cloud-logging-cross-project-configuration

Files (2)

• SKILL.md
• attribution.md · references/

SKILL.md (first lines)

# Configuring Cross-Project Logging

This skill describes how to use `gcloud` commands to configure Cloud Logging so
that you store log data in a central location, regardless of the point of
origin. The skill also describes how to query log data when that data is stored
in multiple projects.

> [!IMPORTANT] **Sandbox Network Limitation (CRITICAL for Agent Testing):**
> During evaluation or in restricted sandboxed environments, network traffic to
> GCP APIs is blocked. Do **NOT** run network discovery commands to find
> resource names, project IDs, or organization IDs. Always use the exact project
> IDs or placeholders provided in the user prompt or instructions, for example,
> `{project_id}`, `{source_project_id}`, `{central_project_id}`. Assume these
> resources exist and proceed directly with configuration commands. Running
> these discovery commands will cause the execution to hang and timeout.

## Attribution

**MUST READ**: Before executing any `gcloud`, REST API, or MCP command, you
MUST read [Usage Attribution](references/attribution.md). For commands executed
directly in the terminal (never in generated scripts, Terraform, or user code),
prefix `gcloud` with
`CLOUDSDK_METRICS_REQUEST_ATTRIBUTION="agent-skills/1.0 (skill:cloud-logging-cross-project-configuration)"`
and append `agent-skills/1.0 (skill:cloud-logging-cross-project-configuration)` to `User-Agent` headers.

## Safety and Confirmation Tiers (CRITICAL)

Before executing any commands on behalf of the user, you MUST adhere to the
following safety tiers based on the action requested:

1.  **Tier R: Read-Only**
    *   **Description:** Commands that only read state or query logs.
    *   **Example commands:**
        *   `gcloud logging read`
        *   `gcloud logging buckets list`
    *   **Rule:** No confirmation needed. You may execute these commands
        immediately to gather information.
2.  **Tier M: Mutation (Non-Billing)**
    *   **Description:** Configuration modifications or free metadata creations
        that do not incur direct storage or billing costs and do not affect
        resource security/access policies.
    *   **Example commands:**
        *   `gcloud logging views create`
        *   `gcloud logging views update`
        *   `gcloud logging scopes create`
        *   `gcloud logging buckets create`
    *   **Rule:** No confirmation needed. You may execute these commands
        immediately to apply configurations.
3.  **Tier B: Billing and Security-Sensitive Mutations (High-Risk)**
    *   **Description:** Operations that create billing-inducing resources or
        integrations, or modify security and IAM access control policies
        (presenting a risk of privilege escalation).
    *   **Example commands:**
        *   `gcloud logging metrics create`
        *   `gcloud logging links create`
        *   `gcloud projects add-iam-policy-binding`
    *   **Rule:** **Interactive confirmation required.** These commands create
        resources that incur billing costs or alter security access. You MUST
        present the exact, literal command and receive user confirmation before
        executing. NEVER execute in the same turn as asking.
4.  **Tier D: Causes irreversible data loss**
    *   **Description:** Actions that permanently discard or delete logs, for
        example sink exclusions.
    *   **Example commands:**
        *   `gcloud logging buckets delete`
        *   `gcloud logging sinks update --add-exclusion`
    *   **Rule:** **Explicit typed confirmation required.** These commands
        discard or delete logs immediately and irreversibly, or they may result
        in log data not being stored. You MUST ask for explicit typed
        confirmation, for example, "Yes, discard logs", and halt execution until
        the user replies.

## Decision Matrix: Centralized Storage vs. Distributed Storage with Read-Time Aggregation

Use this decision matrix to evaluate and choose between **Centralized Storage**
and **Read-Time Aggregation**. With centralized storage, log data is routed to
one log bucket, regardless of where the data originates. You write queries
against the centralized log bucket. With read-time aggregation, log data is
stored by the resource where it originates. However, a single query aggregates
the data by querying all resources.

After you have determined the optimal architecture for handling cross-project
logs, follow the corresponding configuration steps detailed below.

| Criterion             | Centralized Storage      | Read-Time Aggregation    |
| :-------------------- | :----------------------- | :----------------------- |
| **GCP Project Scale** | Scales to thousands of   | Best for < 375 projects. |
:                       : projects.                :                          :
| **Log Storage**       | Consolidated in a single | Resides in originating   |
:                       : log bucket.              : resources.               :
| **SQL Analytics**     | Easy; unified querying   | Hard; requires querying  |
:                       : via Observability        : multiple log buckets.    :
:                       : Analytics.               :                          :
| **Access Control**    | Scoped access via log    | Requires IAM access to   |
:                       : views on the centralized : all views on resources   :
:                       : log bucket.              : that store log data.     :
| **Configuration**     | Options vary based on    | Will not interfere with  |
: **Complexity**        : Project, Folder,         : bucket-based log-based   :
:                       : Organization structure.  : metrics.                 :
| **Cost**              | Potential for duplicate  | Cost-effective; no data  |
:                       : storage of log buckets   : replication.             :
:                       : if exclusions aren't     :                          :
:                       : set.                     :                          :

## Architecture

### Centralized Storage (Log Routing)

```mermaid
graph LR
    subgraph "Source Project(s)"
        Log[Resource Logs] --> Sink["Sink: route-to-central-project"]
    end

    subgraph "Central Project"
        Sink --> Bucket["Bucket: central-logs-bucket (us-central1)"]
    end
```

From google/skills, Apache-2.0.

More from google/skills