
> arrowTable.batches[0].data.children[9]
Data {
  type: Utf8 { typeId: 5 },
  children: [],
  dictionary: undefined,
  offset: 0,
  length: 122880,
  _nullCount: 0,
  stride: 1,
  valueOffsets: Int32Array(122881) [
     0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11,
    12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
    24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
    36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47,
    48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59,
    60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71,
    72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83,
    84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95,
    96, 97, 98, 99,
    ... 122781 more items
  ],
  values: Uint8Array(122880) [
    79, 79, 79, 79, 79, 79, 79, 70, 70, 70, 70, 70,
    70, 79, 70, 70, 70, 70, 79, 79, 79, 79, 79, 79,
    79, 79, 79, 79, 79, 79, 79, 70, 70, 70, 70, 79,
    79, 79, 79, 79, 79, 79, 79, 79, 79, 70, 70, 70,
    79, 79, 79, 79, 79, 79, 79, 70, 70, 79, 79, 70,
    70, 79, 79, 79, 79, 79, 79, 79, 79, 79, 79, 79,
    79, 79, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70,
    70, 70, 79, 79, 79, 79, 79, 79, 70, 70, 70, 70,
    70, 70, 70, 70,
    ... 122780 more items
  ],
  nullBitmap: Uint8Array(0) []
}
simline commented 2 months ago

the bug might be:

options specified columns, but when in get step, the field schema still use the one without columns

the default column postion 4 is:

  Field {
    name: 'l_quantity',
    type: Decimal { typeId: 7, scale: 2, precision: 15, bitWidth: 128 },
    nullable: true,
    metadata: Map(0) {}
  }

with specified columns at position 4, but the real position of whole arrowTable.schema.fields at postion 9:

  Field {
    name: 'l_linestatus',
    type: Utf8 { typeId: 5 },
    nullable: true,
    metadata: Map(0) {}
  }
kylebarron commented 2 months ago

I'm sorry I still don't understand what your issue is. Are you able to provide a reproducible data file along with your code? Can you remove all the memory reporting that is irrelevant to this issue, and just include the minimum amount of code to show your issue? See: https://stackoverflow.com/help/minimal-reproducible-example

kylebarron commented 2 months ago

Also note that I believe the ordering of the names in columns is not taken into effect. That only determines which columns to select, but the original order in the dataset is used.

simline commented 2 months ago

Also note that I believe the ordering of the names in columns is not taken into effect. That only determines which columns to select, but the original order in the dataset is used.

yes,default postions are 0~15,columns option pick 4~10 with same order, not work

and the tpch lienitem parquet, you can generate from duckdb

kylebarron commented 2 months ago

yes,default postions are 0~15,columns option pick 4~10 with same order, not work

I still don't know what you're saying doesn't work. It looks like you're able to access the data correctly. You need to check the schema to find the positions of the columns in the output data.

and the tpch lienitem parquet, you can generate from duckdb

I am not going to go out of my way to generate data without a reproducible example. Please supply a minimal, reproducible example or I'll close this issue.