Use precompiled Pattern instances

[otbutz]

I'm currently testing the import with all our new encoded values enabled and JFR reported that a lot of Pattern instances are being allocated.

As String.replaceAll() is not reusing it's Pattern instance, we create a new one each time it is called, which quickly becomes expensive.

[karussell]

Were you able to measure any speed ups for the import due to these changes?

[otbutz]

Ran the import twice using the following setup:

• source: germany-latest.pbf
• CH disabled
• LM enabled
• Xms23G Xmx23G
• all available encoded values
• Flightrecorder with preset profile
• measured from first log line until Dropwizard lists the available endpoints

Results

• unpatched: 17m44s
• with precompiled patterns: 16m40s (:arrow_down: ~6%)

Memory allocation according to JFR without precompiled Patterns (109GiB total int[] allocation)

Memory allocation according to JFR with precompiled Patterns (83.3GiB total int[] allocation)

[karussell]

Thanks!

This looks good :)