Merancang Ketersediaan Tinggi

What is High Availability (HA)? It's about designing systems that keep running even when parts fail. Erlang and OTP are built from the ground up to achieve this.

Imagine a critical service like an online store. If it goes down, sales are lost! HA aims to minimize downtime, ensuring your application remains operational and accessible to users.

Achieving High Availability relies on several key design principles:

• Redundancy: Having multiple components capable of performing the same task.
• Fault Tolerance: The ability to continue operating despite failures.
• Automatic Recovery: Systems that detect failures and recover or switch automatically.
• No Single Point of Failure (SPOF): Eliminating any component whose failure would bring down the entire system.

The Active-Passive pattern, also known as Hot Standby, involves one primary (active) component and one or more secondary (passive) components.

The active component handles all requests. If it fails, a passive component takes over, becoming the new active. This provides redundancy and minimizes downtime, but the passive component is idle until needed.

We can simulate an active-passive setup using Erlang processes and monitors. Here, a 'standby' process monitors a 'primary'. If the primary dies, the standby takes over. This is a simplified example of role switching.

Try running this code:

In an Active-Active pattern, multiple components are simultaneously active, sharing the workload. This offers both redundancy and improved scalability by distributing tasks.

If one active component fails, the others continue processing requests, often with a slight performance degradation. This setup requires careful state management and load balancing to ensure requests are distributed efficiently.

A major challenge in HA is maintaining consistent state across redundant components. If an active component fails, its replacement needs access to the most up-to-date information.

Strategies include:

• Replication: Copying state changes to standby or other active components (e.g., using Mnesia or custom replication logic).
• Shared Storage: Storing state in a highly available external database accessible by all nodes.
• Stateless Design: Making components stateless, so any instance can handle any request without needing prior state.

A Single Point of Failure (SPOF) is any part of a system whose failure would stop the entire system from working. Identifying and eliminating SPOFs is crucial for HA.

Common SPOFs include:

• A single database server.
• A single network switch.
• A central coordinator process without a backup.

Design your system with redundancy at every critical layer, from hardware to software components.

To enable automatic recovery and failover, components need a way to detect if others are still alive and healthy. This is done through heartbeating and health checks.

• Processes can send periodic "I'm alive" messages.
• Monitors can detect process crashes immediately (as seen in our example).
• Nodes can monitor other nodes using net_kernel:monitor_nodes/1 for cluster-wide health.

Sometimes, even in an active-active system, a single coordinator or "leader" is needed to manage a shared resource or ensure global consistency. If this leader fails, a new one must be chosen.

Leader Election is the process of dynamically selecting a new leader from a set of potential candidates in a distributed system. Erlang's global module can help with simple global registration, but for robust election algorithms, custom solutions or libraries are often used.

We've explored how to design highly available Erlang OTP systems:

• Understood the pillars: redundancy, fault tolerance, automatic recovery, and no SPOF.
• Examined Active-Passive and Active-Active patterns.
• Discussed state replication and consistency.
• Learned about heartbeating and leader election concepts.

By applying these advanced OTP principles, you can build robust, resilient applications that remain operational even in the face of failures.