NIFI-2142 Cache compiled XSLT in TransformXml

[jfrazee]

Two other tidbits:

1. This also modifies the XSLT_FILENAME attribute to support expression language.
2. This updated the test files to have 3 space indents because Saxon defaults to that and AFAICT the free version doesn't let you change that -- it's either no indent or 3 space. Could be wrong though.

See #591 for PR against 0.x and past discussion.

[JPercivall]

Hey @jfrazee, looks like you have contrib check issues that cause the build to fail:

[WARNING] src/main/java/org/apache/nifi/processors/standard/TransformXml.java24:8 UnusedImports: Unused import - java.util.Collection.
[WARNING] src/test/java/org/apache/nifi/processors/standard/TestTransformXml.java19 AvoidStarImport: Using the '.' form of import should be avoided - java.io..
[WARNING] src/test/java/org/apache/nifi/processors/standard/TestTransformXml.java36:8 UnusedImports: Unused import - org.junit.Ignore.

[jfrazee]

@JPercivall Oh, right. I think these are corrected now but just noticed the image bundle has a bunch of license violations (even after rebasing).

[JPercivall]

@jfrazee the image bundle is no longer in master (replaced by the media bundle). Do a "git clean -i -d" to remove any old bundles.

[JPercivall]

Is this property a max character length or data size limit?

[JPercivall]

Also, this is a integer validator but the property is interpreted as a long on line 171

[JPercivall]

Actually my new guess is that this is the maximum number of stylesheets stored (the description could be added to, lol)

[jfrazee]

@JPercivall Yes, that's correct. So "Maximum number of stylesheets to store in the cache"?

[JPercivall]

@jfrazee I've made a couple comments. Looks good overall, mainly just documentation comments. Do you by chance have a template and/or some example xml/XSLT that I could use for testing to verify the changes?

[jfrazee]

1. @JPercivall I addressed all your comments above and added a template to https://gist.github.com/jfrazee/a501336826aa77d0daa74e8e0aa3afd0.
2. There is one change in behavior now -- a cache size of 0 disables the cache whereas it was just allowing for an arbitrarily large cache.
3. There's a screenshot of the example data flow added to https://issues.apache.org/jira/browse/NIFI-2142 which illustrates the difference between running with and without caching. It's not 100% apples to apples with the older TransformXml, but behaves the same way. The queues back up and there's a stark contrast in Tasks/Time.
4. I rebased on the latest master, but contrib-check is failing because of the commenting style on AbstractSNMPProcessor.

[JPercivall]

The cache will still be created and on line 213 it will try to use it to look up an xslt if the cache size is 0. I'd suggest not even creating it in that case.

[jfrazee]

@JPercivall Addressed your latest comments and added a test with the cache disabled since the logic varies depending on that value now.

[JPercivall]

Sorry for taking so long @jfrazee, I got caught up with something I thought was an issue but turned out it was just me messing up.

+1

Visually verified code, and did a contrib check build. In standalone and a secure 3 node cluster tested the caching using the provided template. Even with just 5 stylesheets being used I saw speed improvements of ~10x! Thanks @jfrazee, I will merge it in.