spark-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Wenchen Fan <cloud0...@gmail.com>
Subject Re: Error while reading hive tables with tmp/hidden files inside partitions
Date Thu, 23 Apr 2020 11:28:51 GMT
Yea, please report the bug on a supported Spark version like 2.4.

On Thu, Apr 23, 2020 at 3:40 PM Dhrubajyoti Hati <dhruba.work@gmail.com>
wrote:

> FYI we are using Spark 2.2.0. Should the change be present in this spark
> version? Wanted to check before opening a JIRA ticket?
>
>
>
>
> *Regards,Dhrubajyoti Hati.*
>
>
> On Thu, Apr 23, 2020 at 10:12 AM Wenchen Fan <cloud0fan@gmail.com> wrote:
>
>> This looks like a bug that path filter doesn't work for hive table
>> reading. Can you open a JIRA ticket?
>>
>> On Thu, Apr 23, 2020 at 3:15 AM Dhrubajyoti Hati <dhruba.work@gmail.com>
>> wrote:
>>
>>> Just wondering if any one could help me out on this.
>>>
>>> Thank you!
>>>
>>>
>>>
>>>
>>> *Regards,Dhrubajyoti Hati.*
>>>
>>>
>>> On Wed, Apr 22, 2020 at 7:15 PM Dhrubajyoti Hati <dhruba.work@gmail.com>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> Is there any way to discard files starting with dot(.) or ending with
>>>> .tmp in the hive partition while reading from Hive table using
>>>> spark.read.table method.
>>>>
>>>> I tried using PathFilters but they didn't work. I am using spark-submit
>>>> and passing my python file(pyspark) containing the source code.
>>>>
>>>> spark.sparkContext._jsc.hadoopConfiguration().set("mapreduce.input.pathFilter.class",
>>>> "com.abc.hadoop.utility.TmpFileFilter")
>>>>
>>>> class TmpFileFilter extends PathFilter {
>>>>   override def accept(path : Path): Boolean = !path.getName.endsWith(".tmp")
>>>> }
>>>>
>>>> Still in the detailed logs I can see .tmp files are getting considered
>>>> in the detailed logs:
>>>> 20/04/22 12:58:44 DEBUG MapRFileSystem: getMapRFileStatus
>>>> maprfs:///a/hour=05/host=abc/FlumeData.1587559137715
>>>> 20/04/22 12:58:44 DEBUG MapRFileSystem: getMapRFileStatus
>>>> maprfs:///a/hour=05/host=abc/FlumeData.1587556815621
>>>> 20/04/22 12:58:44 DEBUG MapRFileSystem: getMapRFileStatus
>>>> maprfs:///a/hour=05/host=abc/.FlumeData.1587560277337.tmp
>>>>
>>>>
>>>> Is there any way to discard the tmp(.tmp) or the hidden files(filename
>>>> starting with dot or underscore) in hive partitions while reading from
>>>> spark?
>>>>
>>>>
>>>>
>>>>
>>>> *Regards,Dhrubajyoti Hati.*
>>>>
>>>

Mime
View raw message