Why does the cleanup step take so long?

[raihle]

I have a pretty simple workflow that installs dependencies with jf npm ci. This leads to build info being published automatically, which is what I want (eventually I will also publish an artifact, but one thing at a time).

The build info is published as part of the cleanup (post) step at the end of the workflow, and that part finishes pretty quickly (within seconds). After this, my server config / credentials are deleted. For some reason, this last step (which should not be much more complicated than deleting a single file?) consistently takes 3 minutes. Why? What can I do to avoid this?

Relevant workflow steps:

      - name: Setup JFrog CLI with OIDC
        uses: jfrog/setup-jfrog-cli@v4
        id: jfrog_oidc
        env:
          JF_URL: ***
          JF_PROJECT: ***
        with:
          oidc-provider-name: ***
          oidc-audience: ***
          custom-server-id: xyz
          
      - name: Install dependencies
        run: jf npm ci

Example logs with timestamps:

Tue, 03 Jun 2025 07:54:26 GMT ::group::Cleanup JFrog CLI servers configuration
Tue, 03 Jun 2025 07:54:26 GMT Cleanup JFrog CLI servers configuration
Tue, 03 Jun 2025 07:54:26 GMT The value of custom is: 'xyz'
Tue, 03 Jun 2025 07:54:26 GMT ##[debug]Removing custom server ID: 'xyz'...
Tue, 03 Jun 2025 07:54:26 GMT /home/runner/_work/_tool/jf/2.75.0/x64/jf c rm xyz --quiet
Tue, 03 Jun 2025 07:54:26 GMT ::endgroup::
Tue, 03 Jun 2025 07:57:26 GMT ##[debug]Node Action run completed with exit code 0
Tue, 03 Jun 2025 07:57:26 GMT ##[debug]JFROG_CLI_COMMAND_SUMMARY_OUTPUT_DIR=''
Tue, 03 Jun 2025 07:57:26 GMT ##[debug]JFROG_CLI_COMMAND_SUMMARY_OUTPUT_DIR='/home/runner/_work/_temp'
Tue, 03 Jun 2025 07:57:26 GMT ##[debug]JFROG_CLI_USAGE_AUTO_BUILD_PUBLISHED='TRUE'
Tue, 03 Jun 2025 07:57:26 GMT ##[debug]Finishing: Post Setup JFrog CLI with OIDC

[raihle]

After some more experiments, it seems that the summary is the problem. With disable-job-summary enabled (and similarly, if there is no build info to publish and thus nothing to summarize), there is no 3 minute delay. It seems odd that this would be the culprit, because as far as I can tell the summary is already output before the final cleanup happens, and it's not particularly long (which might otherwise cause some delay on GitHub's side while processing the output)

I have found that I can create a good enough summary myself without incurring the delay, and will probably proceed that way. But I think the original question stands ... why?

[raihle]

More experiments...

      - name: Publish JFrog Build Info
        env:
          JFROG_CLI_COMMAND_SUMMARY_OUTPUT_DIR: "${{runner.temp}}/jfrog-summary"
        run: |
          jf rt bce
          jf rt bag
          jf rt bp
          jf gsm
          cat $JFROG_CLI_COMMAND_SUMMARY_OUTPUT_DIR/jfrog-command-summary/markdown.md >> $GITHUB_STEP_SUMMARY

This includes most of the generated build summary, seemingly missing only header and footer - and there is no delay.

[p4rs3c]

Hey @raihle

The problem most likely arose on GHES environments that can be air-gapped and in which there is no transport to media.jfrog.com where the PNG for the summary header is located. The problem arose due to incorrect processing of socketTimeout in the http-client.

#302

A fix was applied in 4.6.1.

[raihle]

@p4rs3c I can confirm that the current version has fixed the delay. We are not using GHES in this case, but we do use self-hosted runners.