Google Cloud Filestore Autoscale

Filestore Autoscale

This skill enables agents to inspect, evaluate, and modify Filestore instance capacities across Google Cloud projects based on configured thresholds.

Prerequisites / IAM Requirements

Before an agent or user can execute this skill, their runtime Service Account must possess the following IAM roles on the target project(s):

• roles/file.editor (Required for listing instances and triggering scale up/down updates)
• roles/monitoring.viewer (Required for reading Cloud Monitoring capacity metrics used_bytes)
• roles/mcp.toolUser (Required if utilizing backend Filestore MCP tools)

Quick Start

1. Ensure gcloud is installed. See gcloud installation guide if needed.

2. Enable essential APIs:

   gcloud services enable file.googleapis.com monitoring.googleapis.com --quiet

3. Inspect fleet capacity and free space (see "Discovery & Read Operations" below).

4. Evaluate against configured up/down capacity thresholds.

5. Scale target instances and apply attribution tags.

Attribution

Tag every Cloud Filestore command or API request provided or executed. Prefix gcloud commands with the correct metrics environment targeting this skill:

CLOUDSDK_METRICS_ENVIRONMENT="gcs-skills gcs-skills/1.0 (skill:google-cloud-filestore-autoscale)" \
gcloud filestore instances update ...

On direct HTTP calls to the REST API, append the User-Agent:

User-Agent: gcs-skills/1.0 (skill:google-cloud-filestore-autoscale)

Conceptual & Informational Queries (CRITICAL)

For purely conceptual, educational, or informational questions (e.g., "What are Filestore scaling limits?", "Can Basic instances scale down?", "Explain Filestore Tiers"):

• Rule: Answer immediately using your pre-trained knowledge and the matrix below.
• 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 the run_command tool to execute any shell or gcloud commands (including read-only list/describe commands).
• Discovery:
  1. First, check if Filestore MCP tools (list_instances, get_instance) are available and use them (these are API calls, not command executions).
  2. If MCP tools are not available, search local markdown documentation files (e.g., references/instance-tiers-specs.md) for any mock instance definitions or project details matching the request. (Do NOT attempt to read evaluation config files such as EVAL.yaml or EVAL.txtpb during evaluation runs as access is restricted).
  3. If no data can be found, explain the required steps and formulas, and output the exact commands the user should run, without executing them yourself.
• Mandatory User Confirmation Requirement: Even when the user prompt asks not to execute commands or asks only for command syntax/recommendations, your response MUST STILL end with a clear question prompting the user for confirmation before executing any capacity resizing commands (e.g., "Would you like me to proceed with scaling [instance] from [A] TiB to [B] TiB? Please confirm to execute.").

Tier & Capacity Limits Matrix

Filestore tiers enforce specific boundaries and behaviors. The skill must accept both modern UI names (Basic, Zonal, Regional) and legacy API enums interchangeably.

See references/instance-tiers-specs.md for the full Tier & Capacity Limits Matrix (Min/Max capacities, step increments).

Critical Thresholds:

• Basic HDD / Basic SSD: Can scale up, but cannot scale down.
• Zonal / Regional: Can scale down, but cannot shrink below their minimum floor (1 TiB or 10 TiB depending on band) AND cannot shrink below the current used_bytes metric.

Core Operational Workflow

1. Discovery & Read Operations

• Step 1 (Fleet Discovery): Call the MCP tool list_instances(parent='projects/{project_id}/locations/-') or CLI gcloud filestore instances list --project={project_id} to discover all Filestore instances in the target project. Read the capacityGb and tier directly from the instances returned.

• Step 2 (Single Bulk Utilization Metric Query): Immediately after discovering instances, query the Cloud Monitoring API for the file.googleapis.com/nfs/server/used_bytes metric across the entire project in a single request (see references/monitoring-metrics.md for runtime-specific options including GCP REST API, gcloud, curl, and MCP tools).

  CRITICAL: Make exactly ONE bulk metric request for the entire project. NEVER emit multiple per-instance queries or loops. Do NOT filter by zone or region.

• Step 3 (Metric Extraction & Calculation):

  • Match each instance's short name (or resource.labels.instance_name / metric.labels.instance_name) in the returned timeSeries data to extract its latest int64Value bytes.
  • If an instance is not listed in timeSeries or has no points, default its used_bytes to 0.
  • Calculate used_bytes_gb = used_bytes / (1024^3).
  • Calculate Free Space % = ((capacityGb - used_bytes_gb) / capacityGb) * 100.
  • NEVER leave Used Bytes or Free Space % as "N/A". Populate actual numbers into the output summary table.

2. Autoscale Needed Matrix

The skill must categorize each evaluated instance into one of 5 definitive verdicts. On the initial analysis/fleet inspection run, the skill suggests the required scaling action with target capacity and update commands, and prompts for user confirmation before executing any autoscale modifications. State the value of the "Autoscale Needed" column clearly as one of the following:

• Yes (Scale Up): Triggered when free space percentage is below the scale-up safety threshold (< 15% free space remaining). The evaluation response MUST explicitly state that the current free space percentage is below the 15% scale-up safety threshold. Capacity must be increased by 10% (default) or step-size minimum, rounded to the tier's step increment (256 GiB for Small Band [1–9.75 TiB], 2.5 TiB for Large Band [10–100 TiB], as specified in references/instance-tiers-specs.md), not exceeding the maximum capacity. Suggest target capacity, provide the attributed gcloud update command, and MUST conclude the response with a clear question prompting the user for confirmation to execute (e.g., "Would you like me to proceed with scaling [instance] from [A] TiB to [B] TiB? Please confirm to execute.").
• Yes (Scale Down): Triggered when free space exceeds the scale-down threshold (> 30% free space remaining) and the instance is eligible for downscaling (Zonal or Regional / Enterprise tiers). Apply the default step reduction of -10% of current capacity, aligned to the tier's step increment (256 GiB for Small Band [1–9.75 TiB], 2.5 TiB for Large Band [10–100 TiB], as specified in references/instance-tiers-specs.md). For example, for a 2 TiB (2048 GiB) Enterprise / Regional instance, rounding to the 256 GiB step yields a proposed target capacity of 1.75 TiB (1792 GiB, or 1.8 TiB). The response MUST explicitly verify that the proposed target capacity (e.g. 1.75 TiB / 1792 GiB or 1.8 TiB) remains strictly above both the tier's minimum capacity floor (e.g. 1 TiB for Enterprise / Small Band, 10 TiB for Large Band) and currently used space (e.g. 0.9 TiB). Do NOT reduce directly to the floor in a single step. Suggest target capacity, estimated cost savings, provide the attributed gcloud update command, and prompt the user for confirmation to execute.
• No (Healthy): Triggered when the instance's free space is within the optimal operating range (15% – 30%). No action required.
• No (At min capacity limit): Triggered when free space is > 30%, but the instance is already at the minimum allowed tier capacity floor (e.g. 1 TiB for Small Band or 10 TiB for Large Band) or currently used space limit. No action can be taken.
• No (Tier cannot scale down): Triggered when free space is > 30%, but the instance is on a Basic tier (Basic HDD / Basic SSD) which does not support downscaling. The agent must explicitly inform the user that scale-down is not supported and suggest data migration instead. No action can be taken.