Challenges with distributed systems

Ideas you will meet

Distributed Systems

Consensus, time, failure and the fundamentals of scale-out.

Resilience

Timeouts, retries, circuit breakers, bulkheads and load shedding.

What we already know about this

Our own lessons on these ideas. Read them here before you start, so it makes sense.

• Advanced Distributed Systems · 13 minResilience: timeouts, retries, bulkheads, backpressure and circuit breakersHow to stop one slow or failing dependency from taking everything down: the patterns, how they fit together and where each one goes wrong.Read hereClose

  The core problem

  In a system of many services, slowness is more dangerous than failure. A dependency that fails fast is easy to handle. One that responds in 30 seconds makes callers hold threads, connections and memory until they too stop responding: a cascading failure. Resilience patterns limit how far a problem can spread.

  The toolbox

  100%

  Figure SDA-ADS-06-1 · Resilience layers around a call · © forcsguy.com

  Timeouts

  Every network call needs a timeout, including connect, read and overall. Pick from measured latency (for example slightly above the 99th percentile), not a guess, and make the caller's budget smaller than its own caller's so work is abandoned in an orderly way. Propagate deadlines across calls.

  Retries

  Retry only transient failures and only idempotent operations (or use idempotency keys). Use exponential backoff with jitter (random spread) so clients do not retry in lockstep, cap the attempts, and apply a retry budget (for example no more than 10 percent extra traffic). Retry at one layer, not at every layer, or attempts multiply.

  Circuit breaker

  A breaker watches failures or slow calls. Closed: normal. After too many failures it opens: calls fail immediately (or use the fallback) so the dependency can recover and callers stop wasting resources. After a cool-down it goes half-open: a few trial calls decide whether to close again.

  100%

  Figure SDA-ADS-06-2 · Circuit breaker states · © forcsguy.com

  Bulkheads

  Partition resources so one dependency cannot consume them all: separate thread pools or connection pools per dependency, separate queues per tenant or per priority, or separate instances for critical paths. A slow recommendation service then cannot starve checkout.

  Backpressure and load shedding

  If producers outpace consumers, work piles up. Without limits the buffer grows until memory is exhausted or latency becomes huge. Options:

  • Bounded queues that reject or block when full.
  • Signalling upstream to slow down (credit-based flow in streaming systems, TCP windows, reactive streams).
  • Load shedding: deliberately drop low-priority work early (return 429 or 503 quickly) to protect the rest.
  • Rate limiting at the edge (see the rate-limiter lesson).

  Failing fast and visibly under overload is better than accepting everything and failing slowly.

  Fallbacks and graceful degradation

  Return something useful when a dependency fails: cached data, a default, a simpler algorithm, or hide a non-essential section. Decide which features are essential before an incident, and test the degraded mode.

  How they interact

  • A timeout is the minimum; without it nothing else works.
  • Retries inside a breaker inside a bulkhead is the usual nesting.
  • Everything needs metrics: breaker state, rejected calls, queue depth, timeout counts. Alert on them.
  • Test it: inject latency and failures (chaos experiments) to confirm the patterns work before production does it for you.

  Common mistakes

  Retries with no backoff; timeouts longer than the caller's; unbounded queues; a shared thread pool for everything; breakers that never close; fallbacks that call the same failing service; and testing only the happy path.

  Real-world reading

  Related reading

  • Timeouts, retries, and backoff with jitterAmazon Builders' Library · Amazon Web ServicesBlogs
  • Using load shedding to avoid overloadAmazon Builders' Library · Amazon Web ServicesBlogs
  • Circuit BreakerMartin FowlerBlogs
  • Circuit Breaker patternMicrosoft · Microsoft AzureDocumentation

  Show 6 moreShow fewer

  • Google SRE Book: Addressing Cascading FailuresGoogle SRE team · GoogleBlogs
  • Chaos Monkey (source)Netflix · GitHubGitHub
  • Bulkheads: Partition and Tolerate Failure in One PartSkife engineers · SkifeCase Studies
  • Protector: Circuit Breaker for Time Series Databases at TrivagoTrivago engineers · TrivagoCase Studies
  • Autoscaling Jenkins at TrivagoTrivago engineers · TrivagoCase Studies
  • Troubleshooting a Connection Timeout Issue with tcp_tw_recycle Enabled at eBayeBay engineers · eBayCase Studies

  Open this lesson with its quiz and exercise

• Messaging and Microservice Patterns · 13 minReliable messaging: retries, dead letters, ordering and the outboxThe patterns that turn an at-least-once broker into a correct system: idempotent consumers, retry policies, dead letter queues, ordering and the transactional outbox.Read hereClose

  The honest baseline

  Most brokers give at-least-once delivery: nothing is lost, but a message can arrive more than once and, with competing consumers, out of order. A correct system is built on that, not on wishes.

