Chapter 7 checkpoint

Chapter 7 checkpoint: asynchronous processing and events

Every question from this chapter’s lessons. Your first answer to each is the one that counts; afterward you can practice again as often as you like. Passing needs 75% right on the first attempt.

0 of 17 answered0 right on the first attempt

Question 1 of 17

A server has 200 worker threads, and each request holds a thread for 5 seconds while a PDF renders. At most how many requests per second can it sustain?

Your answerRPS

Answers within 2% count.

Question 2 of 17

An upload endpoint now puts a transcoding job on a queue instead of doing the work. What should it answer straight away?

Nothing until the job finishes202 Accepted, with a URL where the job’s status can be checked503 Service Unavailable, until a worker is free200 OK, after waiting for the transcode to finish

Question 3 of 17

Match each system to how it treats messages.

Amazon SQS or RabbitMQApache KafkaCelery

Question 4 of 17

A Kafka topic has 12 partitions, and a consumer group runs 20 consumers on it. How many of them are actually consuming?

Your answerconsumers

Answers within 1% count.

Question 5 of 17

A payment consumer reads charge events from an at-least-once queue. How should it make sure a redelivered event never charges a customer twice?

Switch the queue to exactly-once deliveryRecord the message id under a unique key in the same database transaction as the chargeCheck a Redis key with SET NX and skip the event if it existsAcknowledge each message before processing it

Question 6 of 17

Match each delivery guarantee to what it means.

At most onceAt least onceExactly once

Question 7 of 17

A consumer first sets a Redis key for each message id with SET NX, then applies the payment in the database. After a crash, some payments were never applied, though their messages were redelivered. Why?

The key was set before the payment committed, so retries skipped messages that had never been appliedRedis lost the keysThe broker did not redeliver the messagesThe messages arrived out of order

Question 8 of 17

Each customer’s deposits and transfers must be processed in order, but customers are independent of each other. How do you get that ordering without a single bottleneck?

Send every event through one queue with one consumerPartition events by customer idSort events by timestamp in each consumerSpread events randomly across partitions

Question 9 of 17

For 10 seconds, 12,000 jobs per second arrive. There are 20 workers, each handling 500 jobs per second. Afterward arrivals drop to 2,000 per second. How many seconds does the backlog take to drain?

Your answer

Answers within 5% count, in any unit you choose.

Question 10 of 17

One malformed message makes the consumer of a Kafka partition throw every time it reads it, and nothing behind that message is processed. What fixes it?

After a set number of failed attempts, move it to a dead-letter queue and continueRetry it fasterDelete and recreate the topicAdd consumers to the group

Question 11 of 17

A client retries with full jitter: it waits a random time between 0 and

\min(M, B \cdot 2^i)

, with base

B = 100

ms, cap

M = 30

s and

i

the attempt number. What is the longest it can wait on attempt

i = 5

, in seconds?

Your answer

Answers within 2% count, in any unit you choose.

Question 12 of 17

Why add random jitter to exponential backoff?

To preserve message orderingSo each retry happens soonerSo clients that failed together do not all retry at the same momentTo send failures to the dead-letter queue sooner

Question 13 of 17

A service inserts an order into its database and then publishes an OrderPlaced event to Kafka from application code. What can go wrong?

A crash between the two steps leaves an order that downstream services never hear aboutNothing, if both calls are awaitedThe only risk is publishing the event twicePublishing to Kafka is too slow for an order flow

Question 14 of 17

Put the steps of the transactional outbox in order.

A relay reads the outbox, by polling or from the change log
The relay publishes the event to the broker
Commit the transaction
Consumers apply the event idempotently
Insert the order and an outbox row in one transaction

Question 15 of 17

What is event sourcing?

Writing an outbox row with each business updateReading the database’s change log to publish eventsUsing separate models for reads and writesStoring every change as an event and treating that sequence as the source of truth

Question 16 of 17

When does Martin Fowler suggest using CQRS?

Whenever you need transactionsIn every microservice, as a defaultOnly on parts of a system where the read and write needs really differNever

Question 17 of 17

Producers publish 12,000 events a second, 1 KB each, to a Kafka topic with 6 partitions. One consumer group reads them and writes each to a database. Which component falls behind?

ProducersPublish 12,000 events/s

Kafka topic, 6 partitionsEach partition takes up to 10 MB/s

Consumer group of 6Each consumer processes 1,500 events/s

DatabaseHandles up to 20,000 writes/s