File System based Bank implementation¶

Problem description¶

Currently we only suppport Swift as a bank implementation. This means that anyone that uses Karbor must also install Swift. This might be unacceptable or over complicated for some deployments. Furthe more, having a many options for bank backends is always a good things.

I suggest adding an FS based implementation. It will use files for objects storing objects and object metadata.

Use Cases¶

As explained, deployers might not want or will be unable to install Swift in their cloud.

Proposed change¶

Objects would be stored under a file name with their ID having / be defined as a directory separator.

For example::

Object ID: /checkpoints/2fd14f87-46bd-43a9-8853-9e1a84ebee3d/index.json

Since object names might contain chars that are unavailable as regular files we will need to escape some chars so that they can be used as file names.

We propose the following encoding escape sequence non ascii chars would be modified to %[XX..] where XX are Hex representations of the utf-8 encodinf of the characters.

This avoids using back-slash for escape.

Example::

object*with%wierd*id => object%[2A]with%[25]wierd%[2A]id

The metadata files will be in a JSON format. The name and format of these files are same as the meatadata objects in the swift bank.

For example::

/checkpoints/3a4d76e7-f8d8-4f2f-9c1d-107d88d7a815/ <- directory /checkpoints/3a4d76e7-f8d8-4f2f-9c1d-107d88d7a815/metadata <- md file /checkpoints/3a4d76e7-f8d8-4f2f-9c1d-107d88d7a815/status

Implementation¶

Dependencies¶

None

Testing¶

Unit tests in Karbor.

Documentation Impact¶

New docs to explain how to use and configure the alternative Bank implementation.

References¶

None