How do you choose metric names and tags without causing a cardinality problem?

Builds stable Micrometer dimensions that answer operational questions without unbounded time series and cost.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I use a stable metric name for the operation and a small set of bounded tags such as outcome, method, dependency name, or normalized route. I never put user IDs, order IDs, raw URLs, exception messages, or other unbounded values in metric tags. Each unique tag combination creates another time series, which can increase memory, storage, query cost, and alert noise.

Full answer below

Practice focus

Meters, tags, low cardinality, naming conventions, URI templates, percentiles, cost, and governance.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
A useful metric groups many events into a bounded number of time series; identifiers destroy that aggregation.
I use a stable metric name for the operation and a small set of bounded tags such as outcome, method, dependency name, or normalized route. I never put user IDs, order IDs, raw URLs, exception messages, or other unbounded values in metric tags. Each unique tag combination creates another time series, which can increase memory, storage, query cost, and alert noise.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
For HTTP metrics I use route templates such as /orders/{id}, not the requested path. I keep tag values from controlled vocabularies and estimate the product of their possible values before shipping. High-cardinality values can belong in sampled traces or structured logs, with privacy controls, rather than metrics. Histograms and percentiles also have cost, so I enable them for important latency signals rather than every meter.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Tagging by customer or request ID.
Using raw exception messages as tag values.
Adding dimensions without estimating combinations.

How do you choose metric names and tags without causing a cardinality problem?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue practice

Try a variant

Tech stacks

Skill areas

Suggest an improvement

Found this answer useful?