How should Amazon EventBridge Pipes documentation be translated when an event is filtered out, an enrichment succeeds but its target fails, or only part of a batch needs retry? The correct language matters because developers can easily mistake a source record for a delivered event, a filter match for an authorized action, or a retried batch for a uniquely processed business transaction.
This comprehensive, world-facing technical localization guide explains Amazon EventBridge Pipes sources, filter patterns, enrichment, input transformations, target calls, IAM execution roles, stream starting positions, batching, partial-batch failure reports, retries, dead-letter queues, source ordering, concurrency and monitoring. It is for professional translators, international cloud teams, technical writers, educators and developers supporting multilingual event-driven systems.
Imagine an international learning platform copying certificate-update events from a source stream to a verification service. A pipe reads an event, applies a filter, calls a synchronous enrichment to find the learner’s course category, then invokes a target. If the target rejects one event, the pipe may retry processing in a manner determined by the source and target configuration. The reader needs to know precisely which stage succeeded. An event observed in a source is not automatically an artifact stored in the target.
Quick answer: source read, filter match and target success are different
Amazon EventBridge Pipes connects one source to one target and can apply optional filtering, transformation and enrichment between them. It does not mean every source record reaches the target: filters can exclude records, enrichment can change payloads and targets can fail. Batch size and partial-failure handling depend on compatible sources and targets. Source ordering guarantees are source-dependent and must not be generalized. Distinguish the pipe running, a record matched, enrichment responded, target accepted, and a downstream business result verified.
A fictional cross-border certificate publisher
A fictional educational publisher streams course-completion events into an EventBridge Pipe. Its filter accepts only records marked ready-for-review, then a Lambda enrichment looks up the educational category and a downstream target indexes approved records. An event with insufficient fields is filtered out; a second event reaches enrichment; a third returns a target-side partial failure. The public interface must not say that every source event was published to the index. Support needs source record identifiers, filter criteria, enrichment responses, target outcomes and a method of identifying repeat attempts.
Amazon EventBridge Pipes: precise terminology for translation
1. EventBridge Pipes
A pipe provides a managed route from a single configured source to a single target, with optional filtering and enrichment.
Translation risk: Do not translate pipe as a broadcast bus that automatically delivers every event to many targets.
Evidence to check: Confirm source, target and configured stages.
2. Source
The source supplies the records that a pipe reads, polls or receives according to its source type.
Translation risk: Do not call the source event a target delivery acknowledgment.
Evidence to check: Inspect the source type and source-specific access settings.
3. Target
The target is the configured destination invoked with the event or transformed payload.
Translation risk: Do not infer that an accepted target invocation guarantees a final user-facing business effect.
Evidence to check: Read target invocation result and downstream records.
4. Single source, single target
One pipe has a single source and a single target even when multiple pipes or an event bus provide broader routing.
Translation risk: Do not translate a single pipe as native fan-out to multiple independent destinations.
Evidence to check: Trace any separate pipes or bus rules explicitly.
5. Filter
A filter pattern selects which source records continue toward enrichment and target.
Translation risk: Do not describe records excluded by filtering as successfully delivered to the target.
Evidence to check: Compare patterns with real sample events.
6. Filter pattern
An event pattern defines field-level matching constraints under documented EventBridge rules.
Translation risk: Do not translate or modify JSON keys and matching operators inside runnable filters.
Evidence to check: Test records that should match and fail to match.
7. Filtered-out record
A record not matching the filter is excluded from downstream processing by that pipe.
Translation risk: Do not describe filtering as an automatic business rollback or archival action.
Evidence to check: Check source-specific treatment of excluded records.
8. Input transformation
An input transformer can construct a target-shaped payload from selected event fields.
Translation risk: Do not confuse transformed content with proof that source data was correct.
Evidence to check: Compare original record and outgoing payload.
9. Enrichment
An optional enrichment invokes a service to add or derive information before the target call.
Translation risk: Do not translate enrichment successful as target processing completed.
Evidence to check: Check enrichment response and later target result.
