Hi @jen-huang, thanks for finding and reporting the duplicate pipeline. What would be the easiest way to test this fix?

I think this is a pretty safe change so I'm happy to test it once this gets deployed out somewhere.

Thanks @jen-huang. I'm good with testing this change once this package is published to the snapshot registry then. WDYT @fearful-symmetry?

Pinging @elastic/integrations (Team:Integrations)

💚 Build Succeeded

the below badges are clickable and redirect to their specific view in the CI or DOCS

@ycombinator I don't know enough how how ingest pipelines are installed to say for sure, but I'm overall fine with it. I'm mostly curious how this has flown under the radar. I don't think we've touched syslog for a while?

I'm mostly curious how this has flown under the radar.

Both pipelines for the system/syslog dataset are very similar (the only material difference I could find was that the .yml pipeline had an extra processor at the end to set event.type which was absent in the .json pipeline). So my guess is that the net effects on the indexed data were the same (or close enough) that no one noticed any issues. 🤷

I think what might help is adding pipeline tests for this data stream. Looks like we have some test files that could be ported over from the corresponding Filebeat fileset.

@jen-huang what happens in Kibana if it detects a default.yml and default.json? I assume it installs both pipelines but they end up having the same names so the first one gets overwritten by the second one? Is there any determinism here, e.g. the .json would always be installed before the .yml one or vice versa?

I think what might help is adding pipeline tests for this data stream. Looks like we have some test files that could be ported over from the corresponding Filebeat fileset.

Just to clarify, I'm adding these to this PR.

@jen-huang what happens in Kibana if it detects a default.yml and default.json? I assume it installs both pipelines but they end up having the same names so the first one gets overwritten by the second one? Is there any determinism here, e.g. the .json would always be installed before the .yml one or vice versa?

Yep, both get installed and the second overwrites the first one. We don't do any sort of logic around the order, just whatever the order happens to be in the package manifest. Looking at system-0.10.9, the .json always seems to appear before .yml, so the .yml version would be the one that is "actually" installed.

Edit: Looking closer at the code, we actually send the ES requests in parallel, so it's possible that either format gets installed.

@fearful-symmetry @mtojek do you have any tips on how to debug the tests failures in this PR: https://beats-ci.elastic.co/blue/organizations/jenkins/Ingest-manager%2Fintegrations/detail/PR-725/8/tests/? Initially I'm focussing on getting the pipeline test for test-darwin-syslog-sample.log to pass. I assume the other test cases have similar issues.

I think you need to write the config to mark new lines for pipeline tests and probably timezone (as the pipeline doesn't accept lack of it). Something like this:

{
  "multiline": {
    "first_line_pattern": "^Dec 13 "
  },
  "fields": {
    "event.timezone": "GMT+2"
  }
}

Thanks for the tip, @mtojek. I had already added the multiline config but looks like I needed the fields config too for the time zone field. Trying that now...

@mtojek I finally gave up and ran elastic-package test pipeline --data-stream syslog --generate. After doing this and making a couple of other tweaks to the test case config files, all pipeline tests for the system/syslog data stream are now passing.

Take a look at the test-darwin-syslog-sample.log-expected.json file and compare it with the test-darwin-syslog-sample.log file. The expected file has only one event while the corresponding raw log file has three events. This looks wrong 🙂 and the tests should not be passing, but they are! I think maybe there is a bug in the pipeline test runner?

@mtojek ignore my previous comment 😄. I had to tweak the multiline.first_line_pattern in the test case config. 🤦

@fearful-symmetry I added test case files for the system/syslog data stream. Would you mind re-reviewing this PR? Thanks!