This section describes how you interact through the
|D s platform|
- HDFS is a scalable file storage system for use across all of the nodes (servers) of a Hadoop cluster. Many interactions with HDFS are similar with desktop interactions with files and folders. However, what looks like a "file" or "folder" in HDFS may be spread across multiple nodes in the cluster. For more information, see https://en.wikipedia.org/wiki/Apache_Hadoop#HDFS.
Uses of HDFS
|D s platform|
- Creating Datasets from HDFS Files: You can read in from a data source stored in HDFS. A source may be a single HDFS file or a folder of identically structured files. See Reading from Sources in HDFS below.
- Reading Datasets: When creating a dataset, you can pull your data from another dataset defined in HDFS. See Creating Datasets below.
- Writing Job Results: After a job has been executed, you can write the results back to HDFS. See Writing Job Results below.
|D s webapp|
NOTE: When the
Before You Begin Using HDFS
Read/Write Access: Your Hadoop administrator must configure read/write permissions to locations in HDFS. Please see the HDFS documentation provided with your Hadoop distribution.
/trifacta/uploadsfor reading and writing data. This directory is used by the
D s webapp
Your Hadoop administrator should provide a place or mechanism for raw data to be uploaded to your Hadoop datastore.
Your Hadoop administrator should provide a writeable home output directory for you, which you can review. See Storage Config Page.
Depending on the security features you've enabled, the technical methods by which
|D s item|
Storing Data in HDFS
Your Hadoop administrator should provide raw data or locations and access for storing raw data within HDFS. All
|D s item|
- Users should know where shared data is located and where personal data can be saved without interfering with or confusing other users.
If JDBC ingest caching has been enabled, users may see a
dataSourceCache folder in their browser. This folder is used to store per-user caches of JDBC-based data that has been ingested into the platform from its source.
NOTE: The datasourceCache folder should not be used for reading and writing of datasets, metadata, or results.
For more information, see Configure JDBC Ingestion.
Reading from Sources in HDFS
You can create a dataset from one or more files stored in HDFS.
|D s dataadmin role|
You can parameterize your input paths to import source files as part of the same imported dataset. For more information, see Overview of Parameterization.
When you select a folder in HDFS to create your dataset, you select all files in the folder to be included. Notes:
- This option selects all files in all sub-folders. If your sub-folders contain separate datasets, you should be more specific in your folder selection.
- All files used in a single dataset must be of the same format and have the same structure. For example, you cannot mix and match CSV and JSON files if you are reading from a single directory.
- When a folder is selected from HDFS, the following file types are ignored:
*_FAILEDfiles, which may be present if the folder has been populated by Hadoop.
- If you have stored files in HDFS that begin with an underscore (
_), these files cannot be read during batch transformation and are ignored. Please rename these files through HDFS so that they do not begin with an underscore.
When creating a dataset, you can choose to read data in from a source stored from HDFS or from a local file.
- HDFS sources are not moved or changed.
- Local file sources are uploaded to
/trifacta/uploadswhere they remain and are not changed.
Data may be individual files or all of the files in a folder. For more information, see Reading from Sources in HDFS.
- In the Import Data page, click the HDFS tab. See Import Data Page.
Writing Job Results
When your job results are generated, they can be stored back in HDFS for you at the location defined for your user account.
- The HDFS location is available through the Output Destinations tab of the Job Details page. See Job Details Page.
- Each set of job results must be stored in a separate folder within your HDFS output home directory.
- For more information on your output home directory, see Storage Config Page.
If your deployment is using HDFS, do not use the
NOTE: Users can specify a default output home directory and, during job execution, an output directory for the current job. In an encrypted HDFS environment, these two locations must be in the same encryption zone. Otherwise, writing the job results fails with a
Access to results:
Depending on how the platform is integrated with HDFS, other users may or may not be able to access your job results.
If user impersonation is enabled, results are written to HDFS through the HDFS account configured for your use. Depending on the permissions of your HDFS account, you may be the only person who can access these results.
- If user impersonation is not enabled, then each
writes results to HDFS using a shared account. Depending on the permissions of that account, your results may be visible to all platform users.
D s item user user
Creating a new dataset from results
As part of writing job results, you can choose to create a new dataset, so that you can chain together data wrangling tasks.
NOTE: When you create a new dataset as part of your job results, the file or files are written to the designated output location for your user account. Depending on how your Hadoop permissions are configured, this location may not be accessible to other users.
Other than temporary files, the
|D s platform|
- Uploaded datasets
- Generated samples
- Generated results
If you are concerned about data accumulation, please contact your HDFS administrator.