In such cases, you need to use the "Dynamic pollsters operations" to process the sample, and retrieve the key as "rgw.main.size". By default, this value ("rgw.main.size") is interpreted as a nested dictionary. I mean, an object with a key "rgw", and then, another one that has a key "main", where there is a dict, with a key "size".
To handle such cases, you would need something similar to: `value_attribute: "usage || value['rgw.main'] | value['size']"`. However, that might not address all use cases. You will also need to handle situations when there is no key "rgw.main" in the response samples.