INFO, LATENCY, SLOWLOG, MEMORY, CLIENT LIST, MONITOR Caveats, and Exported Metrics

Learning outcomes

This lesson turns INFO, LATENCY, SLOWLOG, MEMORY, CLIENT LIST, MONITOR Caveats, and Exported Metrics into an observable AtlasMart workflow with explicit correctness, failure, and production boundaries.

1. Problem: the alert says “Redis latency,” but which layer is slow?

AtlasMart checkout reports a p99 response-time regression. Redis may be executing commands slowly, the client may be queued behind its own connection pool, the network may be delayed, a persistence fork may pause the process, or the application may simply be missing its cache and waiting on another database. Observability means collecting enough independent evidence to distinguish those mechanisms. A service-level objective (SLO) is the target reliability or latency level for a service; an error budget is the amount of allowed failure relative to that objective.

2. Evidence hierarchy: start with cheap snapshots

Use server metadata before intrusive tools. INFO is sectioned and machine-readable; commandstats reports calls and execution time by command, latencystats reports command latency percentiles, memory exposes allocator/process state, replication exposes role/offsets, and keyspace exposes logical-database counts. Redis 8.8+ also adds slow-log aggregate fields to command statistics. These are server-side observations; they do not include the application network round trip.

3. Capture a before-change observability bundle

docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO serverdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO statsdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO commandstatsdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO latencystatsdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO memorydocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO replicationdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin INFO keyspace

Expected evidence includes redis_version:8.10.1, standalone role, command counters, latency percentiles, used-memory/RSS values, and DB 0 key counts. A clean snapshot proves the node state at capture time; it does not prove the absence of short spikes between samples.

4. SLOWLOG answers “which commands blocked Redis?”

The Slow Log records commands whose server execution time crosses slowlog-log-slower-than. Redis explicitly excludes client I/O from that duration, so a slow application request with no Slow Log entry can still be caused by network, client queues, serialization, or another upstream service. Save the original threshold before changing it and restore it after the lab.

OLD_SLOW=$(docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --raw --user academy-admin CONFIG GET slowlog-log-slower-than | tail -1)docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin CONFIG SET slowlog-log-slower-than 0docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-App-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user atlasmart-app SET atlasmart:ch24:l1:probe valuedocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin SLOWLOG GET 5docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin CONFIG SET slowlog-log-slower-than "$OLD_SLOW"

5. LATENCY captures server events that command statistics alone can miss

Redis latency monitoring is disabled by default when latency-monitor-threshold is 0. When enabled, the framework records named event classes such as command, fork, eviction, expiration, and AOF-related events. LATENCY LATEST returns samples; LATENCY DOCTOR turns those samples into a human-readable diagnosis. Choose the threshold from the workload SLO, not a universal number.

OLD_LAT=$(docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --raw --user academy-admin CONFIG GET latency-monitor-threshold | tail -1)docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin CONFIG SET latency-monitor-threshold 1docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin LATENCY LATESTdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin LATENCY DOCTORdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin CONFIG SET latency-monitor-threshold "$OLD_LAT"

6. MEMORY and CLIENT LIST explain pressure around the command engine

MEMORY STATS complements INFO memory, while CLIENT LIST exposes connection identity, age/idle time, authenticated user, RESP version, query buffers, output buffers, and library metadata where the client supplies it. A large omem or tot-mem can reveal slow consumers even when commands themselves are fast.

docker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin MEMORY STATSdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin MEMORY DOCTORdocker exec -e REDISCLI_AUTH=AtlasMart-Ch24-Admin-Lab-Only-2026 atlasmart-redis-ch24 redis-cli --user academy-admin CLIENT LIST

7. MONITOR is an emergency microscope, not a dashboard

MONITOR also does not log every administrative command, and sensitive authentication data is specially handled. Absence from MONITOR is therefore not evidence that no administrative/security event occurred.

8. Export metrics without confusing collection with interpretation

A production metrics system normally scrapes a supported Redis/managed-service exporter. The free local lab instead demonstrates a small JSON snapshot so the evidence pipeline stays reproducible without adding another daemon. The important design is stable metric names, timestamps, labels such as node/role, and a retention policy—not the choice of dashboard product.

import json, time, redisr = redis.Redis(host="127.0.0.1", port=6441, username="academy-admin", password="AtlasMart-Ch24-Admin-Lab-Only-2026", decode_responses=True)info = r.info()payload = {    "captured_at_unix": time.time(),    "redis_version": info.get("redis_version"),    "used_memory": info.get("used_memory"),    "used_memory_rss": info.get("used_memory_rss"),    "connected_clients": info.get("connected_clients"),    "instantaneous_ops_per_sec": info.get("instantaneous_ops_per_sec"),    "keyspace_hits": info.get("keyspace_hits"),    "keyspace_misses": info.get("keyspace_misses"),    "evicted_keys": info.get("evicted_keys"),}print(json.dumps(payload, indent=2, sort_keys=True))

9. Evidence matrix: what each tool proves

10. Production judgment and bridge

Start incidents with low-cost evidence, correlate client and server time, and increase intrusiveness only when the hypothesis requires it. Tail latency, memory/buffer growth, persistence events, replication state, security identity, and workload shape belong in the same timeline. The next lesson decomposes latency mechanisms rather than treating every spike as a Redis command problem.

Summary and next step

INFO, LATENCY, SLOWLOG, MEMORY, CLIENT LIST, MONITOR Caveats, and Exported Metrics is now connected to observable Redis behavior, bounded failure cases, and production tradeoffs. Keep the evidence and cleanup state from this lesson; next, continue with Latency Doctor, Intrinsic Latency, Fork Pauses, Big Commands, Network, CPU, and Storage Diagnostics.

Authoritative references

• Info
• Latency
• Latmon
• Slowlog
• Client
• Monitor
• Benchmark
• Persistence
• Release
• Cli
• Memory