A web proxy monitoring runbook turns scattered dashboards and alerts into a repeatable operating process. This guide explains which health, performance, security, and compliance signals to track; how to set practical alert thresholds; how to test failover and bypass controls; and when to review the monitoring design as infrastructure or traffic changes.
Overview
A proxy can be available while still failing its intended control. Users may reach the internet through an unintended route, authentication may fail for only one identity provider, upstream requests may slow down, or logging may stop even though the proxy process remains healthy. Effective proxy uptime monitoring therefore combines service health checks with behavioral tests and operational review.
Start by documenting the proxy’s role and dependencies. A forward proxy or secure web gateway may enforce outbound access rules, identity controls, malware inspection, or data-loss protections. A reverse proxy may handle TLS termination, routing, rate limits, authentication, and API availability. The monitoring plan should state which functions are mandatory, which failures are tolerable, and who owns each response.
Use a layered model:
- Availability: Can approved clients connect to the proxy and receive a valid response?
- Performance: Are connection, processing, and upstream response times within the normal range?
- Control effectiveness: Are authentication, allowlists, blocklists, routing, and inspection rules behaving as intended?
- Security: Are there signs of bypass, abnormal traffic, credential abuse, or unexpected administrative activity?
- Evidence: Are logs, metrics, alerts, and change records available for investigation and review?
For a broader control review, pair this runbook with a reverse proxy security and compliance checklist or the proxy compliance evidence pack guide.
What to track
1. Health and availability
Monitor the proxy from more than one location when practical. A single internal check can miss a regional routing problem, DNS issue, or client-network failure. Useful checks include:
- TCP or TLS connection success to the expected endpoint
- Successful authentication using a dedicated, restricted test identity
- Resolution of a known permitted destination through the proxy
- Expected blocking of a controlled test destination or policy category
- Availability of administrative health and status endpoints, where supported
- Certificate validity, hostname matching, and time remaining before renewal
- Health of dependent identity, DNS, policy, logging, and upstream services
Separate a process check from a transaction check. A process check confirms that a service responds. A transaction check confirms that the complete path works, including authentication, policy evaluation, routing, and logging.
2. Latency, capacity, and errors
Capture latency as distinct stages rather than treating every delay as a proxy problem. At minimum, distinguish connection time, proxy processing time, upstream connection time, and total request time when the platform exposes those fields. Track percentiles such as p50, p95, and p99 if your telemetry supports them; averages can conceal a smaller group of severely affected requests.
Important measurements include request volume, concurrent connections, connection reuse, queue depth, CPU and memory utilization, disk or buffer pressure, upstream failures, timeout counts, DNS errors, TLS negotiation errors, authentication failures, and policy-denied requests. Record the status-code or error categories used by your platform, but avoid assuming that every 4xx or 5xx response has the same cause.
Set initial thresholds from a baseline rather than copying values from another environment. For example, an alert might fire when successful synthetic checks fail twice within five minutes, when p95 latency remains materially above its normal range for ten minutes, or when error rates exceed the documented operating baseline for a sustained interval. These are starting points for testing, not universal limits. Thresholds should reflect business criticality, traffic patterns, and the cost of false alarms.
3. Security and control signals
Monitor events that could indicate bypasses or misuse:
- Direct outbound traffic from managed clients that should use the proxy
- Unexpected changes to proxy endpoints, routing, DNS, or PAC files
- Repeated authentication failures, unusual administrator activity, or privilege changes
- Sudden increases in blocked requests, policy exceptions, or allowlist changes
- Requests from unknown source networks or unexpected geographic locations
- New destinations, protocols, or ports that fall outside the documented use case
- Gaps in audit logs, delayed log delivery, or changes to retention settings
Do not treat a high block count as automatically malicious. It may reflect a new application, a policy deployment, or a broken client configuration. Pair volume alerts with identity, destination, time, and change data.
4. Privacy and compliance evidence
Monitoring should not create unnecessary personal-data exposure. Define which fields are needed for troubleshooting, who can access them, how long they are retained, and whether sensitive URLs, query strings, credentials, or payloads should be redacted. Review the proxy logging and retention policy guide when changing telemetry or retention settings. If a vendor processes proxy logs, document the relevant processing, subprocessors, transfer arrangements, and deletion expectations during vendor review.
