Skip to main content

Frequently Asked Questions

Which object storage are supported by Fusion​

Fusion currently supports AWS S3 and Google Storage. In the near future it will also support Azure Blob storage.

Can I use Fusion with Minio?​

Yes. Minio, implements a S3-compatible API, therefore it can be used in place of AWS S3. See the documentation how to configure your pipeline execution to use Fusion and Minio. (link to guide TBD).

Can I download Fusion?​

No. Currently, Fusion can only be used by enabling Wave containers in the configuration of your Nextflow pipeline.

Why I need Wave containers to use Fusion?​

Fusion is designed to work at level of job executions. For this reason, it needs to run in containerised job execution context.

This would require to rebuild all containers used by your data pipeline to include the Fusion client each time a new version of the Fusion client is released, and it would make necessary to maintain a custom mirror or existing containers collections, such as BioContainers which is definitively not desirable.

Wave allows adding the Fusion client in your pipeline containers at deploy time, without having to rebuild them or to maintainer a separate container images collection.

How Fusion works behind the scene?​

Fusion is implemented a FUSE driver that mounts the storage bucket in the job execution context as a POSIX file system. This allows the job script to read and write data over the object storage like it were local files.

Can Fusion mount more than one bucket in job file system​

Yes. Fusion any access to an object storage is automatically detected by Fusion and the corresponding bucket is mounted on-demand.

Can Fusion mount buckets of different vendors in the same execution?​

No. Fusion can mount multiple buckets but the must be of the vendor e.g. AWS S3 or Google Storage.

How Fusion can be faster of other existing FUSE driver?​

Fusion is not a general purpose file system. Instead, it has been designed to optimise the data transfer of Nextflow data pipeline taking advantage of the data model used by Nextflow. [to be improved]

I tried Fusion, but I didn't notice any performance improvement. Why?​

Make sure the computing nodes in your cluster have NVMe SSD storage or equivalent technology. Fusion implements an aggressive caching strategy that requires the use of local scratch storage bases on solid-state disks.