Exasol BucketFS Skill

Trigger when the user mentions BucketFS, exapump, bucket, bfsdefault, upload to BucketFS, download from BucketFS, delete from BucketFS, BucketFS path, BucketFS file, or any BucketFS file management task.

BucketFS Concepts

BucketFS is a synchronous distributed file system available on all nodes of an Exasol cluster. Files stored in BucketFS are automatically replicated to every cluster node.

Key concepts:

• Service: A named BucketFS instance. The default service is bfsdefault.
• Bucket: A storage container within a service. The default bucket is default.
• Path inside BucketFS: Files are referenced by the path within the bucket (e.g., models/my_model.pkl).
• Local path inside UDFs: Files are accessible at /buckets/<service>/<bucket>/<path> (e.g., /buckets/bfsdefault/default/models/my_model.pkl).

Important characteristics:

• Writes are atomic — a file is either fully written or not at all.
• No transactions and no file locks; the latest write wins.
• All nodes see identical content after synchronisation.
• BucketFS is not included in database backups — manage backups separately.
• Not suited for very large datasets due to replication overhead.

exapump CLI

The exapump command is the CLI tool for managing BucketFS. All BucketFS operations use the exapump bucketfs subcommand.

Connection Configuration

Connection settings are stored in ~/.exapump/config.toml as named profiles. Example:

[production]
host = "exasol-prod.example.com"
user = "admin"
password = "s3cret"
default = true
bfs_write_password = "bucketpw"
bfs_read_password = "bucketpw"

Key profile fields:

Connection parameters can also be overridden per command via CLI flags (highest priority):

Parameter resolution order: CLI flags → profile values → smart defaults.

Configuration Protocol

Before any BucketFS operation, verify the connection is configured:

1. Check if ~/.exapump/config.toml exists and contains a default profile.
2. If configured, proceed with the operation.
3. If not configured, ask the user for the required connection details (host, port, bucket, passwords). Do not guess or assume any defaults. Help the user create the profile in ~/.exapump/config.toml.

Commands

ls — List Contents

exapump bucketfs ls [PATH] [OPTIONS]
exapump bucketfs ls -r [PATH]            # Recursive listing
exapump bucketfs ls --recursive [PATH]

Examples:

exapump bucketfs ls                      # List bucket root
exapump bucketfs ls models/             # List a directory
exapump bucketfs ls -r models/          # Recursively list all files under models/

cp — Copy / Upload / Download

Direction is automatically determined by the source type (local file vs. BucketFS path).

Upload a local file to BucketFS:

exapump bucketfs cp <local-file> <bucket-path>
exapump bucketfs cp <local-file> <bucket-dir>/    # Preserve filename

Download a file from BucketFS to local:

exapump bucketfs cp <bucket-path> <local-path>

Examples:

exapump bucketfs cp my_model.pkl models/my_model.pkl     # Upload with explicit name
exapump bucketfs cp my_model.pkl models/                 # Upload, preserve filename
exapump bucketfs cp library.jar jars/library.jar         # Upload JAR for UDF
exapump bucketfs cp models/my_model.pkl .                # Download to current dir
exapump bucketfs cp models/my_model.pkl ./local-copy.pkl # Download with rename

rm — Remove a File

exapump bucketfs rm <path-in-bucket>

Examples:

exapump bucketfs rm models/old_model.pkl     # Delete a single file

Typical Use Cases

Upload a JAR for a Java UDF

exapump bucketfs cp my_library.jar jars/my_library.jar

Reference in UDF SQL:

CREATE OR REPLACE JAVA SCALAR SCRIPT my_schema.my_func(input VARCHAR(2000))
RETURNS VARCHAR(2000) AS
  %scriptclass com.example.MyClass;
  %jar /buckets/bfsdefault/default/jars/my_library.jar;
/

Upload an ML Model for a Python UDF

exapump bucketfs cp model.pkl models/model.pkl

Load in Python UDF:

import pickle
with open('/buckets/bfsdefault/default/models/model.pkl', 'rb') as f:
    model = pickle.load(f)

Upload a Custom Script Language Container (SLC)

exapump bucketfs cp my_slc.tar.gz slc/my_slc.tar.gz

Then activate via SQL:

ALTER SESSION SET SCRIPT_LANGUAGES='PYTHON3=localzmq+protobuf:///bfsdefault/default/slc/my_slc?lang=python#buckets/bfsdefault/default/slc/my_slc/exaudf/exaudfclient_py3';

Browse and Clean Up BucketFS

exapump bucketfs ls -r                        # See all files
exapump bucketfs rm old_model.pkl             # Remove an outdated file

Related Skills

• exasol-udfs: For creating UDF scripts that reference files stored in BucketFS.
• exasol-database: For SQL-level operations and connecting to Exasol.