...
I've been able to use gdb to examine call stack backtraces of the running worker child process, during periods when Marmotta requests have stalled See the attached file, below. This deserves further analysis, but I have a couple of backtraces that show garbage collection and memory allocation being performed. I have not yet analyzed the calls higher up the stack leading to the memory allocation and garbage collection.
It's interesting that we don't see the mappings stall out with all of our providers. The current issue is getting Hathi running beyond 24K records, but other providers that we've mapped recently have been able to map hundreds of thousands of records in an acceptable amount of time. Does record size, or some other quality of the record, have something to do with it?
We may benefit from some more time spent here to gather more information.
...