Dependency Reliability

Most production systems depend on databases, queues, third-party APIs, identity providers, and shared platforms. Reliability depends on how those dependencies fail and recover.

Dependency inventory

For each critical dependency, record:

• Owner and support path.
• Expected availability and latency.
• Timeout and retry behavior.
• Quotas, rate limits, and concurrency limits.
• Failure mode and customer impact.
• Fallback, cache, queue, or degraded behavior.
• Monitoring and alert ownership.

Failure controls

Common controls include:

• Timeouts shorter than user-facing deadlines.
• Retries with jitter and bounded attempts.
• Circuit breakers for repeated failures.
• Bulkheads for resource isolation.
• Queues for async buffering.
• Caches for safe stale reads.
• Feature flags to disable optional dependency paths.

Review map

Watchouts

• Default SDK retries can exceed user-facing timeouts.
• Shared dependencies can create correlated failures.
• Optional dependencies often become required accidentally.
• Third-party SLAs do not replace internal failure design.