adapt forked elasticsearch 7 formula

[joeyAghion]

It appears that Elasticsearch is no longer maintaining homebrew formulas for their service at all. They recommend their container images instead.

Most acutely, this means the previously available formula for Elasticsearch 7 fails with undefined method plist_options errors (elastic/homebrew-tap#146). @dskecse has modernized the formula, but that also fails on java-related errors. The new formula here is based on that, with just the addition of an ES_JAVA_HOME environment variable.

This works, but Elasticsearch crashes when exercised by Gravity's tests with the error:

ElasticsearchException[Failure running machine learning native code. This could be due to running on an unsupported OS or distribution, missing OS libraries, or a problem with the temp directory. To bypass this problem by running Elasticsearch without machine learning functionality set [xpack.ml.enabled: false].]

Apparently (elastic/elasticsearch#48490), the way homebrew moves files during installation invalidates the file signatures of ML executables. I'm currently working around this by manually disabling ML features in /opt/homebrew/etc/elasticsearch.yml:

xpack.ml.enabled: false

This^ configuration change should be handled by the formula, so this isn't mergeable yet.

[anandaroop]

So Elasticsearch already knows what to do with ES_JAVA_HOME (as opposed to the previous JAVA_HOME?

[anandaroop]

(Yes, I take it)

[joeyAghion]

JAVA_HOME also works, but generated deprecation warnings recommending ES_JAVA_HOME instead. Neither worked using the old, indirect '/usr/libexec/java_home -v 17' form. Instead I had to actually run the command during installation.

[joeyAghion]

Updated to automatically insert the xpack.ml.enabled: false configuration. This might be mergeable if we don't actually require those ML features. I'll move on to the docker env and real v6 vs. v7 errors next.

[anandaroop]

might be mergeable if we don't actually require those ML features

It looks like a the ML features relate mostly to the analytics-focused use of ES — but also to the NLP tasks we might like to experiment with.

However we are nowhere close to actually doing that, so I don't think that needs to block this interim step.

[joeyAghion]

For now, this formula can be tested by uninstalling elasticsearch 6, pulling down this repo and branch, and installing from source:

brew install --build-from-source ./Formula/[email protected]