Distinguish a rejected Prometheus sample from a failed scrape job

Source facts

Microsoft’s Prometheus troubleshooting table documents different failure scopes. A label name above 511 characters, label value above 1,023 characters or more than 63 labels can fail the entire scrape job. A metric name above 511 characters instead drops the affected series. Labels differing only by case within one sample are treated as duplicates and that sample is dropped during ingestion. Microsoft Learn.

Applicability

Use these documented limits when a target exposes unexpected label shapes or metrics disappear after an exporter change. Establish whether the failure affects one series, one scrape job or a broader collection path before assigning a cause.

DSE recommendation

Inspect the failing sample’s label shape before treating a missing job as a workspace-wide ingestion problem. Have the exporter owner review generated label names, values and counts, especially values assembled from unbounded application input. Preserve a sanitized example and compare it with the documented limit that matches the observed failure. Correct the producer or approved collection configuration rather than merely increasing unrelated workspace capacity.

Verification

In a nonproduction reproduction, compare the target’s scrape status and expected series before and after correcting the offending shape. Check that unrelated metrics in the job return as intended. Record whether the observed result was a whole-job failure or a sample-level drop, and preserve the relevant collector or target diagnostic evidence. Do not infer the scope from an empty dashboard panel alone.

Official references

Microsoft Learn: Troubleshoot Prometheus collection. Source reviewed September 9, 2026.