Resilience Patterns (Retry & Circuit Breaker)

Fluxion leverages Resilience4j across enrichment operators ($httpCall, $sqlQuery) and streaming connectors. This guide consolidates the available options, usage tips, and testing helpers.

---

1. Retry options

Field	Meaning	Default
name	Identifier for the retry instance. Pipelines sharing the same name also share retry state.	Derived from operator/connector name
maxAttempts	Total attempts (initial call + retries).	1
waitDurationMs	Fixed backoff between attempts (milliseconds).	1000
multiplier	> 1.0 enables exponential backoff (waitDurationMs as base).	1.0
retryOn	List of exception class names to include.	[]
ignore	Exceptions to fail fast without retrying.	[]

Example

{
  "retry": {
    "name": "orders-db-retry",
    "maxAttempts": 3,
    "waitDurationMs": 50,
    "multiplier": 2.0,
    "retryOn": ["java.sql.SQLTransientException"],
    "ignore": ["java.sql.SQLException"]
  }
}

---

2. Circuit breaker options

Field	Meaning	Default
enabled	Enables/disables the breaker.	true
name	Identifier for the breaker instance. Shared names share state.	Derived from operator/connector name
failureRateThreshold	Percentage of failures that opens the breaker.	50
slowCallRateThreshold	Percentage of slow calls counted as failures.	100
slowCallDurationThresholdMs	Milliseconds defining a slow call.	60000
waitDurationInOpenStateMs	Time to stay OPEN before half-open probes.	10000
permittedCallsInHalfOpenState	Trial calls while HALF_OPEN.	2
minimumNumberOfCalls	Minimum sample size before metrics apply.	10
slidingWindowSize	Size of metrics window.	100
slidingWindowType	0 for count-based, 1 for time-based.	1
record / ignore	Additional exception class names to record/ignore.	[]

Example

{
  "circuitBreaker": {
    "name": "user-service-breaker",
    "failureRateThreshold": 30,
    "slowCallRateThreshold": 50,
    "slowCallDurationThresholdMs": 2000,
    "waitDurationInOpenStateMs": 30000,
    "minimumNumberOfCalls": 5,
    "permittedCallsInHalfOpenState": 3,
    "slidingWindowSize": 20,
    "slidingWindowType": 0
  }
}

---

3. Usage guidelines

• Keep retries low for idempotent operations; avoid retrying non-idempotent writes without compensating actions.
• Use unique name values per downstream system to isolate metrics; reuse names only when pipelines intentionally share state.
• Combine retry + circuit breaker for noisy/slow dependencies (retry handles transient spikes, breaker protects against sustained failure).
• Surface breaker events to your monitoring/alerting stack (Resilience4j emits metrics you can wire into Micrometer/Prometheus).

---

4. Testing utilities

• $httpCall: HttpCallOperator.resetResilience() resets retry/breaker state in tests.
• $sqlQuery: SqlQueryOperator.resetResilience() does the same for SQL.
• Streaming connectors expose similar reset hooks if needed (Kafka/EventHub).

Run enrichment tests to validate configuration:

mvn -pl fluxion-enrich -am test

---

5. References

Doc	Notes
$httpCall	Shows resilience usage in HTTP calls.
$sqlQuery	SQL enrichment with retry/breaker.
Resilience4j	Official docs.

Apply these settings consistently across operators and connectors to tune resilience without scattering configuration details.