Build 1 publisher2 min readPublished September 20, 2026 at 5:30 PM PDT

A DynamoDB outbox pushes the deduplication work into the consumer's transaction

miruky's console walkthrough puts an order and its outbox event in one DynamoDB transaction, then fails the consumer on purpose to show why redelivery is the consumer's problem. The whole exercise is priced under $5 in us-east-1.

The Engineer · Build desk

Photograph accompanying A DynamoDB outbox pushes the deduplication work into the consumer's transaction — Photo: amazon.science

What happened

miruky built the whole path in the AWS Console, with no terminal required, and deliberately failed the consumer after its transaction to see what a retry does.
The relay cannot atomically send to SQS and advance its stream checkpoint, so the consumer has to tolerate repeated delivery of the same event.
Deduplication keys on the application-level event_id, and a payload digest rejects conflicting content under that identity, independently of transport IDs.
The walkthrough prices a deliberately oversized one-day allowance at about $0.58 before free-tier benefits, against a stated target of under $5 in us-east-1.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

constraint Correctness lands on the consumer side. Every business effect has to sit in the same transaction as its inbox marker, or a redelivered event adds to the sales summary twice.
decision Anyone reproducing this has to build the failure path first: the dead-letter queue must exist before the delivery queue that references it.
cost At under $5 the dedup argument can be settled on an engineer's own card before it reaches a design review, and the person paying is the person testing.
capability A digest over the payload gives the consumer a way to catch two different bodies claiming one event_id, which a transport-level message ID cannot see.

A database commit can succeed while the caller believes processing failed. Retrying then has to answer whether the business effect already happened. miruky's command transaction writes the order and its outbox event together in one table, with key prefixes separating business state from outgoing event intent, and no secondary indexes on it. The consumer's transaction writes an inbox marker and updates the sales summary in the same step.

"An outbox preserves event intent; it does not provide exactly-once transport," miruky wrote.

miruky feeds the relay from the table's own change stream, with New image selected, because a key-only record would not carry the inserted outbox body the relay has to forward. Inbox markers and the sales summary live in a second table, which has no stream.

Queue settings are where the retry budget gets fixed. Visibility timeout is 60 seconds against a 10-second consumer, six times the runtime, which the walkthrough attributes to an AWS recommendation. Maximum receives is 5. So the dedup path has to absorb up to four redeliveries of the same event_id before the message goes to the dead-letter queue. Dead-letter retention is 4 days against the delivery queue's 1 day, four times the window for looking at what failed.

The $0.58 estimate covers a deliberately oversized day: 10,000 Lambda invocations at 256 MB and 10 billed seconds, 100,000 SQS request units, 100,000 DynamoDB write units, 50,000 read units, and 0.1 GB of ingested logs. Divided into the under-$5 target, that is roughly eight such days. miruky notes that transactional reads and writes consume additional units, and that the thing to count is billing units, not business events. For the figure to transfer, you need us-east-1 and the Lambda, SQS, DynamoDB and CloudWatch rates as checked on September 5, 2026, and you need the setup to stay free of a NAT gateway, a customer-managed KMS key, provisioned polling, a public endpoint and a continuous producer. miruky says the $5 is not a spending cap and calls the final billed amount unverified.

Three IAM roles carry inline policies scoped to the table, stream, delivery-queue and log-group ARNs, each log resource ending in :* to reach its streams, instead of broad managed policies. Resource names in the post are random strings, so keep a mapping if you substitute your own. Each function logs to its own group, Standard class, 1-day retention, no customer-managed encryption key.

What to watch

AWS rates for Lambda, SQS, DynamoDB and CloudWatch were checked on September 5, 2026; a rate change moves the $0.58 day.
Whether the final bill lands under $5, which miruky says is unverified and not a spending cap.
Whether the same pattern holds when the consumer's effect is an external API call instead of a DynamoDB write that can share the inbox marker's transaction.

Claim ledger

Ranked by verification strength, evidence, and original report placement.

