One slow database can take down twenty healthy services, and the code that finishes them off is the code you wrote to protect them.
This video treats timeouts, retries, backoff with jitter, retry budgets and circuit breakers as ONE load-control policy, not four features. None of them add capacity to the thing that's failing; they decide who waits, who retries, and who gets told no.
🔧 Free diagram pack (editable D2 + SVG of every diagram in this video): https://github.com/liverandomm/systems-diagram-packs/tree/main/01-retries-timeouts-circuit-breakers
CHAPTERS
0:00 Cold open: one slow database
0:21 The frame: three questions
0:48 Anatomy of a slow call
1:43 Timeouts (and deadline propagation)
2:45 Retries and the multiplication problem
3:54 Backoff and jitter
4:38 Retry budgets
5:26 Circuit breakers: closed, open, half-open
6:16 The skeptic's view of breakers
6:59 Postmortem: Google Cloud, June 2025
7:40 One policy, not four features
8:35 Diagram pack + next video
SOURCES (primary)
- AWS Builders' Library — Timeouts, retries, and backoff with jitter (Marc Brooker): https://aws.amazon.com/builders-library/timeouts-retries-and-backoff-with-jitter/
- AWS Architecture Blog — Exponential Backoff And Jitter: https://aws.amazon.com/blogs/architecture/exponential-backoff-and-jitter/
- Google SRE book, ch. 22 Addressing Cascading Failures: https://sre.google/sre-book/addressing-cascading-failures/
- Google SRE book, ch. 21 Handling Overload: https://sre.google/sre-book/handling-overload/
- Martin Fowler — CircuitBreaker: https://martinfowler.com/bliki/CircuitBreaker.html
- Resilience4j CircuitBreaker docs: https://resilience4j.readme.io/docs/circuitbreaker
- AWS SDKs — Retry behavior (retry quota / token bucket): https://docs.aws.amazon.com/sdkref/latest/guide/feature-retry-behavior.html
- Envoy — CircuitBreakers / retry_budget: https://www.envoyproxy.io/docs/envoy/latest/api-v3/config/cluster/v3/circuit_breaker.proto
- Google Cloud incident report, 2025-06-12: https://status.cloud.google.com/incidents/ow5i3PPK96RduMcb1SsW
Numbers on screen come from those sources (80.4% bimodal-latency example, 243× layered retries, 10% retry budget ≈1.1× vs ≈3×, Resilience4j defaults). Latency histograms and client timelines are labeled illustrations.
Narration uses a synthetic (text-to-speech) voice. Diagrams are built from code (D2) and animated programmatically; no AI-generated images.
#SRE #Microservices #DistributedSystems
