Comment 4 for bug 1831910

I've done some poking around in the db, and the '\r' is in the raw db value for nextcloud, as is the split channels for lxd, so the problem is definitely in the way snapstats miner is generating/storing the aggregates, as opposed to a bad transform when reading them.

I'm looking into fixing the generation, but for bad data already present, it's tricky to know what to do. In this case, on reading the data, we could maybe have a validation of channel names, and just drop bad ones, which should mean the data that is there can be parsed buy the consuming api, although there well be missing data.

Backfilling a fix is tricky, as we don't like to update in cassandra, as it leaves tombstones around.