google-cloud-filestore-autoscale

SKILL.md

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

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

Enable essential APIs:

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

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

Evaluate against configured up/down capacity thresholds.

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:

First, check if Filestore MCP tools (list_instances, get_instance) are available and use them (these are API calls, not command executions).

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).

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.

Output Format

Every status report, evaluation, or recommendation response MUST include a markdown table summarizing the evaluated instances. Even if evaluating a single instance, format it as a table.
The table MUST contain the following columns:

Instance

Service Tier

Provisioned Capacity

Used Bytes

Free Space %

Autoscale Needed (MUST contain one of: Yes (Scale Up), Yes (Scale Down), No (Healthy), No (At min capacity limit), or No (Tier cannot scale down))

Example standard output table:

| Instance | Service Tier | Provisioned Capacity | Used Bytes | Free Space % | Autoscale Needed | Proposed Action |
|---|---|---|---|---|---|---|
| `[instance-name]` | REGIONAL | 2048 GiB | 900 GiB | 56.05% | Yes (Scale Down) | Scale down to 1792 GiB. `CLOUDSDK_METRICS_ENVIRONMENT=... gcloud filestore instances update ...` |

3. Execution & Confirmation Workflow

Analysis & Recommendation (First Run / Inspection):

Calculate step-aligned target capacity adhering to tier ceilings, floors, and basic scale-up only rules.

Present the summary table and proposed actions.

MANDATORY USER CONFIRMATION PROMPT: Whenever recommending target capacity or providing a gcloud filestore instances update command, your response MUST explicitly include a clear question asking the user to confirm execution before any modifications are made (e.g. "Would you like me to proceed with scaling [instance] from [A] TiB to [B] TiB? Please confirm to execute.") to prevent accidental billing spikes or capacity exhaustion.

Do not execute autoscale commands without user confirmation.

Execution upon Confirmation:

Once the user confirms (e.g., "Yes, proceed with scaling", "Scale instance X"), execute the attributed gcloud filestore instances update command on the confirmed instance(s).

Fallback:

If execution fails due to Prod mutation restrictions, output the failure reason and provide the user with the exact attributed gcloud command to run manually, reminding them to confirm before manual execution.

Custom Thresholds

When the user configures or passes custom threshold values in prompts (e.g.
"Scale up if free space drops below 10% with a 20% step", or custom
max_threshold / up_increment):

Global Session Memory Confirmation: The response MUST accept and
acknowledge the custom thresholds and MUST explicitly confirm that custom
thresholds apply globally across projects in session memory, explicitly
mentioning the target project IDs evaluated or active in session memory to
prevent accidental cross-project misconfiguration.

Configuration Summary: The response MUST display the updated active
configuration summary showing all active thresholds and step increments.

Preserve Overrides: The response MUST NOT revert to default thresholds
(15% / 10%) when custom overrides are provided.

Reference Directory

For progressive disclosure of deeper topics, consult the references/ directory:

Instance Tiers & Specs

Monitoring Metrics Formulas

Troubleshooting & Errors