[mlgtcode]

This plugin fetches JSON payloads from configured endpoints.

[renakas]

Nested and non-numeric values go into the metrics as they are

_flatten_payload only flattens the top level. Everything else is passed through unchanged, so for a payload like:

{"temp": 21.5, "wind": {"speed": 4, "dir": "NW"}, "tags": ["a", "b"], "note": null, "city": "Oslo"}

the result contains weather_wind (a dict), weather_tags (a list), weather_note (None) and weather_city (a string). Metrics should be numeric, so these values are either dropped by the backend or break the series. The same applies to a top-level list or string payload (line 213).

Suggestion: flatten recursively with a depth limit and keep only numbers (and booleans as 0/1):

def _flatten_payload(self, prefix, value, depth=3):
    if isinstance(value, bool):
        return {prefix: int(value)}
    if isinstance(value, (int, float)):
        return {prefix: value}
    if isinstance(value, dict) and depth > 0:
        out = {}
        for key, child in value.items():
            out.update(self._flatten_payload('%s_%s' % (prefix, key), child, depth - 1))
        return out
    return {}  # strings, null, lists, too-deep values are skipped

Lists could be indexed (weather_tags_0) or ignored; I'd ignore them unless there's a use case.