10. Synchronous enrichment
EventBridge Pipes waits for a configured enrichment’s response before proceeding to the target.
Translation risk: Do not promise background asynchronous enrichment by default.
Evidence to check: Read supported enrichment service and invocation type.
11. Lambda enrichment
An AWS Lambda function can enrich data and return a transformed response to continue through the pipe.
Translation risk: Do not claim Lambda’s success proves every individual target item succeeded.
Evidence to check: Inspect returned payload and target response.
12. API Gateway enrichment
An API Gateway endpoint can serve as a configured enrichment endpoint under the supported interface.
Translation risk: Do not hide an upstream 4xx or 5xx as a successful enrichment.
Evidence to check: Check invocation status and mapped response.
13. API destination enrichment
An API destination can provide a managed external-API enrichment path with configured connection and permissions.
Translation risk: Do not claim the endpoint can access any external service without authentication or quotas.
Evidence to check: Check destination connection and response handling.
14. Step Functions enrichment
An Express Step Functions state machine can be used as synchronous enrichment under documented constraints.
Translation risk: Do not translate any Step Functions workflow type as interchangeable for enrichment.
Evidence to check: Verify Express compatibility and response shape.
15. Enrichment response
The enrichment’s returned data is forwarded toward the target, potentially after transformation.
Translation risk: Do not assume the number of returned records always equals the number originally read.
Evidence to check: Compare input batch and returned payload shape.
16. Enrichment as filter
An enrichment can suppress target invocation by returning a suitable empty response under documented behavior.
Translation risk: Do not say a suppressed invocation was delivered as an empty business record.
Evidence to check: Distinguish empty result from a purposely sent payload.
17. Empty payload
An empty enrichment response can have special semantics different from a list containing an empty object.
Translation risk: Do not translate all empty representations as equivalent JSON syntax.
Evidence to check: Test the exact payload shape described by AWS.
18. Batch
A batch groups multiple events for supported source, enrichment and target operations.
Translation risk: Do not promise batching is supported by every combination of endpoints.
Evidence to check: Check compatible batch APIs and configured limits.
19. Batch size
The maximum usable batch size depends on source limits and target or enrichment support.
Translation risk: Do not translate a source maximum as a universal end-to-end maximum.
Evidence to check: Check the smallest relevant supported limit.
20. Batch window
Some source integrations allow time-based collection of records before invocation.
Translation risk: Do not translate window elapsed as proof a target has completed processing.
Evidence to check: Read batch window and invocation timing separately.
21. Batchable target
A compatible target API can accept multiple events in one request under its own limits.
Translation risk: Do not assume single-event targets accept arrays automatically.
Evidence to check: Check the target API payload contract.
22. Batch payload array
Batch integrations can pass arrays of records to supported enrichments or targets, including special single-record array behavior.
Translation risk: Do not silently change the array into an object in localized code samples.
Evidence to check: Validate payload schema with the actual integration.
23. Input transformer per record
For a source batch, certain transformations operate on each JSON record rather than a single combined array.
Translation risk: Do not describe per-record mapping as an aggregate transformation without evidence.
Evidence to check: Compare individual items before and after transformation.
24. Partial batch failure
For supported sources and targets, a partial failure report identifies records that should be retried.
Translation risk: Do not translate one failed record as every record lost or every record succeeded.
Evidence to check: Inspect supported response schema and failed IDs.
25. batchItemFailures
The target response can identify failed items through a documented batchItemFailures structure.
Translation risk: Do not translate or misspell the JSON key in runnable examples.
Evidence to check: Validate response object field names.
26. itemIdentifier
Each partial failure entry identifies an original source event by its supported identifier.
Translation risk: Do not use an arbitrary business label when the API expects a source-specific ID.
Evidence to check: Verify messageId or eventID mapping as appropriate.
27. Incorrect failure ID
A wrong identifier can prevent correct partial-failure processing and may cause whole-batch behavior under documented semantics.
Translation risk: Do not treat a malformed report as an accurate per-record result.
