🗓️ 06112025 0031

MICROMETER TO PROMETHEUS MAPPING

Core Concept: Micrometer metrics transform into Prometheus time series with specific naming conventions, label mappings, and multiple related metrics per instrument.

Why It Matters

Understanding the mapping helps you write correct PromQL queries and debug why metrics appear (or don't appear) in Prometheus. One Micrometer metric often becomes multiple Prometheus time series.

Naming Convention

Micrometer uses dot notation, Prometheus uses underscores:

Micrometer: http.request.duration
Prometheus: http_request_duration_seconds

Rules:

• . → _
• CamelCase → snake_case
• Base unit appended automatically (_seconds, _bytes)

Counter Mapping

Micrometer:

Counter.builder("http.requests")
    .tag("method", "GET")
    .tag("status", "200")
    .register(registry);

Prometheus Export:

# TYPE http_requests_total counter
http_requests_total{method="GET",status="200"} 1543.0

Key Changes:

• Suffix _total added automatically
• Tags → Labels {key="value"}
• Single time series per Counter

Query:

rate(http_requests_total{method="GET"}[5m])

Timer Mapping

Micrometer:

Timer.builder("http.request.duration")
    .tag("endpoint", "/api/users")
    .publishPercentiles(0.95, 0.99)
    .register(registry);

Prometheus Export (Histogram mode):

# TYPE http_request_duration_seconds histogram
http_request_duration_seconds_bucket{endpoint="/api/users",le="0.001"} 0
http_request_duration_seconds_bucket{endpoint="/api/users",le="0.01"} 45
http_request_duration_seconds_bucket{endpoint="/api/users",le="0.1"} 234
http_request_duration_seconds_bucket{endpoint="/api/users",le="1.0"} 456
http_request_duration_seconds_bucket{endpoint="/api/users",le="+Inf"} 500
http_request_duration_seconds_sum{endpoint="/api/users"} 123.45
http_request_duration_seconds_count{endpoint="/api/users"} 500
http_request_duration_seconds_max{endpoint="/api/users"} 2.3

Key Changes:

• One Timer → 4 metric types:
  • _bucket{le="x"} - histogram buckets (cumulative)
  • _sum - total duration
  • _count - total events
  • _max - maximum observed value (resets periodically)
• Base unit converted to seconds
• Labels duplicated across all series

Queries:

# Average duration
rate(http_request_duration_seconds_sum[5m]) / 
rate(http_request_duration_seconds_count[5m])

# 95th percentile
histogram_quantile(0.95, 
  rate(http_request_duration_seconds_bucket[5m])
)

# Request rate
rate(http_request_duration_seconds_count[5m])

Gauge Mapping

Micrometer:

Gauge.builder("jvm.memory.used", memoryBean, MemoryMXBean::getHeapMemoryUsage)
    .tag("area", "heap")
    .baseUnit("bytes")
    .register(registry);

Prometheus Export:

# TYPE jvm_memory_used_bytes gauge
jvm_memory_used_bytes{area="heap"} 536870912.0

Key Changes:

• Base unit suffix added (_bytes)
• Single time series
• Value polled on each scrape

Query:

jvm_memory_used_bytes{area="heap"}
avg_over_time(jvm_memory_used_bytes[5m])

DistributionSummary Mapping

Micrometer:

DistributionSummary.builder("request.payload.size")
    .tag("endpoint", "/api/upload")
    .baseUnit("bytes")
    .register(registry);

Prometheus Export (same structure as Timer):

# TYPE request_payload_size_bytes histogram
request_payload_size_bytes_bucket{endpoint="/api/upload",le="1024.0"} 23
request_payload_size_bytes_bucket{endpoint="/api/upload",le="4096.0"} 67
request_payload_size_bytes_bucket{endpoint="/api/upload",le="+Inf"} 100
request_payload_size_bytes_sum{endpoint="/api/upload"} 234567.0
request_payload_size_bytes_count{endpoint="/api/upload"} 100
request_payload_size_bytes_max{endpoint="/api/upload"} 8192.0

Identical to Timer except no time unit conversion.

