Incident Management: Operational Resilience

Incident Management is the process used by DevOps and SRE teams to respond to unplanned events or service interruptions and restore service as quickly as possible. In a high-maturity SRE organization, this isn't just about "fixing the bug"—it's about forensic precision and systemic immunity.

1. Incident Lifecycle and Severities

Every incident must be categorized to dictate the response level:

SEV-1 (Critical): Core service is down for all users (e.g., Checkout is broken). Immediate "War Room" required.
SEV-2 (High): Significant degradation for a subset of users. Response within 30 minutes.
SEV-3 (Medium): Minor bug or non-critical feature failure. Resolved during business hours.

2. The 'Golden Hour' Forensics

In SRE, the Golden Hour refers to the first 60 minutes of a major incident. Actions taken here determine the Mean Time to Recovery (MTTR) and the quality of the subsequent Root Cause Analysis (RCA).

A. Non-Destructive Mitigation

The primary goal is restoration, but the secondary goal is Preservation.

Avoid 'The Big Reboot': Indiscriminately restarting all pods/instances destroys volatile state (memory leaks, thread deadlocks, cache corruption) that is essential for RCA.
Isolation First: Use traffic shifting (canary bypass) or circuit breakers to isolate the failing component rather than killing it.

B. Forensic Data Capture Checklist

Before applying a patch or restarting, capture the following state:

Thread Dumps & Heap Dumps: For JVM/Go/Node processes, capture the state of concurrency and memory.
Kernel/Network State: Use ss -atp (socket state), dmesg, and conntrack tables to identify network saturation or resource leaks.
eBPF Probes: Use tools like bpftrace to capture syscall latency or file I/O spikes that metrics might aggregate away.
Log Snapshot: Tail and preserve the last 10,000 lines of application and system logs before they are rotated or lost.

3. The Response Role: Incident Commander (IC)

The IC is the single source of truth during a SEV-1. They do not write code; they manage the Flow of Information.

Scribe: Documents the timeline in real-time (Slack/Zoom).
Communications Lead: Handles updates to status pages and executive stakeholders.
Operations Lead: Coordinates the engineers actually applying the fix.

4. The PACE Communication Plan

Relying on one channel is an operational risk.

Primary: Slack / Microsoft Teams.
Alternate: Zoom / Google Meet Bridge.
Contingency: Phone bridge / SMS.
Emergency: StatusPage.io (External to infrastructure).

5. Root Cause Analysis (RCA) Patterns

A "Blameless" culture is the prerequisite for effective RCA. We move from "Who did it?" to "Why did the system allow this?".

A. The Five Whys (Systemic Depth)

Human Error: "Engineer ran a destructive SQL command."
Technical Root Cause: "The staging and production environments had identical CLI prompts."
Systemic Failure: "The database access layer lacks a 'read-only by default' policy for interactive shells."

B. RCA Methodology: The Three Perspectives

Proximate Cause: The immediate trigger (e.g., a deployment, a configuration change, a spike in external traffic).
Underlying Cause: The vulnerability that made the trigger catastrophic (e.g., missing retry limits, lack of timeout on a dependency).
Systemic Cause: The organizational or architectural debt that allowed the vulnerability to exist (e.g., lack of automated integration testing for failure modes).

C. Common RCA Anti-Patterns to Avoid

Correlation Fallacy: Assuming two spikes on a dashboard are causal without verifying the sequence of events.
Pointing to 'The Human': Stopping at "human error" is the ultimate failure of incident management. It leaves the trap set for the next engineer.
Dashboard Bias: Looking only at existing metrics. If the metrics were sufficient, the incident might have been predicted or automatically mitigated.

6. Closing the Loop: Action Items

Every post-mortem must result in prioritized tickets.

Detection: "Add alert for P99 latency on Auth service."
Mitigation: "Implement auto-circuit breaker on downstream DB."
Prevention: "Update CI/CD pipeline to reject deployments with missing health checks."

See Also:

Game Day Exercises — Practicing the response.
Monitoring and Alerting — Detecting the incident.
Graceful Degradation — Limiting incident impact.
Blameless Post-Mortems — Cultural foundations.