Evidence to check: Test invalid and missing identifiers safely.
28. Target batch retry
After a partial target failure, supported pipes may retry remaining failed records through the pipe.
Translation risk: Do not guarantee enrichment runs only once when records are retried.
Evidence to check: Trace retry attempts and enrichment calls.
29. Enrichment partial failure
Pipes does not support the same partial-batch failure mechanism for enrichment as for supported targets.
Translation risk: Do not copy target partial-failure rules into enrichment guidance.
Evidence to check: Read the specific supported exception and retry behavior.
30. Retry
A retry attempts processing again after a qualifying error according to the configured source and invocation policy.
Translation risk: Do not promise a retry is exactly-once business execution.
Evidence to check: Use stable deduplication keys in external writes.
31. At-least-once processing
A source or integration path may permit repeated delivery attempts, requiring idempotent consumers.
Translation risk: Do not translate an event ID as proof a handler was invoked only once.
Evidence to check: Compare target records across attempts.
32. Source retention
Some source records remain only for a configured retention window, affecting recovery choices.
Translation risk: Do not promise infinite replay from a stream.
Evidence to check: Check source retention and checkpoint position.
33. Stream starting position
Stream sources may allow LATEST, TRIM_HORIZON or time-based start semantics where supported.
Translation risk: Do not translate LATEST as all historical events.
Evidence to check: Inspect selected starting position and source creation time.
34. LATEST
LATEST generally begins reading new stream records from a contemporary position and may miss earlier records during setup.
Translation risk: Do not translate LATEST as earliest retained event.
Evidence to check: Check stream-specific AWS documentation.
35. TRIM_HORIZON
TRIM_HORIZON directs applicable stream readers toward the oldest retained records currently available.
Translation risk: Do not promise access to records already expired from retention.
Evidence to check: Verify the actual stream retention window.
36. AT_TIMESTAMP
Some stream sources support beginning near a specified timestamp according to their API.
Translation risk: Do not convert timestamps without preserving the time-zone instant.
Evidence to check: Test date formats and stream position.
37. SQS source
An Amazon SQS queue can be a pipe source with visibility-timeout, receive and dead-letter behavior.
Translation risk: Do not treat an SQS source as a Kinesis checkpointed stream.
Evidence to check: Read source-queue redrive settings.
38. SQS visibility timeout
Visibility timeout influences how soon an unacknowledged message may become available again.
Translation risk: Do not promise a failed message immediately reappears or vanishes forever.
Evidence to check: Check source queue’s configured timeout.
39. SQS dead-letter queue
A source SQS queue can use its own redrive policy and dead-letter queue after repeated receives.
Translation risk: Do not confuse this source queue DLQ with the pipe’s separate asynchronous invocation DLQ.
Evidence to check: Inspect where each DLQ is attached.
40. Pipe asynchronous DLQ
A configured pipe dead-letter destination can handle specific asynchronous delivery failures under its documented model.
Translation risk: Do not claim it automatically replaces every source-specific retry or queue DLQ.
Evidence to check: Review pipe and source configurations separately.
41. SQS FIFO message group
Messages with the same FIFO group have ordering rules preserved through the supported pipe source behavior.
Translation risk: Do not turn a per-group ordering statement into a global order across groups.
Evidence to check: Inspect messageGroupId and concurrent groups.
42. Kinesis stream source
Kinesis provides records organized in shards with source-specific ordering and checkpoint behavior.
Translation risk: Do not promise global stream order or infinite event retention.
Evidence to check: Read shard and starting-position settings.
43. DynamoDB Streams source
A DynamoDB stream provides table change records read with source-specific ordering and stream lifecycle considerations.
Translation risk: Do not equate table write success with target indexing success.
Evidence to check: Trace event IDs and target outputs.
44. Kafka source
Amazon MSK or self-managed Kafka sources have topic, partition and consumer-processing semantics that shape ordering.
Translation risk: Do not assume behavior matches SQS or DynamoDB Streams automatically.
Evidence to check: Check topic, partition, consumer and authentication settings.
