Multi Cloud Virtual Directory
Be part of a team to create a multicloud virtual directory.
Your goal will be to to develop an API, secure REST, and commandline tool that easily interfaces with storage servoces such as file based services and object stores osted on the internet to create a virtual directory that accesses the files by directory name and filename.
This is espaciellyusefull for data scientists that may want to strore their data on multiple cloud providers and eliminate vendor lockin.
It could also be useful for geographically distributed files that allow services to utilize them based on speed of access or the physical location.
In a previous version of cloudmesh we have developed an integrated approach for compute and storage services. However in this project we like to explore some new features and separate the implementation form the main cloudmesh.
In the past cloudmesh has conducted two such projects
We like to reimplement thise projects and significantly improve them by
- Adding more providers
- Adding a secure rest service for acceassing the virdual directory
- Creating a slick user interface to brows the virdula directry
- Put everything in a container so it can be run on Linux, MAc and Windows.
- Dreate a convenient commandline tool that allows starting of the service, interacting with it and make this real easy to use. THe commandline will hide the docker commands while providing human readable abbreviations.
- Deliver unit tests with pytests
- Deliver a high quality report including benchmarks
In order for you to participate in this project you will need:
- A computer on which you can run docker (Windows Home will not work)
- Significant python knowledge
- Be highly motivated
- Be willing to have meetings on this project once or twice a week
- SHowcase significant progress over the lifetime of the project.
- be knowledgable with github (a repository will be provided to which Dr. von Laszewski will contribute)
- Conduct task management in GitHub (Gregor will explain)
- Be honest and not hiding problems or implementation bugs.
- You must be able to do a videoconference and be able to share your screen (I typically use google meet or zoom).
One of the two projects used a single file taansfer logic, while the other used a parallel file transfer option. It will be up to you to decide how to proceed.
In the first week collect all clouds and services that could be used in this project. THis work should be split between the teammembers and a detailed analysis including features including authentication and security must be conducted.
You will need to evaluate the previous efforts and decide how to proceed.
One way is to select several cloudas and define an abstraction that has to be implemented for each cloud. this can be actually bes demonstrated by an example commandline such as
vdir mkdir DIR vdir cd [DIR] vdir ls [DIR] vdir add [FILEENDPOINT] [DIR_AND_NAME] vdir delete [DIR_OR_NAME] vdir status [DIR_OR_NAME] vdir get NAME DESTINATION
From the commandline derive some API interface and use FastAPI to implement it.
Identify a mechnism on how to deal with the security. AN encrypted file with your cloud credentials will be sufficient.
One of the limitations of the previous vdir project is that the fill add function is done only on a single file but not on a recursive file tree. Naturally we need to upload, single files, multiple files and filetrees, similar to the unix command rsync. This has been implemented in the storage command
storage [--storage=SERVICE] [--parallel=N] create dir DIRECTORY storage [--storage=SERVICE] [--parallel=N] get SOURCE DESTINATION [--recursive] storage [--storage=SERVICE] [--parallel=N] put SOURCE DESTINATION [--recursive] storage [--storage=SERVICE] [--parallel=N] list [SOURCE] [--recursive] [--output=OUTPUT] storage [--storage=SERVICE] [--parallel=N] delete SOURCE storage [--storage=SERVICE] search DIRECTORY FILENAME [--recursive] [--output=OUTPUT] storage [--storage=SERVICE] sync SOURCE DESTINATION [--name=NAME] [--async] storage [--storage=SERVICE] sync status [--name=NAME] storage config list [--output=OUTPUT] storage [--parallel=N] copy SOURCE DESTINATION [--recursive]
We like naturally to (re-)implement the storage command.
Before you start implementing describe and showcase a couple of commands on how to used it. Our command will be called cdir (for cloudmesh dir so it is easy to distinguish)
Make sure that you provide a comprehensive list of potentail cloud storage ervices. Distinguish between for pay and free services.
The list could include:
- AWS Drive
- AWS Object Store
- An ssh accesible computer
- Google Drive
- Google docs/drive
- Your local computer
Are there companies or services that already offer this?
When defining the REST API we like to be able to use a dirname and a basename within the api similar to python. to deal with provider specific filenames we do have two urls for a file.
The first allows duplication of files between services, the latter defines a “preferred” url based on some criteria. It could be that we just set the preferred provider for the file.
Note that the introduction of duplicating files is new
It also requires an update so that the file with the newest timestamp gets updated on all registered serrvices.
Please note that the project addresses the ability to integrate object store and regular file system based storage.
In case of qeuestions, pleas econtact Gregor at