  Idempotent consumers

  An operation is idempotent if doing it twice has the same effect as once. Ways to get there:

  • Natural idempotency: "set status to PAID" is safe to repeat; "add 10 to balance" is not.
  • Deduplication table: record the message id (or an idempotency key) in the same transaction as the work; skip if already seen.
  • Conditional updates: update only if the version or state matches.

  async function handle(msg: { id: string; orderId: string }) {
    await db.tx(async (t) => {
      const fresh = await t.insertIfAbsent("processed_messages", { id: msg.id });  // unique key on id
      if (!fresh) return;                                // duplicate: do nothing
      await t.markOrderShipped(msg.orderId);             // the real work, in the same transaction
    });
    // acknowledge only after the transaction commits
  }
  

  Acknowledge after the work commits; if the process dies in between, the broker redelivers and the dedup record makes it safe.

  Retries and dead letter queues

  100%

  Figure SDA-MSV-02-1 · Retry and dead-letter flow · © forcsguy.com

  • Separate transient failures (a dependency is down: retry later) from permanent ones (invalid payload: send to the DLQ straight away).
  • Use exponential backoff with jitter, and cap retries.
  • A poison message retried forever blocks a partition or hogs a worker; the DLQ prevents it.
  • Alert on DLQ depth and age; build a safe replay tool.
  • In Kafka, which has no per-message acknowledgement, implement retry and DLQ with extra topics ("retry-5s", "dlq").

  Ordering

  Total order across a system is expensive. Aim for ordering only where needed, usually per entity.

  • Partition by key (all events for one order go to one partition and stay ordered).
  • A queue with competing consumers can reorder; use FIFO queues with message groups, or one consumer per key.
  • Make handlers tolerate late or out-of-order events with versions or timestamps ("ignore if older than what I have").
  • Retries break order: a failed message retried later arrives after its successors. Decide whether to block the key or tolerate reordering.

  The dual-write problem and the outbox

  A service must often change its database and publish an event. Doing both separately is unsafe: if the database commit succeeds and publishing fails, the event is lost; if the order is reversed, an event may announce something that did not happen.

  100%

  Figure SDA-MSV-02-2 · Transactional outbox · © forcsguy.com

  The inbox side mirrors it: consumers record processed message ids in their own database transaction.

  Before: write the database, then publish

  100%

  After: transactional outbox

  100%

  What changed

  • The order and its event are saved in one transaction
  • A crash can no longer leave an order without an event
  • Consumers must still tolerate duplicates

  Putting the guarantees together

  • Outbox gives "event published if and only if the data changed" (at least once).
  • Idempotent consumers make duplicates harmless.
  • DLQ and monitoring make failures visible.
  • Per-key ordering where order matters.

  Together these give effectively exactly-once results without relying on a mythical exactly-once network.

  Other practical points

  • Message design: include an id, a type, a version and a timestamp; keep payloads small or reference large data; evolve schemas compatibly (add optional fields; use a schema registry).
  • Backpressure: bounded queues and consumer concurrency limits (see the resilience lesson).
  • Visibility timeouts and acknowledgement deadlines must exceed processing time, or the broker redelivers work still in progress.
  • Monitoring: queue depth, age of the oldest message, consumer lag, DLQ size, processing time.

  Real-world reading

  Related reading

  • Exactly-once Semantics Is Possible: Here's How Apache Kafka Does ItNeha Narkhede · ConfluentBlogs
  • Making retries safe with idempotent APIsAmazon Builders' Library · Amazon Web ServicesBlogs
  • Pattern: Transactional outboxChris Richardson · microservices.ioDocumentation
  • Confluent Developer: Apache Kafka 101ConfluentCourses

  Show 6 moreShow fewer

  • Confluent: Apache Kafka TalksConfluentYouTube
  • Apache Kafka DocumentationApache Kafka · Apache Software FoundationDocumentation
  • Avoiding insurmountable queue backlogsAmazon Builders' Library · Amazon Web ServicesBlogs
  • 10 Tips for Building Resilient Payment SystemsShopify Engineering · ShopifyCase Studies
  • Challenges with distributed systemsAmazon Builders' Library · Amazon Web ServicesBlogs
  • Resiliency and Disaster Recovery with Kafka at eBayeBay engineers · eBayCase Studies

  Open this lesson with its quiz and exercise