45. Amazon MQ source
An Amazon MQ broker can provide messages through a pipe according to broker and source-specific constraints.
Translation risk: Do not equate broker acknowledgment with target business completion.
Evidence to check: Review source configuration and retry semantics.
46. Source ordering
Ordering preservation depends on what the particular source guarantees and how records move through the pipe.
Translation risk: Do not claim a global total order for unrelated sources and partitions.
Evidence to check: Name source ordering scope in the translation.
47. Pipe execution
Each record or batch moving to enrichment or target can constitute a pipe execution under the platform’s terminology.
Translation risk: Do not equate execution counted with unique customer transaction completed.
Evidence to check: Correlate source identifiers and target evidence.
48. STARTED state
A started pipe is active and can poll or process events according to its configuration.
Translation risk: Do not translate STARTED as all backlog fully drained.
Evidence to check: Read source backlog and delivery metrics.
49. STOPPED state
A stopped pipe does not actively perform normal event processing while stopped.
Translation risk: Do not translate stopping the pipe as deleting the source records or target artifacts.
Evidence to check: Check source retention and resume behavior.
50. Source update restriction
Some source definitions cannot be swapped in place after pipe creation, and source resources can have lifecycle constraints.
Translation risk: Do not promise an effortless source ARN replacement on an existing pipe.
Evidence to check: Verify allowed update operations for that source.
51. IAM execution role
A pipe uses configured permissions to read source data and invoke enrichment or target operations.
Translation risk: Do not translate permission denied as a filter mismatch or successful processing.
Evidence to check: Inspect role trust and permissions.
52. Credential and secret
Integrations may depend on connection credentials or stored secrets for protected services.
Translation risk: Do not copy live credential values into public training material.
Evidence to check: Use synthetic placeholders and verify secret-access permissions.
53. Encryption and privacy
Source payloads can carry personal or confidential information through several systems and logs.
Translation risk: Do not expose private learner records in localized troubleshooting examples.
Evidence to check: Redact samples and review each stage’s data policy.
54. Metrics
Monitoring can show invocations, failures and traffic but may not represent final customer receipt.
Translation risk: Do not translate a rising invocation count into number of completed customer orders.
Evidence to check: Correlate metrics with business records.
55. CloudWatch logs
Configured logs can help explain pipe behavior and failure points, subject to logging level and permissions.
Translation risk: Do not treat a missing log line as proof no event was processed.
Evidence to check: Read logging setup and trace identifiers.
56. Input schema evolution
Changing source fields can cause filters, enrichments and targets to behave differently.
Translation risk: Do not translate an old field name into a replacement without updating executable patterns.
Evidence to check: Test historical and new schema versions.
57. Poison event
An event that repeatedly fails parsing or processing can block progress within particular source semantics.
Translation risk: Do not call every poison event a corrupted source stream.
Evidence to check: Identify the failing record and recovery path.
58. Idempotent target
A target uses stable identifiers to avoid duplicate external effects when the same source event is attempted again.
Translation risk: Do not promise the pipe itself guarantees an exactly-once payment or certificate write.
Evidence to check: Verify target uniqueness constraints and event keys.
59. Business completion
The intended business effect exists only when the target-side system confirms it according to its own rules.
Translation risk: Do not infer this from a filter match, enrichment response or pipe state alone.
Evidence to check: Check the authoritative customer-facing record.
