Connected infrastructure

Interfaces and recovery plans that survive ordinary failures.

Interfaces

Choose an API boundary

An API exposes a responsibility that another system can rely on. A useful boundary follows a stable business capability rather than the current arrangement of database tables. Internal storage can then change without requiring every caller to change with it.

Interfaces
Plan a compatible interface change

An interface change is compatible only relative to the clients that use it. Adding a field can still break a strict parser, and adding an enum value can break an exhaustive switch. Compatibility needs evidence about real client behaviour.

Interfaces
Keep pagination consistent

Pagination divides a large result into smaller responses. If the underlying records change between requests, a caller may otherwise miss a record or see it twice. The ordering and consistency model are part of the interface contract.

Interfaces
Share an end-to-end time budget

A request can cross several services before returning a response. Independent generous timeouts at every hop can make the total wait much longer than the caller expects. A deadline makes the remaining budget visible along the path.

Interfaces
Treat a webhook as an untrusted message

A webhook lets another system notify an endpoint when something changes. The public endpoint receives input from outside its trust boundary. Authentication, replay handling and input limits are therefore part of the message design, not optional finishing work.

Interfaces

Reliability

Separate readiness from liveness

A running process is not necessarily ready to serve traffic. Conversely, a temporary dependency problem does not always mean restarting the process will help. Distinct health signals let the platform choose between waiting, removing traffic and restarting.

Reliability
Stop feeding a failing dependency

Repeated calls to an unhealthy dependency consume time and resources while adding pressure to the failing component. A circuit breaker temporarily rejects some calls after a defined failure pattern, then permits controlled probes to see whether service has recovered.

Reliability
Drain work before stopping

Stopping a process can interrupt requests, lose unacknowledged work or cause clients to retry a completed action. A graceful shutdown sequence gives the process a bounded opportunity to finish or return ownership of its work before it exits.

Reliability
Budget a scarce resource

A service can fail before its CPU is full. Connections, memory, queue slots or a downstream quota may become the limiting resource first. Capacity planning starts by identifying which resource bounds the workload and how demand grows with concurrency.

Reliability
Rehearse a small recovery

A recovery document is a hypothesis until someone follows it. A bounded drill checks whether people, access and procedures work together under realistic conditions. The goal is to discover missing steps before an outage makes every minute expensive.

Reliability

Search the library