3.0 KiB
MinIO HDFS Gateway
MinIO HDFS gateway adds Amazon S3 API support to Hadoop HDFS filesystem. Applications can use both the S3 and file APIs concurrently without requiring any data migration. Since the gateway is stateless and shared-nothing, you may elastically provision as many MinIO instances as needed to distribute the load.
NOTE: Intention of this gateway implementation it to make it easy to migrate your existing data on HDFS clusters to MinIO clusters using standard tools like
mc
oraws-cli
, if the goal is to use HDFS perpetually we recommend that HDFS should be used directly for all write operations.
Run MinIO Gateway for HDFS Storage
Using Binary
Namenode information is obtained by reading core-site.xml
automatically from your hadoop environment variables $HADOOP_HOME
export MINIO_ACCESS_KEY=minio
export MINIO_SECRET_KEY=minio123
minio gateway hdfs
You can also override the namenode endpoint as shown below.
export MINIO_ACCESS_KEY=minio
export MINIO_SECRET_KEY=minio123
minio gateway hdfs hdfs://namenode:8200
Using Docker
Using docker is experimental, most Hadoop environments are not dockerized and may require additional steps in getting this to work properly. You are better off just using the binary in this situation.
docker run -p 9000:9000 \
--name hdfs-s3 \
-e "MINIO_ACCESS_KEY=minio" \
-e "MINIO_SECRET_KEY=minio123" \
minio/minio gateway hdfs hdfs://namenode:8200
Test using MinIO Browser
MinIO gateway comes with an embedded web based object browser. Point your web browser to http://127.0.0.1:9000 to ensure that your server has started successfully.
Test using MinIO Client mc
mc
provides a modern alternative to UNIX commands such as ls, cat, cp, mirror, diff etc. It supports filesystems and Amazon S3 compatible cloud storage services.
Configure mc
mc config host add myhdfs http://gateway-ip:9000 access_key secret_key
List buckets on hdfs
mc ls myhdfs
[2017-02-22 01:50:43 PST] 0B user/
[2017-02-26 21:43:51 PST] 0B datasets/
[2017-02-26 22:10:11 PST] 0B assets/
Known limitations
Gateway inherits the following limitations of HDFS storage layer:
- No bucket policy support (HDFS has no such concept)
- No bucket notification APIs are not supported (HDFS has no support for fsnotify)
- No server side encryption support (Intentionally not implemented)
- No server side compression support (Intentionally not implemented)
- Concurrent multipart operations are not supported (HDFS lacks safe locking support, or poorly implemented)