Learn how to describe and discuss fault injection techniques in chaos engineering: latency injection, error injection, resource exhaustion, and network partition.
0 / 5 completed
1 / 5
A team says: 'We are going to inject latency into the payment service dependency.' What will they do?
Latency injection is one of the most common and valuable chaos experiments: real production failures often manifest as slow dependencies rather than hard errors. Latency injection tests whether callers have appropriate timeouts, circuit breakers, and fallback logic. 'Injecting 2 seconds of latency' means adding a 2-second artificial delay to the targeted service's responses — not crashing it, not returning errors, just slowing it down.
2 / 5
What is the correct way to describe an error injection experiment?
Error injection means making a service return specific error responses (HTTP 4xx/5xx, gRPC error codes, etc.) at a defined rate — without actually breaking the underlying service. The description must include: the error type (HTTP 500), the injection rate (30%), the target (recommendation service), and the expected resilience behavior being tested (graceful degradation on the product page). 'Turning off the database' is not controlled error injection — it is a broader, less targeted disruption.
3 / 5
An engineer reports: 'We observed resource exhaustion on the primary node.' What happened?
Resource exhaustion experiments test a different failure mode from latency or error injection: they simulate the resource constraint failures that happen in production due to memory leaks, runaway processes, disk-filling logs, or connection pool exhaustion. When reporting, specify the resource type: 'We exhausted available memory on the primary node to 95% utilization' or 'We filled the disk to 100% to test log rotation alerts and database failover.'
4 / 5
What is a 'network partition' and how is it typically described in chaos engineering?
A network partition (also called a split-brain scenario) blocks network communication between specific components. It is more targeted than latency injection — no traffic reaches the destination rather than slow traffic. Description should specify: which services are partitioned from each other, the protocol/port affected, and the expected behavior being tested (fallback, failover, error handling). Network partitions are particularly important for distributed systems, databases, and consensus algorithms.
5 / 5
Which sentence correctly compares two fault types?
Distinguishing fault types is important for experiment design: latency injection tests timeout configuration, circuit breaker sensitivity, and degradation under slow dependencies. Error injection tests error handling code paths, fallback logic, and retry behavior. Resource exhaustion tests alerting, auto-scaling, and graceful degradation under resource pressure. Network partition tests split-brain handling, failover, and distributed consensus. Each reveals different resilience gaps.