The Guardian software engineering recommendations

This repository document principles, standards and guidelines.

Principles

Security principles

Service should follow the principle of least privilege regarding user access and consider the sensitivity of any data or operations the service makes available. For example a service which provides read-only data on what's for lunch in the canteen may be accessed by all Guardian staff, but a service that publishes content to our site must be tightly controlled to a known set of users.
Service should follow a zero-trust approach. Service should not rely on the network level (3) only to grant privilege or make assumption but rather ensure each request is authenticated at application level (7), and may use multiple parameters (location, time, user, authentication method, behaviours) to grant or deny access.
Service should employ multiple techniques and have defense in depth: typesafe language, virtual machine, operating system hardening, code reviews, integration tests, safe libraries, code static analysis, vulnerabilities scanning, user input sanitising, browsers built-in restrictions, encryption, audits and pentesting.
All services should be registered in a service catalog with defined technical owners, state of the service and links to related artefacts.
All services in production should be maintained and kept up to date to latest version in alignment with vulnerabilities management standards and with the aim to reduce exposure windows to a minimum.
Technical debt should be actively managed and reduce in line with risk and business strategy.

Reliability principles

Publish services availability SLA to communicate expectations to users or dependent systems and identify area of improvements.

Design for simplicity and single responsibility. Great designs model complex problems as simple discrete components. Monitoring, self-healing and graceful degradation are simpler when responsibilities are not conflated.
Design for failures. All things break, so the behaviour of a system when any of its components, collaborators or hosting infrastructure fail or respond slowly must be a key part of its design.
Ensure applications are swim-laned to enable graceful degradation and prevent cascading failure. Communication via idempotent events should be preferred over remote procedure calls (RPC) as they allow us to replay events in case of failure. When sending event use at-least-once semantic to ensure a message will always be sent even in case of a failure.
Software and infrastructure are not separated concerns. The simplest and most reliable solutions often require changes to both, so a team must be responsible for both and they should be designed together.
Leverage the vendors's recovery and resilience features in the underlying platforms in preference to implementing custom processes. Vendors SLAs usually provide higher levels of availability than could be achieved using our own processes:
- For example, prefer using RDS rather than hosting databases on our own AWS instances. With RDS, AWS manages patching, backups and replication of data to a standby instance in a different Availability Zone (AZ), providing resilience to data centre outages.
- For example, we use use BigQuery which SLA details a monthly uptime percentage superior to 99.99%.
Out of hours support. Team are responsible for their services but there is additional 24/7 compensated support to deal with unexpected issues.

Scalability principles

Design for horizontal scaling a system that can be scaled by adding additional servers provides simpler scalability and greater robustness.
Scaling state is complex avoid state where possible and where it is unavoidable, have a clear plan for how it can scale including CAP considerations, or why scaling is unnecessary.
Service should provide coherent features set with explicit documentation including SLA for use by other services.

Operational principles

All services should be using for defined common standards for infrastructure management, deployment, configuration management, logging, and observability.
Workflow and processes should be automated to enable users to self-serve based on their needs without unneeded time consuming interactions.
Deployment of changes, including infrastructure, should be automated, repeatable, auditable and continuous.
All services should be monitored to alert on business or technical issue and reduce time to restore service.
Reducing time to detect and restore service (availability to users) should be optimised rather than number of failures (reliability of services).

Data principles

Service should be designed with data minimisation and data protection by design and default principles.
Data is classified and protected according to classification and associated standard regarding access, storage and protections.
Service should implement data retention policy in accordance with data governance standards.
Service should ensure compliance with GDPR, PECR, CCPA, and others data privacy regulations when creating or maintaining service through following defined processes (e.g Data Council), by using existing libraries and by integrating with existing capabilities for data ingestion (Fivetrans), transformation (DBT), storage (BigQuery) and orchestration of end users privacy requests (Baton).

guardian / recommendations

The Guardian software engineering recommendations

Principles

Security principles

Reliability principles

Scalability principles

Operational principles

Data principles

Guidelines

About

Languages