[jira] [Comment Edited] (ARROW-17373) [R] copying dataset and immediatly writing the copy to a different location fails

["Adam Black (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-17373?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17583184#comment-17583184 ] 

Adam Black edited comment on ARROW-17373 at 8/22/22 8:26 PM:
-------------------------------------------------------------

Here is another reprex. I think this only happens when the write location is the same as the current location of the open dataset. In the example above `writePath` and `savePath` are equal.

 


{{df <- data.frame(x = replicate(1,sample(0:1, 100e6, rep=TRUE)))}}
{{savePath <- file.path(tempdir(), 'arrowTest')}}
{{{}if (!dir.exists(savePath)){}}}{{{}{ dir.create(savePath) }{}}}

{{arrow::write_feather(df, file.path(savePath, 'part-0.feather'))}}

{{# writing a dataset to a new directory ('arrowTest2') works }}{{fine}}

{{writePath <- file.path(tempdir(), 'arrowTest2')}}

{{if (!dir.exists(writePath)) \{ dir.create(writePath) }}}

{{dataset <- arrow::open_dataset(savePath, format='feather')}}
{{nrow(dataset)}}
{{#> [1] 100000000}}

{{arrow::write_dataset(dataset = dataset, path = writePath, format = 'feather')}}
{{dataset2 <- arrow::open_dataset(writePath, format='feather')}}
{{nrow(dataset2)}}
{{#> [1] 100000000}}

{{trying to write an open dataset to it's own path gives an error}}
{{arrow::write_dataset(dataset = dataset2, path = writePath, format = 'feather')}}
{{{}#> Error: Invalid: Expected to read 144 metadata bytes but got 0{}}}{{{}# and it modifies the dataset{}}}
{{nrow(dataset2)}}
{{#> [1] 1966080}}

{{# But with a smaller dataset there seems to be no issue }}
{{arrow::write_dataset(dataset = head(dataset, 1000), path = savePath, format = 'feather')}}
{{dataset3 <- arrow::open_dataset(savePath, format='feather')}}
{{nrow(dataset3)}}
{{#> [1] 1000}}


was (Author: JIRAUSER289460):
Here is another reprex. I think this only happens when the write location is the same as the current location of the open dataset. In the example above `writePath` and `savePath` are equal.

 

``` r
df <- data.frame(x = replicate(1,sample(0:1, 100e6, rep=TRUE)))
savePath <- file.path(tempdir(), 'arrowTest')
if (!dir.exists(savePath)) {
  dir.create(savePath)
}
arrow::write_feather(df, file.path(savePath, 'part-0.feather'))

# writing a dataset to a new directory works fine
writePath <- file.path(tempdir(), 'arrowTest2')
if (!dir.exists(writePath)) {
  dir.create(writePath)
}
dataset <- arrow::open_dataset(savePath, format='feather')
nrow(dataset)
#> [1] 100000000

arrow::write_dataset(dataset = dataset, path = writePath, format = 'feather')
dataset2 <- arrow::open_dataset(writePath, format='feather')
nrow(dataset2)
#> [1] 100000000

# trying to write an open dataset to it's own path gives an error
arrow::write_dataset(dataset = dataset2, path = writePath, format = 'feather')
#> Error: Invalid: Expected to read 144 metadata bytes but got 0

# and it modifies the dataset
nrow(dataset2)
#> [1] 1966080

# But with a smaller dataset there seems to be no issue 
arrow::write_dataset(dataset = head(dataset, 1000), path = savePath, format = 'feather')
dataset3 <- arrow::open_dataset(savePath, format='feather')
nrow(dataset3)
#> [1] 1000
```

<sup>Created on 2022-08-22 with [reprex v2.0.2](https://reprex.tidyverse.org)</sup>

> [R] copying dataset and immediatly writing the copy to a different location fails
> ---------------------------------------------------------------------------------
>
>                 Key: ARROW-17373
>                 URL: https://issues.apache.org/jira/browse/ARROW-17373
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: R
>    Affects Versions: 9.0.0
>         Environment: Ubuntu 22.04
>            Reporter: Egill Axfjord Fridgeirsson
>            Priority: Major
>
> When I copy large feather files, open a dataset from that file and immediately write that dataset to a new location I get the following error:
>  
> ```Error: Invalid: Expected to read 144 metadata bytes but got 0```
>  
> I have made a reproducible example below:
>  
> ``` r
> df <- data.frame(replicate(1,sample(0:1,100e6,rep=TRUE)))
> savePath <- file.path(tempdir(), 'arrowTest')
> if (!dir.exists(savePath)) {
>   dir.create(savePath)
> }
> arrow::write_feather(df, file.path(savePath, 'part-0.feather'))
> copyPath <- file.path(tempdir(),'arrowTest')
> if (!dir.exists(copyPath)) {
>   dir.create(copyPath)
> }
> writePath <- file.path(tempdir(), 'arrowTest')
> if (!dir.exists(writePath)) {
>   dir.create(writePath)
> }
> arrow::copy_files(savePath, copyPath)
> dataset <- arrow::open_dataset(copyPath, format='feather')
> arrow::write_dataset(dataset = dataset, path = writePath, format = 'feather')
> ```



--
This message was sent by Atlassian Jira
(v8.20.10#820010)