[
https://issues.apache.org/jira/browse/SPARK-31962?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Christopher Highman updated SPARK-31962:
----------------------------------------
Description:
When using structured streaming with a FileDataSource, I've encountered a
number of occasions where I want to be able to stream from a folder containing
any number of historical files in CSV format. When I start reading from a
folder, however, I might only care about files that were created after a
certain time.
{code:java}
spark.readStream
.option("header", "true")
.option("delimiter", "\t")
.format("csv")
.load("/mnt/Deltas")
{code}
In
[https://github.com/apache/spark/blob/f3771c6b47d0b3aef10b86586289a1f675c7cfe2/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InMemoryFileIndex.scala],
there is a method, _listLeafFiles,_ which builds FileStatus objects containing
an implicit _modificationDate_ property. We may already iterate the resulting
files if a filter is applied to the path. In this case, its trivial to do an
primitive comparison against modificationDate. Without the filter specified,
we would be expending less effort than if the filter were applied by itself.
Having the ability to provide an option specifying a timestamp by which to
begin globbing files would result in quite a bit of less complexity needed on a
consumer who leverages the ability to stream from a folder path but does not
have an interest in reading what could be thousands of files that are not
relevant.
One example to could be "filesModifiedAfterDate" accepting a UTC datetime like
below.
{code:java}
spark.readStream
.option("header", "true")
.option("delimiter", "\t")
.option("filesModifiedAfterDate", "2020-05-01 00:00:00")
.format("csv")
.load("/mnt/Deltas")
{code}
If this option is specified, the expected behavior would be that files within
the _"/mnt/Deltas/"_ path must have been created at or later than the specified
time in order to be consumed for purposes of reading files in general or for
purposes of structured streaming.
I have unit tests passing under _CSVSuite_ and _FileIndexSuite_ in the
_spark.sql.execution.datasources_ package.
was:
When using structured streaming with a FileDataSource, I've encountered a
number of occasions where I want to be able to stream from a folder containing
any number of historical delta files in CSV format. When I start reading from
a folder, however, I might only care about files were created after a certain
time.
{code:java}
spark.readStream
.option("header", "true")
.option("delimiter", "\t")
.format("csv")
.load("/mnt/Deltas")
{code}
In
[https://github.com/apache/spark/blob/f3771c6b47d0b3aef10b86586289a1f675c7cfe2/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSource.scala],
there is a method, _checkAndGlobPathIfNecessary,_ which appears create an
in-memory index of files for a given path. There may a rather clean
opportunity to consider options here.
Having the ability to provide an option specifying a timestamp by which to
begin globbing files would result in quite a bit of less complexity needed on a
consumer who leverages the ability to stream from a folder path but does not
have an interest in reading what could be thousands of files that are not
relevant.
One example to could be "createdFileTime" accepting a UTC datetime like below.
{code:java}
spark.readStream
.option("header", "true")
.option("delimiter", "\t")
.option("createdFileTime", "2020-05-01 00:00:00")
.format("csv")
.load("/mnt/Deltas")
{code}
If this option is specified, the expected behavior would be that files within
the _"/mnt/Deltas/"_ path must have been created at or later than the specified
time in order to be consumed for purposes of reading the files in general or
for purposes of structured streaming.
> Provide option to load files after a specified date when reading from a
> folder path
> -----------------------------------------------------------------------------------
>
> Key: SPARK-31962
> URL: https://issues.apache.org/jira/browse/SPARK-31962
> Project: Spark
> Issue Type: Improvement
> Components: SQL, Structured Streaming
> Affects Versions: 3.1.0
> Reporter: Christopher Highman
> Priority: Minor
>
> When using structured streaming with a FileDataSource, I've encountered a
> number of occasions where I want to be able to stream from a folder
> containing any number of historical files in CSV format. When I start
> reading from a folder, however, I might only care about files that were
> created after a certain time.
> {code:java}
> spark.readStream
> .option("header", "true")
> .option("delimiter", "\t")
> .format("csv")
> .load("/mnt/Deltas")
> {code}
> In
> [https://github.com/apache/spark/blob/f3771c6b47d0b3aef10b86586289a1f675c7cfe2/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InMemoryFileIndex.scala],
> there is a method, _listLeafFiles,_ which builds FileStatus objects
> containing an implicit _modificationDate_ property. We may already iterate
> the resulting files if a filter is applied to the path. In this case, its
> trivial to do an primitive comparison against modificationDate. Without the
> filter specified, we would be expending less effort than if the filter were
> applied by itself.
> Having the ability to provide an option specifying a timestamp by which to
> begin globbing files would result in quite a bit of less complexity needed on
> a consumer who leverages the ability to stream from a folder path but does
> not have an interest in reading what could be thousands of files that are not
> relevant.
> One example to could be "filesModifiedAfterDate" accepting a UTC datetime
> like below.
> {code:java}
> spark.readStream
> .option("header", "true")
> .option("delimiter", "\t")
> .option("filesModifiedAfterDate", "2020-05-01 00:00:00")
> .format("csv")
> .load("/mnt/Deltas")
> {code}
> If this option is specified, the expected behavior would be that files within
> the _"/mnt/Deltas/"_ path must have been created at or later than the
> specified time in order to be consumed for purposes of reading files in
> general or for purposes of structured streaming.
>
> I have unit tests passing under _CSVSuite_ and _FileIndexSuite_ in the
> _spark.sql.execution.datasources_ package.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