Twelve worked failure cases and their safe translations
1. An event fails the filter
A course-update record does not match the ready-for-review event pattern. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translated dashboard says every source event was delivered to the target. A technically defensible explanation is: Explain that the record was excluded by the pipe’s filter and did not proceed to that target. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Check filter rule, source-specific removal and matching fields. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
2. The enrichment returns an empty response
An enrichment intentionally suppresses target invocation for a nonqualifying record. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The localized status says an empty document was stored at the destination. A technically defensible explanation is: Explain that the target was not invoked because the enrichment result suppressed the event. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Inspect actual response shape. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
3. A target accepts only a smaller batch
The source could supply many records, but the target supports a smaller batch maximum. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translated guide uses the source maximum for the whole pipe. A technically defensible explanation is: Explain that the compatible batch size is limited by the entire route. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Check target batch API and configured pipe size. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
4. One batch item fails at the target
A supported target returns a valid partial-failure report naming one original event. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The localized report says all batch items failed or none requires retry. A technically defensible explanation is: Explain that the named item is eligible for retry while successful items should not be conflated with it. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Validate batchItemFailures and source IDs. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
5. An enrichment is invoked again during retry
A target-side partial failure sends a record through the pipe again and enrichment may run again. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translated incident says enrichment always happens only once per record. A technically defensible explanation is: State that retries can repeat enrichment and external side effects must be designed accordingly. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Count enrichment calls for the source event. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
6. An itemIdentifier is wrong
A target returns a business order number instead of the required original event identifier. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The localized tutorial calls the response a valid partial failure report. A technically defensible explanation is: Explain the schema error and possible whole-batch failure handling. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Check source-specific messageId or eventID. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
7. An SQS message is retried
The target fails, visibility timeout expires and the source queue makes the message available again. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translation says a new independent customer order was submitted. A technically defensible explanation is: Explain redelivery of the same logical message and use a stable deduplication key. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Compare message ID and business key. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
8. A stream starts at LATEST
A new pipe starts reading a stream and earlier retained records are not necessarily replayed. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The localized help claims the pipe backfilled all history automatically. A technically defensible explanation is: Explain the chosen starting position and the risk of missing records around creation or updates. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Inspect source offset or sequence and setup time. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
9. An SQS DLQ is mistaken for pipe DLQ
A failing message needs source queue redrive, but the operator configures only a pipe asynchronous DLQ. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translated runbook promises all poison source messages are archived there. A technically defensible explanation is: Explain that the source queue DLQ and pipe DLQ serve different failure paths. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Review where each redrive policy is attached. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
10. FIFO group order is generalized
One SQS FIFO message group is processed sequentially while other groups proceed independently. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translated diagram promises a single total order across the entire queue. A technically defensible explanation is: Explain ordering scope at the message-group level. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Check messageGroupId and source-specific concurrency. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
11. A pipe role lacks permission
The execution role cannot invoke the enrichment Lambda although the source is readable. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translation blames a filter mismatch and says there were no events. A technically defensible explanation is: Explain authorization failure in the enrichment invocation stage. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Inspect IAM policies and error logs. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
12. A target returns success but business state is pending
A destination accepts an invocation but another service must finish producing a certificate. A pipe can move data between systems, but the boundary between source, enrichment and destination is not a single atomic business transaction. Readers need the correct stage of evidence before taking another action.
An inaccurate localization would say: The translated portal says the reader can already download the final file. A technically defensible explanation is: Explain that delivery to the target and final artifact availability are separate. That distinction becomes especially important during retries, when repeating a source event may repeat an enrichment or external write.
Verification should include: Verify the final document and stable customer record. If the records cannot be correlated, report the result as not yet verified rather than assuming success or failure.
