Repository navigation
Using None as a label value breaks exposure of all metrics #1011
Description
If a metric is added and provided a label value of None, the response generated for scraping raises an AttributeError, preventing any attribute from being scraped, not just the misconfigured one.
Sample code:
from prometheus_client import Info, generate_latest
info_metric = Info("metric", "description")
info_metric.info({"label": None})
_ = generate_latest()
Output:
Traceback (most recent call last):
File "/home/mol/git/tango/test_metrics.py", line 7, in <module>
_ = generate_latest()
File "/home/mol/miniconda3/lib/python3.10/site-packages/prometheus_client/exposition.py", line 282, in generate_latest
output.append(sample_line(s))
File "/home/mol/miniconda3/lib/python3.10/site-packages/prometheus_client/exposition.py", line 239, in sample_line
['{}="{}"'.format(
File "/home/mol/miniconda3/lib/python3.10/site-packages/prometheus_client/exposition.py", line 240, in <listcomp>
k, v.replace('\\', r'\\').replace('\n', r'\n').replace('"', r'\"'))
AttributeError: ("'NoneType' object has no attribute 'replace'", Metric(metric, description, info, , [Sample(name='metric_info', labels={'label': None}, value=1.0, timestamp=None, exemplar=None)]))
The misconfigured metric not getting exposed is fine -- None is not a valid label value. But the other metrics not getting exposed either is painful. There's several easy ways to prevent this, any of those would do:
- detect label values of
None(or non-strings) and raise a ValueError when they are set - cast label value to string (
"None"in this case) before parsing them as a string (usestr(k)instead ofkin the traceback below) - filter out misconfigured metrics when generating the HTTP response
Since error checking is light in prometheus_client and duck-typing is a thing, one of the last two options seem most attractive?
Activity