Update support for iRODS

[korydraughn]

This PR updates support for iRODS by replacing Jargon (legacy iRODS library) with irods4j.

The foundational work was implemented by @MINGYJ, a recent iRODS intern. My commits are mainly polish and corrections around the use of the irods4j library.

Basic functionality is working - i.e. single stream uploads/downloads, renames, editing, etc.

The parallel transfer implementation doesn't appear to be working as intended. This is likely due to not having a full understanding of how the Cyberduck components fit together - i.e. Read/WriteFeature vs Upload/DownloadFeature.

Here are the steps for performing parallel transfer (i.e. multipart uploads) in iRODS.

• https://docs.irods.org/5.0.1/system_overview/data_objects/#how-to-perform-a-parallel-write-over-the-zone-port

Putting in draft for now. Feedback and guidance on how to implement proper support for parallel transfer would be greatly appreciated.

Resolves #14449.

[dkocher]

Putting in draft for now. Feedback and guidance on how to implement proper support for parallel transfer would be greatly appreciated.

Discussed in #14449 (comment)

[korydraughn]

TODO: Update documentation for iRODS - https://docs.cyberduck.io/protocols/irods/

[trel]

updating the docs is a separate repo, would have its own PR

[korydraughn]

Rebased PR on top of master.

Verified the following:

• Parallel/Single stream uploads and downloads works
• Config options defined in new iRODS profile are honored
• PAM authentication works via the pam_password auth scheme
• Other operations work too - rename, delete, etc.

With this PR, users will not be able to calculate checksums on data in an iRODS zone. Cyberduck will report checksums in their hex form if they exist in iRODS. We could add an option to the iRODS profile which allows users to instruct Cyberduck to calculate checksums following an upload, but that didn't feel like the correct approach, mainly because profiles seem to be loaded once on program start and never reread.

As for the unit/integration tests, I'm not sure how the implementation can be tested without a real iRODS server.

Finally, this PR bumps the minimum iRODS version requirement to 4.3.2. A new version of irods4j will be needed before this is merged. See irods/irods4j#126. Will get a new version released this week soon.

[korydraughn]

@dkocher When I move a file in or out of a directory, I see a new iRODS connection get created. It disappears eventually, but not cleanly.

Can you explain why Cyberduck creates a new connection every time a file is moved?

The move class is very similar to other implementations so it's not clear to me what causes the new connection.

[dkocher]

@dkocher When I move a file in or out of a directory, I see a new iRODS connection get created. It disappears eventually, but not cleanly.

Can you explain why Cyberduck creates a new connection every time a file is moved?

The move class is very similar to other implementations so it's not clear to me what causes the new connection.

You will need to return Statefulness.stateless in IRODSProtocol#getStatefulness.

[korydraughn]

@dkocher When would someone choose stateless or stateful?
If I change the protocol to stateless, how will that affect other operations?
If I continue using stateful, what needs to change to make sure the additional connections are closed immediately?

To provide some context, a single iRODS connection cannot be used to execute multiple API operations in parallel. A single request is sent and the server returns a response.

[dkocher]

If you can share credentials with us for your test environment we can run daily integration tests from our CI.

[dkocher]

@dkocher When would someone choose stateless or stateful? If I change the protocol to stateless, how will that affect other operations? If I continue using stateful, what needs to change to make sure the additional connections are closed immediately?

To provide some context, a single iRODS connection cannot be used to execute multiple API operations in parallel. A single request is sent and the server returns a response.

We will have to keep it stateful then as otherwise it will be attempted to use a single connection for multiple actions in parallel, i.e. when the user is browsing folders or a file transfer with multiple files in parallel.

I will need to review how we can still support the native copy feature implementation for iRODS that would not require a new connection.

[korydraughn]

If you can share credentials with us for your test environment we can run daily integration tests from our CI.

We don't have a CI system for people to hook into yet. We have a small set of tools which make it easy to launch one or more iRODS servers for testing. It's likely overkill for your needs though.

With that said, building a containerized environment for testing iRODS is pretty easy. I'm happy to put together a Docker compose project for the iRODS component. That will allow you to launch it on a local computer and have it available for testing.

We will have to keep it stateful then as otherwise it will be attempted to use a single connection for multiple actions in parallel, i.e. when the user is browsing folders or a file transfer with multiple files in parallel.

I will need to review how we can still support the native copy feature implementation for iRODS that would not require a new connection.

So, because the IRODSProtocol class reports iRODS as being stateful, does that mean every operation results in a new iRODS connection being created?

If that's true, why is it that the Move operation results in connections which do not disconnect? No other operation shares that behavior.

I added some log statements to my local build to try and box in what is leading to the additional connections, but it didn't help. However, it did reveal a high number of instantiations of IRODSMoveFeature. Any idea why the move operation is instantiated so much?

[dkocher]

We have other usages of Docker Compose containers in integration tests, thus that should be feasible.

[korydraughn]

Where should I place the Docker Compose project? Is the test directory for irods appropriate?

I figure you can move things around if needed.

[korydraughn]

@dkocher What triggers the Copy implementation?

Using the Duplicate option within the context menu doesn't appear to trigger it.

[korydraughn]

Docker compose project added under test directory.

Squashing everything down.

[dkocher]

1. Due to the IRODSProtocol implementation reporting as "stateful", Cyberduck will not use the same iRODS connection for parallel operations, correct?

That is correct. Connections are leased from a pool.

2. Will Cyberduck block when the user attempts to execute another operation while the connection is in use?

No, a second connection will be opened.

[dkocher]

Will need to be removed when rebasing with #17640 merged.

Suggested change

.withLocalCompose(true)

[korydraughn]

Hmm, the tests fail when that line is removed. They pass when it is included.

Is that expected?

[dkocher]

Have you rebased to origin/master? With the updated testcontainers:2.x dependency the API should no longer be available.

[korydraughn]

So far, all of my GUI testing has been "click button, wait for operation to finish, check results and log". I have not tried executing multiple operations in parallel using the GUI.

The simplest way to test manually is to expand multiple folders at once.

I'm probably misunderstanding, but wouldn't I need to expand multiple folders fast enough such that the operations overlap?

And if that's the case, is there a trick to doing that in the GUI?

[dkocher]

So far, all of my GUI testing has been "click button, wait for operation to finish, check results and log". I have not tried executing multiple operations in parallel using the GUI.

The simplest way to test manually is to expand multiple folders at once.

I'm probably misunderstanding, but wouldn't I need to expand multiple folders fast enough such that the operations overlap?

And if that's the case, is there a trick to doing that in the GUI?

Select multiple folders and press the right arrow key to expand.

[korydraughn]

Wow, can't believe I forgot about multi-select.

Will give that a try.

[dkocher]

Please use

Suggested change

	IRODSApi.setApplicationName("Cyberduck");
	IRODSApi.setApplicationName(PreferencesFactory.get().getProperty("application.name"));

as the library is not limited to usages in Cyberduck.

[dkocher]

[ERROR] /home/runner/work/cyberduck/cyberduck/irods/src/main/java/ch/cyberduck/core/irods/IRODSConnectionUtils.java:[69,16] cannot find symbol
[ERROR]   symbol:   variable trustManagers
[ERROR]   location: variable options of type org.irods.irods4j.low_level.api.IRODSApi.ConnectionOptions

Is there a dependency version update required?