Use NaN-safe numpy functions in metric aggregations

[Copilot]

Metric aggregations returned NaN when any data point contained NaN, making the aggregation functions unusable for real-world metrics with missing data.

Changes

• Core aggregation logic: Replaced standard numpy functions with NaN-safe variants:

  • sum → nansum (treats NaN as 0)
  • max/min → nanmax/nanmin (ignores NaN)
  • average → nanmean (ignores NaN)
  • percentile → nanpercentile (ignores NaN)
  • std/var → nanstd/nanvar (ignores NaN)

• Consistency fix: Changed percentile calculation to use np_array instead of query_values list for alignment with other operations

Example

# Before: returns all NaN values
pc = PrometheusConnect(url="http://prometheus:9090")
result = pc.get_metric_aggregation(
    query="my_metric",  # Contains [1.0, NaN, 2.0, 3.0]
    operations=["sum", "max", "average"]
)
# {'sum': nan, 'max': nan, 'average': nan}

# After: returns meaningful statistics
# {'sum': 6.0, 'max': 3.0, 'average': 2.0}

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Handling of NaN values in metric aggregations</issue_title>
<issue_description>Describe the bug
Not really sure if a bug or a design decision but I could not find this case covered in the tests.

When using numpy to calculate the different operations in get_metric_aggregation the functions chosen will return NaN if there is a value in the metric that is NaN.

For example for:

• sum there is nansum - Treats NaN values as 0
• max there is nanmax - Ignores NaN values
• min there is nanmin - Ignores NaN values
• average there is nanmean - Ignores NaN values (Since there are no weights passed to average it behaves the same as mean)
• percentile there is nanpercentile - Ignores NaN values
• std there is nanstd - Ignores NaN values
• var there is nanvar - Ignores NaN values

To Reproduce
Steps to reproduce the behavior:

1. Pass a metric with a NaN value and several numerical values to get_metric_aggregation
2. Check the result and you should see that it is NaN

Expected behavior
Handle NaN values to return a statistically meaningful number

Additional context
I could submit a PR if you think this is something that can be changed.
It could even be possible to pass a boolean flag handle_nans as an argument to get_metric_aggregation to choose the behavior.
</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Handling of NaN values in metric aggregations #264

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.