LongTaskTimer Mapping

Micrometer:

LongTaskTimer.builder("batch.job.duration")
    .tag("job", "export")
    .register(registry);

Prometheus Export:

# TYPE batch_job_duration_seconds_active_count gauge
batch_job_duration_seconds_active_count{job="export"} 2.0

# TYPE batch_job_duration_seconds_duration_sum gauge
batch_job_duration_seconds_duration_sum{job="export"} 45.3

# TYPE batch_job_duration_seconds_max gauge
batch_job_duration_seconds_max{job="export"} 30.1

Key Changes:

• One LongTaskTimer → 3 gauge metrics:
  • _active_count - currently running tasks
  • _duration_sum - total duration of active tasks
  • _max - longest running task
• All are gauges (not histograms)

Queries:

# Currently running jobs
batch_job_duration_seconds_active_count{job="export"}

# Average duration of active tasks
batch_job_duration_seconds_duration_sum / 
batch_job_duration_seconds_active_count

# Alert on stuck jobs
batch_job_duration_seconds_max > 3600

Tag/Label Mapping

Micrometer Tags → Prometheus Labels

// Micrometer
Counter.builder("requests")
    .tag("method", "GET")      // → {method="GET"}
    .tag("endpoint", "/users") // → {endpoint="/users"}
    .tag("status", "200")      // → {status="200"}
    .register(registry);

Prometheus:

requests_total{method="GET",endpoint="/users",status="200"} 42

Rules:

• Tag keys/values must be strings
• Empty values become ""
• High cardinality tags (user IDs, UUIDs) cause storage explosion

DANGER

Avoid high-cardinality labels like user IDs or request IDs. Each unique label combination creates a new time series.

Base Unit Suffixes

Micrometer automatically adds unit suffixes:

Base Unit	Prometheus Suffix
seconds	_seconds
milliseconds	_seconds (converted)
nanoseconds	_seconds (converted)
bytes	_bytes
kilobytes	_bytes (converted)
megabytes	_bytes (converted)
(none)	(no suffix)

Example:

// Micrometer: baseUnit("milliseconds")
Timer.builder("response.time")
    .baseUnit("milliseconds")
    
// Prometheus: converted to seconds
response_time_seconds 0.123

Percentile Configuration

Client-side percentiles (Summary mode):

Timer.builder("http.duration")
    .publishPercentiles(0.5, 0.95, 0.99)
    .register(registry);

Prometheus Export:

http_duration_seconds{quantile="0.5"} 0.023
http_duration_seconds{quantile="0.95"} 0.145
http_duration_seconds{quantile="0.99"} 0.312
http_duration_seconds_sum 123.45
http_duration_seconds_count 500

Trade-off: Pre-calculated but cannot aggregate across instances.

Server-side percentiles (Histogram mode - default):

Timer.builder("http.duration")
    .serviceLevelObjectives(
        Duration.ofMillis(10),
        Duration.ofMillis(50),
        Duration.ofMillis(100),
        Duration.ofMillis(500)
    )
    .register(registry);

Prometheus Export:

http_duration_seconds_bucket{le="0.01"} 12
http_duration_seconds_bucket{le="0.05"} 45
http_duration_seconds_bucket{le="0.1"} 234
http_duration_seconds_bucket{le="0.5"} 456
http_duration_seconds_bucket{le="+Inf"} 500

Query with histogram_quantile() for flexible percentiles.

Common Naming Patterns

Micrometer	Prometheus	Type
http.requests	http_requests_total	Counter
http.request.duration	http_request_duration_seconds_*	Timer → Histogram
jvm.memory.used	jvm_memory_used_bytes	Gauge
cache.size	cache_size	Gauge
request.size	request_size_bytes_*	DistributionSummary → Histogram
batch.job.duration	batch_job_duration_seconds_active_count etc.	LongTaskTimer → 3 Gauges

Cardinality Impact

Each unique label combination = new time series

// BAD: High cardinality
Timer.builder("api.duration")
    .tag("userId", userId)        // 1M users
    .tag("requestId", requestId)  // Infinite
    // → Millions of time series → OOM

// GOOD: Low cardinality
Timer.builder("api.duration")
    .tag("endpoint", "/api/users")  // ~100 endpoints
    .tag("method", "GET")           // 4-5 methods
    .tag("status", "200")           // ~10 status codes
    // → ~5000 time series → Manageable

Quick Reference

// Counter: metric_total
Counter → metric_name_total{labels}

// Timer: 4 metrics
Timer → metric_name_seconds_bucket{labels,le}
        metric_name_seconds_sum{labels}
        metric_name_seconds_count{labels}
        metric_name_seconds_max{labels}

// Gauge: single metric
Gauge → metric_name_unit{labels}

// DistributionSummary: same as Timer
DistributionSummary → metric_name_unit_bucket{labels,le}
                      metric_name_unit_sum{labels}
                      metric_name_unit_count{labels}
                      metric_name_unit_max{labels}

// LongTaskTimer: 3 gauges
LongTaskTimer → metric_name_seconds_active_count{labels}
                metric_name_seconds_duration_sum{labels}
                metric_name_seconds_max{labels}

---

References

• Micrometer Prometheus Documentation
• Prometheus Naming Best Practices