Using Allas with Rclone
This chapter contains instructions for using Allas with Rclone in the Puhti and Mahti computing environments. Rclone provides a very powerful and versatile way to use Allas and other object storage services. It is able to use both the S3 and Swift protocols (and many others), but in the case of Allas, the Swift protocol is preferred. It is also the default option on the CSC servers.
WARNING: Rclone should not be used to copy, move or rename objects inside Allas. Rclone provides commands for these operations but they don't work correctly for files larger than 5 GB.
WARNING: If a rclone data upload process for a over 5 GB file gets interrupted, please remove the partially uploaded object before restarting the upload process. Otherwise rclone sometimes reports a successful data upload even though not all data has been copied to Allas.
The basic syntax of Rclone:
rclone subcommand optons source:path dest:path
The most frequently used Rclone commands:
- rclone copy – Copy files from the source to the destination, skipping what has already been copied.
- rclone sync – Make the source and destination identical, modifying only the destination.
- rclone move – Move files from the source to the destination.
- rclone delete – Remove the contents of a path.
- rclone mkdir – Create the path if it does not already exist.
- rclone rmdir – Remove the path.
- rclone check – Check if the files in the source and destination match.
- rclone ls – List all objects in the path, including size and path.
- rclone lsd – List all directories/containers/buckets in the path.
- rclone lsl – List all objects in the path, including size, modification time and path.
- rclone lsf – List the objects using the virtual directory structure based on the object names.
- rclone cat – Concatenate files and send them to stdout.
- rclone copyto – Copy files from the source to the destination, skipping what has already been copied.
- rclone moveto – Move the file or directory from the source to the destination.
- rclone copyurl – Copy the URL's content to the destination without saving it in the tmp storage.
A more extensive list can be found on the Rclone manual pages or by typing the command
rclone in Puhti.
The first step is to authenticate to a project in Allas. Rclone can use both Swift and S3 protocols but these connections will have different names in rclone commands.
In this document we describe how Rclone is used in CSC computing environment (Puhti and Mahti). You can use rclone also in your local computer. Intructions of configuring locally installed Rclone are here
Rclone with swift on CSC supercomputers
The default protocol of Allas is Swift. In Puhti and Mahti Swift based Allas connection is activated with commands:
module load allas allas-conf
allas-confcommand asks for your CSC password (the same you use to login to CSC servers). It lists your projects in Allas and asks you to define the project that will be used. Then allas-conf generates a Rclone configuration file for the Allas service and authenticates the connection to the selected project. In Rclone command this swift based connection is referred with remote name
allas:. The authentication information is stored in the shell variables
OS_STORAGE_URLthat are valid for up to eight hours. However, you can refresh the authentication at any time by running allas-conf again. The environment variables are available only for that login session, so if you login to Puhti in another session, you need to authenticate again to access Allas.
Rclone with S3 on CSC supercomputers
If you want to use Allas with S3 protocol in stead run the allas-conf command with option --mode s3cmd.
module load allas allas-conf --mode s3cmd
s3allas:. In the examples below the swift based
allas:remote definition is used, but if you have S3 connection defined, you could replace it with
s3allas:. Note that you can have both
s3allas:functional in the same time and that they can still use different Allas projects. However, you should avoid mixing protocols. If an object is loaded using
allas:do also all operations with
Create buckets and upload objects
The data in Allas is arranged into containers called buckets. You can consider them top level directories. All buckets in Allas must have unique names – you cannot create a bucket if some other project has already used that bucket name. It is a good rule of thumb to have something project- or user-specific in the bucket name, e.g. 2000620-raw-data. See the checklist for how to name a bucket.
In the case of Rclone, create a bucket:
rclone mkdir allas:2000620-raw-data
rclone copy file.dat allas:2000620-raw-data/
rclone moveinstead of
rclone copy, the local version of the uploaded file (file.dat) is deleted after copying.
The copy and move subcommands only work with files. If you would like to copy all files in a directory, use the copyto or moveto subcommands.
During upload files that are larger than 5 GB will be split and stored as several objects. This is done automatically to a bucket that is named by adding extension
_segments to the original bucket name. For example, if you would upload a large file to 2000620-raw-data the actual data would be stored as several pieces into bucket 2000620-raw-data_segments. The target bucket (2000620-raw-data) would contain just a front object that contains information what segments make the stored file. Operations performed to the front object are automatically reflected to the segments. Normally users don't need to operate with the segments buckets at all and objects inside these buckets should not be deleted or modified.
List buckets and objects
List all the buckets belonging to a project:
rclone lsd allas: 0 2019-06-06 14:43:40 0 2000620-raw-data
List the content of a bucket:
rclone ls allas:2000620-raw-data 677972 file.dat
Use the same
rclone copy and
rclone copyto commands to download a file:
rclone copy allas:2000620-raw-data/file.dat
If you include a destination parameter in the download command, Rclone creates a directory for the download:
rclone copy allas:2000620-raw-data/file.dat doh
ls doh file.dat
ls -ld doh drwxr-xr-x 3 user staff 96 Jun 6 14:58 doh
Synchronizing a directory
One way of moving data between Allas and the computing environment is synchronization. The difference between copying and synchronizing is that while copying only adds new objects or files from the source to the destination, synchronization can also remove data from the destination, in order to make the destination match the source. This feature makes synchronization very effective but also potentially very dangerous.
For example, a folder named mydata has the following structure:
ls -R mydata mydata/: file1.txt setA setB mydata/setA: file2.txt mydata/setB: file3.txt file4.txt
An example of using sync (note that the destination parameter requires the folder name (mydata)):
rclone sync mydata allas:2000620-raw-data/mydata
rclone ls allas:2000620-raw-data 677972 mydata/file1.txt 10927 mydata/setA/file2.txt 1116 mydata/setB/file3.txt 5075 mydata/setB/file4.txt
Let us assume that we are storing new data (file5.txt and file6.txt) in the subdirectory mydata/setC and simultaneously removing the file mydata/setB/file3.txt. When the rclone sync command is executed again, the new data is added to Allas and the object mydata/setB/file3.txt is removed.
rclone sync mydata allas:2000620-raw-data/mydata rclone ls allas:2000620-raw-data 677972 mydata/file1.txt 10927 mydata/setA/file2.txt 5075 mydata/setB/file4.txt 1265 mydata/setC/file5.txt 4327 mydata/setC/file6.txt
In the examples above, Allas has been used as the destination that is changed. However, the command can be used in the reverse direction as well:
rclone sync allas:2000620-raw-data/mydata mydata
This command returns the uploaded data from Allas to the mydata directory. Note however that if you have added new data to mydata after synchronizing the directory with Allas, this data will be erased.