Skip to content

Latest commit

 

History

History
64 lines (47 loc) · 4.05 KB

VeloxS3.md

File metadata and controls

64 lines (47 loc) · 4.05 KB
Raw
layout title nav_order parent
page
Using S3 with Gluten
3
Getting-Started

Object stores offered by CSPs such as AWS S3 are important for users of Gluten to store their data. This doc will discuss all details of configs, and use cases around using Gluten with object stores. In order to use an S3 endpoint as your data source, please ensure you are using the following S3 configs in your spark-defaults.conf. If you're experiencing any issues authenticating to S3 with additional auth mechanisms, please reach out to us using the 'Issues' tab.

Working with S3

Configuring S3 endpoint

S3 proivdes the endpoint based method to access the files, here's the example configuration. Users may need to modify some values based on real setup.

spark.hadoop.fs.s3a.impl                        org.apache.hadoop.fs.s3a.S3AFileSystem
spark.hadoop.fs.s3a.aws.credentials.provider    org.apache.hadoop.fs.s3a.SimpleAWSCredentialsProvider
spark.hadoop.fs.s3a.access.key                  XXXXXXXXX
spark.hadoop.fs.s3a.secret.key                  XXXXXXXXX
spark.hadoop.fs.s3a.endpoint                    https://s3.us-west-1.amazonaws.com
spark.hadoop.fs.s3a.connection.ssl.enabled      true
spark.hadoop.fs.s3a.path.style.access           false

Note if testing with a mock AWS S3 environment(like Minio/Ceph), users may required to modify some of the values. E.g., on Minio setup, below config is required: spark.hadoop.fs.s3a.path.style.access true

Configuring S3 instance credentials

S3 also provides other methods for accessing, you can also use instance credentials by setting the following config

spark.hadoop.fs.s3a.use.instance.credentials true

Note that in this case, "spark.hadoop.fs.s3a.endpoint" won't take affect as Gluten will use the endpoint set during instance creation.

Configuring S3 IAM roles

You can also use iam role credentials by setting the following configurations. Instance credentials have higher priority than iam credentials.

spark.hadoop.fs.s3a.iam.role  xxxx
spark.hadoop.fs.s3a.iam.role.session.name xxxx

Note that spark.hadoop.fs.s3a.iam.role.session.name is optional.

Other authentatication methods are not supported yet

Local Caching support

Velox supports a local cache when reading data from HDFS/S3. The feature is very useful if remote storage is slow, e.g., reading from a public S3 bucket and stronger performance is desired. With this feature, Velox can asynchronously cache the data on local disk when reading from remote storage, and the future reading requests on already cached blocks will be serviced from local cache files. To enable the local caching feature, below configurations are required:

spark.gluten.sql.columnar.backend.velox.cacheEnabled      // enable or disable velox cache, default false.
spark.gluten.sql.columnar.backend.velox.memCacheSize      // the total size of in-mem cache, default is 128MB.
spark.gluten.sql.columnar.backend.velox.ssdCachePath      // the folder to store the cache files, default is "/tmp".
spark.gluten.sql.columnar.backend.velox.ssdCacheSize      // the total size of the SSD cache, default is 128MB. Velox will do in-mem cache only if this value is 0.
spark.gluten.sql.columnar.backend.velox.ssdCacheShards    // the shards of the SSD cache, default is 1.
spark.gluten.sql.columnar.backend.velox.ssdCacheIOThreads // the IO threads for cache promoting, default is 1. Velox will try to do "read-ahead" if this value is bigger than 1 
spark.gluten.sql.columnar.backend.velox.ssdODirect        // enbale or disable O_DIRECT on cache write, default false.

It's recommended to mount SSDs to the cache path to get the best performance of local caching. On the start up of Spark context, the cache files will be allocated under "spark.gluten.sql.columnar.backend.velox.cachePath", with UUID based suffix, e.g. "/tmp/cache.13e8ab65-3af4-46ac-8d28-ff99b2a9ec9b0". Gluten is not able to reuse older caches for now, and the old cache files are left there after Spark context shutdown.