Prometheus scraping fails with negative untyped values

[boomer41]

Bug Report

Describe the bug

The following happens with both prometheus_textfile and prometheus_scrape inputs.
When having untyped metrics, parsing the prometheus text fails when said metrics have negative values.

Example:

# HELP node_zfs_arc_memory_available_bytes kstat.zfs.misc.arcstats.memory_available_bytes
# TYPE node_zfs_arc_memory_available_bytes untyped
node_zfs_arc_memory_available_bytes -8.58914944e+08

fluent-bit outputs:

[error] [input:prometheus_textfile:prometheus_textfile.0] error parsing file prom.txt: 'cmt_untyped_set failed'

To Reproduce

• Put above prometheus text in a file called prom.txt
• Use the following configuration:

service:
  log_level: debug
  daemon: off

pipeline:
  inputs:
    - name: prometheus_textfile
      tag: prom
      path: prom.txt
      scrape_interval: 1

  outputs:
    - name: stdout
      match: '*'

Expected behavior

Metrics to be successfully scraped.

Screenshots

Your Environment

• Version used: 5.0.7
• Configuration: YAML, see above
• Environment name and version: Debian Package on Trixie
• Server type and version: Debian
• Operating System and version: Trixie
• Filters and plugins: prometheus_textfile

Additional context

We are using Fluent Bit to scrape local metrics and push them via Open Telemetry to a central monitoring instance.
When a single metric with above values exist, the whole exporter fails causing our Monitoring to go haywire.

[boomer41]

I have isolated the problem to https://github.com/fluent/fluent-bit/blob/master/lib/cmetrics/src/cmt_untyped.c#L115
I will create an issue in the cmetrics repo.