A twelve-step technical localization workflow
Step 1: Map source through target
Draw the one-source, optional filter, optional enrichment, transformation and one-target route. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Write status words for each boundary separately. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 2: Check source semantics
Identify SQS, Kinesis, DynamoDB Streams, Kafka or MQ and its retry, retention and ordering model. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Do not import guarantees from a different source. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 3: Protect event schema
Keep JSON keys, patterns, identifiers and payload shapes exactly valid in runnable examples. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Translate explanations and labels, not the machine contract. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 4: Test positive and negative filters
Use sample events that should pass and fail; note source-specific consequences of filtering. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Readers must know whether excluded records are discarded, retained or retried. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 5: Validate enrichment output
Test empty, single-record and batch-shaped responses as supported by the configured path. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
A returned array can change target behavior. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 6: Determine batch limits
Compare source settings and target capabilities before publishing batch examples. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
The tightest compatible limit governs end-to-end batching. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 7: Test partial failure response
Return correctly shaped batchItemFailures with original source identifiers in a safe sandbox. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Incorrect keys can change retry scope. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 8: Inspect repeated attempts
Exercise target timeouts and retry flows to see when enrichment and target operations run again. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Protect non-idempotent writes with stable keys. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 9: Separate dead-letter paths
Document source queue redrive and pipe asynchronous failure handling according to the actual source. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Do not use dead-letter queue as one universal term without context. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 10: Audit permissions and secrets
Check role trust, source access, enrichment permission, target permission and credential handling. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
A missing permission is not an event-filter result. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 11: Trace delivery evidence
Correlate source ID, pipe execution, enrichment response, target invocation and final business record. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
A target response may still be provisional. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Step 12: Complete multilingual review
Have a technical reviewer replay each example and a language editor inspect all promise verbs. Work from a real configuration sample or a synthetic equivalent rather than a diagram stripped of source-specific details. The correct translation depends on each stage’s actual API contract.
Preserve conditional meaning and unknown outcomes across languages. Include at least one negative example in the target language so readers see what the service does not guarantee. A reliable support article helps users avoid duplicated actions and missing data.
Reader practice: repair overconfident claims
Practice 1: An event fails the filter
Read the incident: A course-update record does not match the ready-for-review event pattern. A draft help article claims: The translated dashboard says every source event was delivered to the target. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain that the record was excluded by the pipe’s filter and did not proceed to that target. Substantiate the wording by checking: Check filter rule, source-specific removal and matching fields. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 2: The enrichment returns an empty response
Read the incident: An enrichment intentionally suppresses target invocation for a nonqualifying record. A draft help article claims: The localized status says an empty document was stored at the destination. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain that the target was not invoked because the enrichment result suppressed the event. Substantiate the wording by checking: Inspect actual response shape. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 3: A target accepts only a smaller batch
Read the incident: The source could supply many records, but the target supports a smaller batch maximum. A draft help article claims: The translated guide uses the source maximum for the whole pipe. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain that the compatible batch size is limited by the entire route. Substantiate the wording by checking: Check target batch API and configured pipe size. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 4: One batch item fails at the target
Read the incident: A supported target returns a valid partial-failure report naming one original event. A draft help article claims: The localized report says all batch items failed or none requires retry. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain that the named item is eligible for retry while successful items should not be conflated with it. Substantiate the wording by checking: Validate batchItemFailures and source IDs. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 5: An enrichment is invoked again during retry
Read the incident: A target-side partial failure sends a record through the pipe again and enrichment may run again. A draft help article claims: The translated incident says enrichment always happens only once per record. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: State that retries can repeat enrichment and external side effects must be designed accordingly. Substantiate the wording by checking: Count enrichment calls for the source event. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 6: An itemIdentifier is wrong
Read the incident: A target returns a business order number instead of the required original event identifier. A draft help article claims: The localized tutorial calls the response a valid partial failure report. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain the schema error and possible whole-batch failure handling. Substantiate the wording by checking: Check source-specific messageId or eventID. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 7: An SQS message is retried
Read the incident: The target fails, visibility timeout expires and the source queue makes the message available again. A draft help article claims: The translation says a new independent customer order was submitted. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain redelivery of the same logical message and use a stable deduplication key. Substantiate the wording by checking: Compare message ID and business key. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 8: A stream starts at LATEST
Read the incident: A new pipe starts reading a stream and earlier retained records are not necessarily replayed. A draft help article claims: The localized help claims the pipe backfilled all history automatically. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain the chosen starting position and the risk of missing records around creation or updates. Substantiate the wording by checking: Inspect source offset or sequence and setup time. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 9: An SQS DLQ is mistaken for pipe DLQ
Read the incident: A failing message needs source queue redrive, but the operator configures only a pipe asynchronous DLQ. A draft help article claims: The translated runbook promises all poison source messages are archived there. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain that the source queue DLQ and pipe DLQ serve different failure paths. Substantiate the wording by checking: Review where each redrive policy is attached. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 10: FIFO group order is generalized
Read the incident: One SQS FIFO message group is processed sequentially while other groups proceed independently. A draft help article claims: The translated diagram promises a single total order across the entire queue. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain ordering scope at the message-group level. Substantiate the wording by checking: Check messageGroupId and source-specific concurrency. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 11: A pipe role lacks permission
Read the incident: The execution role cannot invoke the enrichment Lambda although the source is readable. A draft help article claims: The translation blames a filter mismatch and says there were no events. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain authorization failure in the enrichment invocation stage. Substantiate the wording by checking: Inspect IAM policies and error logs. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Practice 12: A target returns success but business state is pending
Read the incident: A destination accepts an invocation but another service must finish producing a certificate. A draft help article claims: The translated portal says the reader can already download the final file. Decide whether that sentence is about an observed source event, filter result, enrichment call, destination result or downstream business outcome.
