Different result for cardinality aggregation [6.3.0] in Python 3.6 plugin

[Wietse]

Hi

I'm trying to extract every hour the amount of unique users who logged in the past 24 hours. For this I use the cardinality aggregation with a range on the time. I use ELK 6.3.

GET logins/_search
{
  "_source": "user_id", 
  "query": {
    "range": {
      "time": {
        "gte": 1537016400.0,
        "lt": 1537102800.0
      }
    }
  },
  "aggs": {
    "unique_users": {
      "cardinality": {
        "field": "user_id"
      }
    }
  }
}

This yields me the same result every time, when I raise the precision threshold I get the same result still (254202 unique users).

However, when I execute this query using the Python API (Python 3.6), I get three different results (254202, 202847, 209580 unique users). These values alternate consistently. The correct value is among them, but I need to know which one it is of course. Am I missing something? An extra option in Python? Raising the precision threshold in Python did not help.

Any help is much appreciated.