Loggers serialization safety

[pomadchin]

Even though loggers are serializable, we have to be sure that there would be no problems during multiple ser / deser. Spark internal loggers all marked with @transient, all Hadoop loggers are final static. There is still no unit tests related to that issue, though you can notice it during Spark memory shuffle and using DISK_ONLY or MEMORY_AND_DISK RDD persistence levels.

Signed-off-by: Grigory Pomadchin [email protected]

[hjaekel]

We tested your patch with our code and the NullPointerException is gone, everything works fine now.

Just for the documentation, this was the exception we get when GeoTrellis is using the com.typesafe.scalalogging.LazyLogging:

java.lang.NullPointerException
        at org.slf4j.impl.Log4jLoggerAdapter.isWarnEnabled(Log4jLoggerAdapter.java:391)
        at geotrellis.raster.ArrayTile$class.convert(ArrayTile.scala:46)
        at geotrellis.raster.UShortArrayTile.convert(UShortArrayTile.scala:24)

We could not reproduce this issue in unit tests, too. The NPEs just occur while executing bigger processes with limited memory.

[lossyrob]

Thanks for looking into that @hjaekel.

👍

[echeipesh]

I don't think this is a breaking change or even api-change. We replaced a protected field with a protected field.

And addition of @transient does not break binary compatibility: https://docs.oracle.com/javase/specs/jls/se7/html/jls-13.html#jls-13.4.11