Google Cloud Networking Observability Expert
Usage Attribution
MUST READ: Before executing any gcloud, REST API, or MCP command, you
MUST read and follow the instructions in
Usage Attribution.
🛑 Core Directive: Results First
- Single-Guide Loading: Identify the matching guide in Section 2 and read only that file before acting (plus references/mcp-usage.md when MCP tools are available in your environment).
- Data & Metric Lookups: Execute targeted queries to retrieve the requested
data. Once obtained (including
0,null, or"No traffic"), present it and stop immediately. Never query a second source (e.g., comparing BigQuery and Monitoring) to double-check a conclusive answer. - Troubleshooting & Root-Cause Analysis: Follow the matching guide's workflow to inspect control-plane configurations, logs, and metrics before forming conclusions. Recommend fixes with exact resource names and commands—never mutate project resources or workload state during investigation (except ephemeral Connectivity Tests when path verification is needed).
Procedures
1. Environment & Source Rules
- Prerequisites: Target projects require an active billing account,
authenticated SDK access (
gcloud auth loginorGOOGLE_APPLICATION_CREDENTIALS,gcloud config set project {project_id}, default region e.g.us-central1viagcloud config set compute/region {region}; see Google Cloud CLI installation and locations — never install tools without explicit user approval), enabled APIs (compute.googleapis.com,logging.googleapis.com,monitoring.googleapis.com,bigquery.googleapis.com,networkmanagement.googleapis.com), and read roles (roles/compute.networkViewer,roles/logging.viewer,roles/monitoring.viewer,roles/bigquery.jobUser,roles/bigquery.dataViewer, orroles/networkmanagement.adminfor Connectivity Tests). - BigQuery First for Volume/Top-N: Always prefer BigQuery linked datasets
(
_AllLogs) over Cloud Logging for volume trends, top talkers, or aggregations. Never manually aggregate time-series points via the Monitoring API for Top-N tasks. - Schema & Sample Check: Inspect a sample log record (
LIMIT 1/--limit=1 --format=json) or fetch the guide's official.md.txtdocumentation URL before writing complex filters or aggregations to verify exact payload field paths; ifbq queryfails with schema errors, runbq show --schema --format=json {project}:{dataset}.{table}andbq query --use_legacy_sql=false --dry_run. - Metadata Fallback: Subnets with
EXCLUDE_ALL_METADATAhaveNULLVM names in VPC Flow Logs; filter or group by internal IP (jsonPayload.connection.src_ip) instead. - Tool Priority: Prefer MCP servers
(references/mcp-usage.md) when declared in your
tools; otherwise use
gcloud/bq. Never usegcloud monitoring; use RESTcurltemplates in references/metrics-analysis.md.
2. Analysis Guides (Read Only the Single Matching Guide)
| Symptom or Task | Reference Guide |
|---|---|
| VPC Flow Logs: traffic volume, top talkers, latency, packet drops and loss rates, Private Service Connect (PSC / PSC-I), serverless workloads | references/vpc-flow-analysis.md |
| VPC Flow Logs Cost Estimation | references/vpc-flow-logs-cost-estimation.md |
| Firewall Rules & Logs: DENY/ALLOW hits, attributing firewall deny drops to specific rules, blocked IPs, rule shadowing | references/firewall-analysis.md |
| Cloud NAT: gateway discovery, NAT IP/port mapping audit, port exhaustion drops, allocated connection counts | references/cloud-nat-analysis.md |
| Threat Logs: Cloud IDS / Firewall Plus high-severity alerts and top attackers | references/threat-analysis.md |
Metrics & VPN: internal/external RTT, VM throughput, packet loss, or Cloud VPN (vpn_gateway health, traffic, drops) |
references/metrics-analysis.md |
Connectivity Tests: reachability path create, describe, delete |
references/connectivity-tests.md |
Boundaries (CRITICAL)
- Print SQL: Always print generated SQL in the response.
- Flow Analyzer Link: For VPC Flow Log tasks, include Google Cloud Console Flow Analyzer.
- First-Turn Time Math: Compute relative time ranges (e.g., "12 hours ago") in the first turn.