Metrics to track: RED, USE and business metrics with Prometheus

Backend Live Coding November 13, 2026 python

For services, RED: Rate, Errors, Duration (p50/p95/p99, not the average).

python 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
# ======================================================================
# Q: Which metrics do you measure? Write an example.
#
# For services, RED: Rate, Errors, Duration (p50/p95/p99, not the average). For
# resources, USE: Utilization, Saturation, Errors. And business metrics: imports
# per hour, rejected rows, queue length, delay from event to alert.
# Watch out for cardinality: never user_id or tenant_id as a label in Prometheus
# (millions of time series). Those go to logs/traces.
# ======================================================================

from prometheus_client import Counter, Histogram

IMPORT_ROWS = Counter('import_rows_total', 'Rows processed',
                      ['source', 'result'])          # result: ok|rejected
EXTERNAL_LATENCY = Histogram('external_api_seconds', 'External API latency',
                             ['provider'])

with EXTERNAL_LATENCY.labels(provider='geoip').time():
    data = client.lookup(ip)
IMPORT_ROWS.labels(source='edr', result='ok').inc(len(good))

Related Posts

Python shallow copy vs deep copy
Python shallow copy vs deep copy
Python closures and late binding in lambdas
Python closures and late binding in lambdas
Storing heartbeats from 250,000 screens with Redis write coalescing
Storing heartbeats from 250,000 screens with Redis write coalescing
Writing a retry decorator with exponential backoff in Python
Writing a retry decorator with exponential backoff in Python
SQL: what a CTE is and how to find duplicate records
SQL: what a CTE is and how to find duplicate records
Threads vs processes vs asyncio: when to use each
Threads vs processes vs asyncio: when to use each