If I recall correctly, Brian mentioned at the MAMI SIGCOMM tutorial that metadata from different sets with different metadata keys are merged; metadata with same keys and the same value is merged; but identical keys with different values are dropped.
I think dropping information isn't a good strategy since the metadata is probably important (why put it in the metadata otherwise?), and would propose either indicating that information was removed ('key': '/removed due to conflicting values/'), or creating an array of both (all) values (probably breaking existing parsers), or renaming the keys (key1, key2, etc) (also breaking parsers expecting the original key name. Anyway, some ideas in case you haven't considered this earlier.
If I recall correctly, Brian mentioned at the MAMI SIGCOMM tutorial that metadata from different sets with different metadata keys are merged; metadata with same keys and the same value is merged; but identical keys with different values are dropped.
I think dropping information isn't a good strategy since the metadata is probably important (why put it in the metadata otherwise?), and would propose either indicating that information was removed ('key': '/removed due to conflicting values/'), or creating an array of both (all) values (probably breaking existing parsers), or renaming the keys (key1, key2, etc) (also breaking parsers expecting the original key name. Anyway, some ideas in case you haven't considered this earlier.