AWS Lambda Supports Parallelization Factor for Kinesis and DynamoDB Event Sources
0 likes3,563 views
The document provides an overview of AWS Lambda and its integration with streaming event sources, specifically focusing on Amazon Kinesis and DynamoDB Streams. It covers topics such as scaling, monitoring, error handling, and performance optimization in a chalk-talk format aimed at enhancing serverless applications. Key insights include strategies for managing lambda functions, issues related to iterator age, throughput limits, and the use of enhanced fan-out for improved data delivery.
AWS Lambda Supports Parallelization Factor for Kinesis and DynamoDB Event Sources
- 2. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Mastering AWS Lambda streaming event sources Adam Wagner S V S 3 2 3 - R Solutions Architect Amazon Web Services
- 3. Related breakouts SVS317-R – Serverless stream processing pipeline best practices SVS401-R – Optimizing your serverless applications SVS335-R – Serverless at scale: Design patterns and optimizations API304 – Scalable serverless event-driven applications using Amazon SQS & Lambda
- 4. Agenda Introduction to streaming event sources for AWS Lambda Scaling Monitoring and error handling Common issues Performance and optimization
- 5. Session expectations • Chalk-talk format – Please ask questions • What we will cover • The details of using Lambda with streaming event sources • Scaling • Monitoring • Error handling • Performance and optimization • What we won’t cover • What is serverless? • What is Lambda? • Event sources outside of Amazon Kinesis Data Streams and Amazon DynamoDB Streams
- 6. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 7. Amazon Kinesis Easily collect, process, and analyze video and data streams in real time Capture, process, and store video streams for analytics Load data streams into AWS data stores Analyze data streams with SQL Build custom applications that analyze data streams Amazon Kinesis Video Streams Amazon Kinesis Data Streams Amazon Kinesis Data Firehose Amazon Kinesis Data Analytics
- 8. Amazon DynamoDB Amazon DynamoDB Document or key-value Scales to any workload Fully managed NoSQL Access control Event-driven programming Fast and consistent
- 9. DynamoDB Streams DynamoDB DynamoDB stream ✓ Stream of item changes ✓ Exactly once, guaranteed delivery ✓ Strictly ordered by key ✓ Durable, scalable ✓ Fully managed ✓ 24-hour data retention ✓ Sub-second latency ✓ Event source for Lambda DynamoDB Streams
- 10. What we’re talking about today Kinesis Data Streams Lambda Data Produce r Data Produce r Data producer Downstream system Clients Produce r Clients Produce r Clients DynamoDB DynamoDB stream Lambda Downstream system
- 11. Kinesis Data Streams Kinesis Data Streams Lambda service Data ProducerData ProducerData producer Lambda function A Lambda function B
- 12. DynamoDB Streams Clients Produce r Clients Produce r Clients DynamoDB DynamoDB stream Lambda service Lambda function A Lambda function B
- 13. Kinesis data stream shard detail Data ProducerData ProducerData producer FunctionKinesis data stream Shard Shard Shard Shard Lambda service
- 14. Kinesis data stream Data ProducerData ProducerData producer FunctionKinesis data stream Shard Shard Shard Shard
- 15. Kinesis data stream FunctionKinesis data stream Shard Shard Shard Shard
- 16. Kinesis data stream shard-level detail Shard 1. Lambda service polls the shard once per second for a set of records. Then synchronously invokes the Lambda function with the batch of records. 2. If the Lambda returns successfully, the Lambda service advances to the next set of records and repeats step 1. 3. If the Lambda errors, by default the Lambda service invokes the function with the same set of records and will continue to do so until it succeeds or the records age out of the stream.