A better version says: Explain that delivery to the target and final artifact availability are separate. Substantiate the wording by checking: Verify the final document and stable customer record. Protect identifiers and JSON syntax when translating the example. If the destination effect has not been confirmed, leave that uncertainty explicit.
Frequently asked questions
Does EventBridge Pipes send every source record to the target?
No. Filter patterns can exclude records, and processing failures can prevent delivery. Source-specific handling determines what happens to filtered or failed records. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Can one pipe connect many unrelated sources to many targets?
A pipe is configured for one source and one target. Multiple pipes or an event bus can support broader routing architectures. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Does enrichment mean the target has already succeeded?
No. Enrichment happens before the target invocation. Its success proves only that the enrichment stage responded according to its contract. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Can enrichment be used to suppress events?
Yes. Supported empty responses can prevent invoking the target. Preserve the difference between no target invocation and intentionally delivering an empty JSON object. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Does batch size mean the source maximum alone?
No. Compatible enrichment and target capabilities can constrain the configured batch size. Check the entire route. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
What is a partial batch failure response?
A supported target reports specific failed items using the required structure and original source identifiers, allowing appropriate retry handling. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Can retry invoke enrichment again?
Yes, for applicable target-side partial-failure retry flows, records may pass through enrichment again. Make enrichment and target side effects safe for repetition. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Is ordering guaranteed globally?
No general total order should be inferred. Any preservation of order depends on the source and its ordering scope, such as a shard or FIFO message group. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Are source SQS and pipe DLQs equivalent?
No. A source queue’s DLQ handles its redrive path; the pipe’s asynchronous invocation DLQ addresses a different path. Read the relevant AWS guidance. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Does LATEST read all historical stream records?
No. It is a starting-position choice oriented toward contemporary data. For backfills, consider the source’s supported oldest or timestamp positioning and retention. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Does STARTED mean the target has caught up?
Not necessarily. The pipe may be polling with a backlog or encountering failures. Check source lag and target results separately. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
What belongs in a translated incident record?
Include source type, event ID, filter outcome, enrichment response, target status, retry attempt, DLQ configuration and verified final business effect. Translate the stated scope, such as source type, batch support or conditional retry behavior, without upgrading it into a universal promise.
Official documentation and further reading
These AWS references document the distinctions used here. Available sources, targets, quotas and error handling can change. Confirm the particular deployed pipe and the current AWS reference before issuing production instructions.
Amazon EventBridge Pipes concepts
EventBridge Pipes batching and concurrency
EventBridge Pipes with SQS sources
EventBridge Pipes with DynamoDB Streams
Related: Cloudflare Queues localization and redelivery
Conclusion: report the exact stage, not a promise of delivery
Amazon EventBridge Pipes makes event connections easier to implement, but a pipe does not erase the meanings of source order, filtering, enrichment, target invocation and final business completion. The strongest translation keeps event IDs and runnable syntax intact, describes supported partial failures precisely, distinguishes dead-letter paths and treats retries as possible repeated attempts rather than automatic exactly-once success. Readers can then make sensible decisions about monitoring, recovery and verification.