NIFI-11472 Make PutFTP processor more multithread friendly

[MormonJesus69420]

Add an extra check during directory creation to see if directory wasn't already created in another thread.

From Issue:

Problem happens when a PutFTP is set to run several concurrent tasks and two (or more ) FlowFiles come in and both need to create the same directory. One of them will create directory and succeed immediately while the other will try to create directory, but fail since it already exist, throw an error, the FlowFile will then be penalized and on second run will succeed.

While it is not the biggest error, as files are getting transferred in the end, but the bulletins and errors are annoying, especially in production environment where you don't want to get unnecessary errors.

We found that the solution involves a simple change to the FTPTransfer.java class in:
nifi-nar-bundles/nifi-standard-bundle/nifi-standard-processors/src/main/java/org/apache/nifi/processors/standard/util/FTPTransfer.java
On line 398 and ensureDirectoryExists method you can simply add another if check which double checks that the directory exists when it fails to create one.

final boolean cdSuccessful = setWorkingDirectory(remoteDirectory);

if (!cdSuccessful) {    
  if (client.makeDirectory(remoteDirectory)) {        
    logger.debug("Remote Directory not found: created directory [{}]", remoteDirectory);    
  } else if (!setWorkingDirectory(remoteDirectory)) { 
         // Double check that the dir exists as it might have been created in another thread        
    throw new IOException("Failed to create remote directory " + remoteDirectory);    
  }
}

Summary

NIFI-11472

Tracking

Please complete the following tracking steps prior to pull request creation.

Issue Tracking

• Apache NiFi Jira issue created

Pull Request Tracking

• Pull Request title starts with Apache NiFi Jira issue number, such as NIFI-00000
• Pull Request commit message starts with Apache NiFi Jira issue number, as such NIFI-00000

Pull Request Formatting

• Pull Request based on current revision of the main branch
• Pull Request refers to a feature branch with one commit containing changes

Verification

Please indicate the verification steps performed prior to pull request creation.

Build

• Build completed using mvn clean install -P contrib-check
  • JDK 11
  • JDK 17

Licensing

• New dependencies are compatible with the Apache License 2.0 according to the License Policy
• New dependencies are documented in applicable LICENSE and NOTICE files

Documentation

• Documentation formatting appears as expected in rendered files

[arpadboda]

Hello @MormonJesus69420 , thanks for your contribution!
I'm not sure if using multiple threads to transfer to the same ftp makes any sense, this operation is limited by bandwith anyway.
So I would prefer to restrict it to single threaded usage, but the current implementation allows uploading to multiple hosts based on flowfile attributes, in which case multiple threads might make sense, so I'm not against this change.
Weak +1, I'm ok to merge with a 2nd approval.

[MormonJesus69420]

Hi @arpadboda I might not have described it so well, but the issue we face is when we configure processor to run several concurrent tasks.

When we change the number to two or more we start receiving "errors" about processor being unable to create directory as it already exists. While the issue resolves itself, it is rather distracting to see it, when it's not a "real" error.

We have noticed a significant performance boost when using a PutFTP processor with more than one concurrent task. I don't have the numbers on me at the moment, but switching to two or three concurrent tasks significantly sped up the transfer time.

[MormonJesus69420]

I am sorry, I managed to make the most basic mistake in such a small change, I forgot to add a ! to the setWorkingDirectory(remoteDirectory) method call.

[MormonJesus69420]

Strange, I don't understand why it failed on the Windows action. I don't have the ability to test it on Windows either, since we use Linux for development at work. Also my branch is based off of the latest nifi/main branch.

[exceptionfactory]

Thanks for the contribution @MormonJesus69420, and thanks for the review @arpadboda!

The Windows build often takes longer than the others, so it timed out, I restarted it to try again.

The use case of multiple concurrent tasks for PutFTP makes sense, the introducing the additional call to setWorkingDirectory seems reasonable under the circumstances. Within the context of a single NiFi server, a more robust solution might be to add synchronized to this method, but that would not help when running PutFTP across multiple NiFi nodes in a cluster.

With that background, attempting to change to the directory if the makeDirectory command fails seems like the best approach. I will monitor the Windows build status for verification.

[exceptionfactory]

Thanks again @MormonJesus69420! +1 merging

[MormonJesus69420]

Hi, I see that the PR is already closed, but thought I might add the example where we tested running PutFTP processor using 1 and 2 concurrent tasks. As long as we were able to provide the data to send, you can see that upload speed was almost doubled. As such the bottleneck was how fast we were able to provide data to send, rather than sending of data.