Cadence and checkpoints
A monitoring program works best when routine checks are assigned to named roles and recorded in a ticket, operations log, or compliance evidence repository.
Every shift or day
- Review availability, synthetic checks, critical alerts, and unresolved incidents.
- Check whether error, latency, and authentication trends are within the current baseline.
- Confirm that logs are arriving in the monitoring or SIEM destination.
- Review failed health checks for correlated DNS, identity, network, or upstream events.
Weekly
- Sample blocked and allowed requests to confirm policy behavior.
- Review alert noise, duplicate notifications, and alerts that lacked an owner.
- Check certificate, storage, capacity, and license or quota indicators where applicable.
- Inspect changes to allowlists, blocklists, routes, PAC files, and administrative roles.
The allowlist and blocklist policy guide can help structure exception testing and approval records.
Monthly
- Recalculate performance baselines using recent normal traffic.
- Test at least one permitted path, one blocked path, and one authentication failure path.
- Verify alert routing, escalation contacts, maintenance suppression, and recovery notifications.
- Review recurring incidents and convert confirmed patterns into dashboards, runbooks, or permanent fixes.
- Confirm that monitoring data collection remains consistent with the approved logging and retention design.
Quarterly or after material change
- Run a failover or resilience exercise within an approved maintenance window.
- Test whether clients can bypass the proxy through alternate DNS, direct egress, or unmanaged routes.
- Review privileged access and administrative activity using the proxy access-control review workflow.
- Reassess vendor telemetry, health endpoints, SIEM integrations, and supported alert behavior.
- Update diagrams, ownership records, escalation rules, and evidence requirements.
How to interpret changes
Investigate changes by comparing four dimensions: time, scope, dependency, and recent change. A sudden increase in latency immediately after a routing update points to a different investigation than a gradual increase that follows traffic growth. A failure affecting every location suggests a service or dependency issue; a failure affecting one site suggests local DNS, firewall, identity, or network conditions.
Use correlation without assuming causation. For a latency alert, compare proxy processing time with upstream time, DNS resolution, connection counts, and resource utilization. For authentication failures, check identity-provider health, certificate or clock errors, account status, and recent policy changes. For a suspected bypass, compare endpoint configuration, DNS answers, firewall egress logs, and client telemetry.
Classify alerts by impact:
- Critical: The proxy is unavailable for a critical population, a known bypass is active, or a security control has materially failed. Page the on-call responder immediately and open an incident.
- High: A major function is degraded, such as authentication, policy enforcement, logging, or a significant route. Escalate to the service owner and relevant security or network owner.
- Medium: The service remains usable but is outside a sustained baseline, has recurring errors, or has a capacity risk. Assign an owner and deadline for investigation.
- Low: A certificate, dashboard, documentation, or nonurgent configuration issue needs attention. Track it through normal operational work.
Every escalation should include the affected proxy or route, start time, observed symptoms, scope, recent changes, relevant alert links, and the next diagnostic action. Avoid sending raw proxy logs broadly; use the minimum data needed and follow approved access controls.
When to revisit
Revisit this runbook monthly for threshold tuning and at least quarterly for ownership, resilience, and compliance review. Update it sooner after a proxy migration, topology or DNS change, identity integration, policy redesign, new region, major application launch, vendor change, security incident, or repeated false alarm.
Use this practical review worksheet:
- List each critical proxy function and its synthetic test.
- Record the current baseline for availability, latency, errors, traffic, and resource use.
- Mark which alerts are paging, ticketing, or dashboard-only, and name the owner for each.
- Review the last month of incidents, false positives, missed detections, and unowned alerts.
- Confirm that failover, bypass, authentication, policy, and logging tests still reflect the live architecture.
- Check whether telemetry contains more personal or sensitive data than the monitoring purpose requires.
- Document threshold changes, test results, open risks, and the next review date.
Keep the runbook beside the current architecture diagram and escalation roster. A proxy monitoring checklist is valuable only when its checks run, its alerts reach the right people, and its assumptions are reviewed after the environment changes.