[1]
A database commit can succeed while the caller believes processing failed; retrying then becomes a correctness problem, because the question is whether the business effect already happened.
ReportedSupportedView cited source
[2]
miruky built the setup in the AWS Console and failed the consumer after its transaction; the exercise needs resource-creation and role-passing permissions and no terminal.
ReportedSupportedView cited source
[3]
The command transaction stores the order and its outbox event together.
ReportedSupportedView cited source
[4]
The consumer transaction stores an inbox marker and updates the sales summary together; the queue between producer and consumer can deliver repeatedly.
ReportedSupportedView cited source
[5]
"An outbox preserves event intent; it does not provide exactly-once transport."
ReportedSupportedSource: miruky, writing in the dev.to walkthroughView cited source
[6]
The relay cannot atomically send to SQS and advance its stream checkpoint, so the consumer must tolerate repeated delivery.
ReportedSupportedView cited source
[7]
Deduplication uses the application identifier event_id; a payload digest rejects conflicting content under that identity, independently of transport IDs.
ReportedSupportedView cited source
[8]
The target is below $5 in us-east-1, excluding tax and unrelated usage; the walkthrough states this is not a spending cap and that the final billed amount is unverified.
ReportedSupportedView cited source
[9]
A deliberately oversized one-day allowance is approximately $0.58 before free-tier benefits: 10,000 Lambda invocations at 256 MB and 10 billed seconds, 100,000 SQS request units, 100,000 DynamoDB write units, 50,000 read units, 0.1 GB of ingested logs, and tiny stored data.
ReportedSupportedView cited source
[10]
Transactional reads and writes consume additional units; the walkthrough says to count billing units, not business events.
ReportedSupportedView cited source
[11]
Current rates for Lambda, SQS, DynamoDB and CloudWatch were checked September 5, 2026.
ReportedSupportedView cited source
[12]
The setup needs no NAT gateway, customer-managed KMS key, provisioned polling, public endpoint, or continuous producer.
ReportedSupportedView cited source
[13]
The command table uses PK as partition key and SK as sort key, both String, and needs no secondary indexes; the defaults retained are DynamoDB Standard, On-demand and AWS owned key.
ReportedSupportedView cited source
[14]
The command table stores business state and outgoing event intent, distinguished by different key prefixes.
ReportedSupportedView cited source
[15]
The table's stream is enabled with New image selected, because the relay needs the complete inserted outbox body and key-only records would not contain its input; the stream ARN is retained for the relay policy.
ReportedSupportedView cited source
[16]
The second table uses the same PK and SK string keys, stays on demand, stores inbox markers and the sales summary, and does not need a stream.
ReportedSupportedView cited source
[17]
The dead-letter queue is Standard with a message retention period of 4 days and receive message wait time of 20, encrypted with the Amazon SQS key (SSE-SQS), and is created before the delivery queue that references it.
ReportedSupportedView cited source
[18]
The delivery queue is Standard with visibility timeout 60, retention 1 day and receive wait 20, with the dead-letter queue enabled and Maximum receives set to 5.
ReportedSupportedView cited source
[20]
Three log groups are created in United States (N. Virginia) with Retention setting of 1 day and Standard selected, without a customer-managed encryption key, so each function has its own destination.
ReportedSupportedView cited source
[21]
Three IAM roles are created for the Lambda service and use case, each with a corresponding inline permission policy rather than broad managed policies; placeholders are replaced with the table, current stream, delivery-queue and log-group ARNs, and each log resource ends in :* to address its streams.
ReportedSupportedView cited source
[22]
The walkthrough tells readers to keep a mapping if they substitute the supplied random resource names.
ReportedSupportedView cited source
[d1]
The 60-second visibility timeout is six times the 10-second consumer runtime.
DerivedSupported1 source— sign in to open
[d2]
With Maximum receives set to 5, the consumer's deduplication path must absorb up to four redeliveries of the same event_id before the message is routed to the dead-letter queue.
DerivedSupported1 source— sign in to open
[d3]
The under-$5 target covers roughly eight days of the oversized $0.58 allowance.
DerivedSupported1 source— sign in to open
[d4]
Dead-letter retention is four times the delivery queue's retention.
DerivedSupported1 source— sign in to open
[19]
The 60-second visibility timeout matches AWS's six-times-timeout recommendation for the 10-second consumer used in the walkthrough.
ReportedInsufficientView cited source

Sources

Publishers with included, body-backed reporting in this cluster.

dev.to
1 article · September 20, 2026
Build a DynamoDB Outbox and an Idempotent SQS Consumer in the Console

Ask Clarity

Have a question this piece left open? The answer comes from its reporting and claim ledger — nothing else.

Try

Clarity analysis

How the coverage leans and what the record actually supports. Read the methodology

Perspective Coverage

How much of the coverage speaks to builders, operators and investors. It measures emphasis, not whether the story is right.

The design decision worth copying is where the deduplication lives. miruky writes the inbox marker and the sales-summary update in one DynamoDB transaction and keys the marker on the application's own event_id, with a payload digest that rejects conflicting content under that identity. SQS message IDs never enter the check. He is explicit about why: the relay cannot send to the queue and advance its stream checkpoint in one atomic step, so the consumer has to survive repeats. The stream is enabled with New image, because key-only records would not carry the outbox body the relay forwards. Everything is clicked by hand. That makes the build readable. It also means there is nothing to re-run.

Reality

Evidence, adoption, hype gap, incentives and confidence, each read independently from the current record on the scale shown beneath it. How these are measured.

Evidence58

0100

Reproducible in detail, single-sourced

Every figure and setting traces to one dev.to post, but the post is unusually checkable: named form fields, exact queue values, and three IAM policies quoted as full JSON. An engineer with a console can rebuild it and see whether it behaves as described. Two claims sit outside that reach, the AWS six-times visibility recommendation and the under-$5 total, and miruky says the billed amount is unverified.

AdoptionInsufficient

One personal account

The post describes a single hand-built exercise in the author's own AWS account, and no other engineer in our coverage reports running it. There is nothing to measure use against.

Hype gap−10

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

Entities

build
Conditional PutItem claim on orderId blocks duplicate charges from repeated SQS messages
1 publisher
build
Stage boundaries stop a parse failure from re-running the upload, and a thumbnail failure from re-running the parse
1 publisher
build
A 15-minute IAM token replaces the password Lambda keeps in its environment variables
1 publisher
build
One SQS consumer can empty a default Lambda concurrency pool in under four minutes
1 publisher

What happened

Why it matters

What to watch

Claim ledger

Sources

Let Clarity write the post for you.

Topics and entities

Topics

Entities

Related stories

Claim ledger

Let Clarity write the post for you.

Topics and entities

Topics

Entities

Perspective Coverage

Reality

Related stories

Sources