Google Cloud Filestore Log Troubleshooting

<!-- disableFinding(LINE_OVER_80) -->

Filestore Log-Based Troubleshooting

Diagnoses, troubleshoots, and remediates Filestore client mount failures, permission errors (EACCES), and network timeouts (ETIMEDOUT) across projects.

Prerequisites & Quick Start

Required IAM roles on target project(s) (and Shared VPC host project if applicable):

• Read: roles/file.viewer (instance & export ACLs), roles/compute.networkViewer (VPC firewall rules), roles/logging.viewer (Cloud Audit & GKE CSI logs), roles/mcp.toolUser (if using MCP tools).
• Write (Remediation only): roles/file.editor (export ACL updates), roles/compute.securityAdmin (firewall rule creation).

Authenticate, verify billing/APIs, and configure your environment:

gcloud auth login && gcloud auth application-default login
gcloud billing projects describe {project_id} --format="value(billingEnabled)"
gcloud services enable file.googleapis.com compute.googleapis.com logging.googleapis.com --quiet
gcloud config set project {project_id} && gcloud config set compute/region {region}

Attribution

Prefix every gcloud command provided or executed with the skill metrics environment:

CLOUDSDK_METRICS_ENVIRONMENT="gcs-skills gcs-skills/1.0 (skill:google-cloud-filestore-log-troubleshooting)" \
gcloud filestore instances describe ...

On direct REST API calls, append HTTP header: User-Agent: gcs-skills/1.0 (skill:google-cloud-filestore-log-troubleshooting).

Conceptual & Informational Queries (CRITICAL)

For purely conceptual, architectural, or educational questions (e.g., "What causes EACCES on Filestore?", "Why does GKE Node IP appear instead of Pod IP?", "What ports does Filestore require?", "Explain root squash"):

• Rule: Answer immediately using pre-trained knowledge and the workflow rules below. Answering directly minimizes tool latency and token usage when the user only seeks architectural guidance.
• Constraint: Do not execute external tool calls or API requests for basic knowledge questions.

Handling "No-Command" Constraints (CRITICAL)

If the user prompt contains constraints like "Do not execute commands", "without executing", or "read-only":

• Rule: Strictly avoid calling run_command to execute any shell, Python, or gcloud commands.
• Discovery:
  1. First, check if Filestore MCP tools (get_instance, list_instances) are available and use them (API calls, not command executions).
  2. If MCP tools are unavailable, read references/mock-fleet-data.md only if the requested instance matches one of the evaluation scenarios (finance-share, shared-nfs, ml-data, data-hub, prod-share). Never report mock data as live production state. If the instance is not listed there, state that live access is required and provide the exact commands for the user to run.
  3. Fast-Path Stop Rule: Once you locate the target instance in references/mock-fleet-data.md, stop reading additional files immediately and formulate your response. Do NOT read scripts/quick_diagnose.py, scripts/diagnose_lib.py, _internal/quick_diagnose_test.py, or EVAL.* files when command execution is disabled, as inspecting code/test files wastes turns and triggers timeouts.
  4. Explain the required diagnostic steps and output the exact attributed commands for manual execution.
• Mandatory User Confirmation Requirement: Even when command execution is disabled or the user asks only for recommendations, your response MUST STILL end with a clear question prompting the user for explicit confirmation before applying any remediation (e.g., "Would you like me to proceed with creating the VPC ingress firewall rule [rule_name]? Please confirm to proceed.").

Multi-Runtime Execution Options

Option 1: Bundled Python CLI Script (Recommended for CLI / Terminal Agents)

# Single instance diagnosis
python3 scripts/quick_diagnose.py --instance="<INSTANCE_ID>" --location="<LOCATION_OR_ZONE>" \
    [--project="<PROJECT_ID>"] [--client-ip="<CLIENT_IP>"] [--client-subnet="<CLIENT_SUBNET_CIDR>"]

# Bulk project-wide fleet diagnosis
python3 scripts/quick_diagnose.py --all --project="<PROJECT_ID>" [--json]

Option 2: Filestore MCP Tools / REST API / gcloud CLI

• MCP Tools: Call get_instance(name='projects/{project_id}/locations/{location}/instances/{instance_id}') and inspect fileShares[0].nfsExportOptions and networks[0].network.
• Native REST API (call_gcp_api): Invoke service="file", version="v1", resource_path="projects/{project_id}/locations/{location}/instances/{instance_id}".
• Standard gcloud:

  CLOUDSDK_METRICS_ENVIRONMENT="gcs-skills gcs-skills/1.0 (skill:google-cloud-filestore-log-troubleshooting)" \
  gcloud filestore instances describe {instance_id} --location={location} --project={project_id} --format=json

Core Operational Workflow

Step 1: Parameter Extraction & GKE Node Architecture

• Extract instance, location (region/zone), project, and client_ip / client_subnet. If target parameters are missing, list instances or ask the user to confirm.
• GKE Architecture Rule: PersistentVolumes are mounted by the Linux kernel on the GKE Worker Node, not inside the Pod network namespace. Even if Pod IPs (10.4.0.0/14) are allowlisted, the NFS server sees traffic originating from the GKE Node Internal IP. Always evaluate and allowlist the GKE Node Subnet CIDR in nfsExportOptions and VPC firewall rules.