- 17. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 18. Kinesis data stream scaling FunctionKinesis data stream Shard Shard Shard Shard
- 19. Kinesis data stream scaling aws kinesis update-shard-count --stream-name reinvent19-01 --target-shard-count 8 --scaling-type UNIFORM_SCALING { "StreamName": "reinvent19-01", "CurrentShardCount": 4, "TargetShardCount": 8 } FunctionKinesis data stream Shard Shard Shard Shard
- 20. Kinesis data stream scaling FunctionKinesis data stream Shard Shard Shard Shard Shard Shard Shard Shard aws kinesis update-shard-count --stream-name reinvent19-01 --target-shard-count 8 --scaling-type UNIFORM_SCALING { "StreamName": "reinvent19-01", "CurrentShardCount": 4, "TargetShardCount": 8 }
- 21. Kinesis data stream scaling FunctionKinesis data stream Shard Shard Shard Shard • Scale more than twice per rolling 24-hour period per stream • Scale up to more than double your current shard count for a stream • Scale down below half your current shard count for a stream • Scale up to more than 500 shards in a stream • Scale a stream with more than 500 shards down unless the result is less than 500 shards • Scale up to more than the shard limit for your account
- 22. Kinesis data stream scaling … more detail Function Shard Shard Kinesis data stream Shard Shard Shard Shard • The stream scales up by splitting shards • Splitting a shard creates two new child shards that split the partition keyspace of the parent shard • Lambda will not start receiving records from the child shards until it’s processed all records from the parent shard
- 24. Parallelization Factor FunctionKinesis Stream Shard Shard Shard Shard • Adds Lambda parallelization per shard • Setting of 1 is the same as the current behavior, maximum setting is 10 • Batching via partition keys to maintain in order processing per partition key • Works with both Kinesis Data Streams and DynamoDB Streams --parallelization-factor 1
- 25. Parallelization Factor Kinesis Stream Shard Shard Shard Shard Function --parallelization-factor 2 • Adds Lambda parallelization per shard • Setting of 1 is the same as the current behavior, maximum setting is 10 • Batching via partition keys to maintain in order processing per partition key • Works with both Kinesis Data Streams and DynamoDB Streams
- 27. Kinesis data stream scaling FunctionKinesis data stream Shard Shard Shard Shard • Auto-scale your shard count using Application Auto Scaling: https://aws.amazon.com/blogs/big- data/scaling-amazon-kinesis-data-streams- with-aws-application-auto-scaling/ • Scale conservatively to leave overhead for bursts of traffic • Scale your shard count to match your Lambda throughput and/or use Parallelization Factor • Test! Test! Test! Measure unit tests to watch for performance regressions, and also test at scale!
- 28. DynamoDB Streams scaling FunctionDynamoDB stream Shard Shard Shard Shard DynamoDB
- 29. DynamoDB on-demand vs. provisioned capacity
- 30. DynamoDB on-demand scaling DynamoDB tables using on-demand capacity mode automatically adapt to your application’s traffic volume. On-demand capacity mode instantly accommodates up to double the previous peak traffic on a table. For example, if your application’s traffic pattern varies between 25,000 and 50,000 strongly consistent reads per second where 50,000 reads per second is the previous traffic peak, on-demand capacity mode instantly accommodates sustained traffic of up to 100,000 reads per second. If your application sustains traffic of 100,000 reads per second, that peak becomes your new previous peak, enabling subsequent traffic to reach up to 200,000 reads per second.
- 32. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 33. Kinesis data stream monitoring Kinesis Data Streams Lambda Data ProducerData ProducerData producer Downstream system
- 34. DynamoDB stream monitoring Clients Produce r Clients Produce r Clients DynamoDB DynamoDB stream Lambda Downstream system
- 35. Error handling options A number of new options are available to tune error handling • Maximum retry attempts – min 0, default/max 10,000 • Maximum Record Age in seconds – min 60, default/max 604,800 • Bisect Batch on Function Failure • On-Failure Destination
- 36. Bisect Batch on Function Failure Recursively split the failed batch and retry on a smaller subset of records, eventually isolating the problematic records • Boolean – false by default • These retries do NOT count towards MaximumRetryAttempts • Make sure your function is idempotent
- 37. On-Failure Destination An SNS Topic or SQS Queue, which is sent the metadata about a failed batch of records • Used only after configured retry limit or maximum record age are reached. • Remember the bisected batch retries are not counted towards retry limit. • Does not contain the actual records, but does contain all the information needed to retrieve them!!
- 38. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 39. Common issues • Kinesis Data Streams • IteratorAge is growing rapidly • ReadProvisionedThroughputExceeded throttles • DynamoDB Streams • IteratorAge is growing rapidly • Rapid growth in Lambda concurrency
- 40. Kinesis Data Streams IteratorAge is growing rapidly • Initial questions • How many Lambda functions subscribed to the stream? • Does the Lambda function show any errors? • Does the Lambda function show any throttles? • Is there a large increase in Kinesis Data Streams metrics IncomingRecords or IncomingBytes?
- 41. Kinesis Data Streams IteratorAge is growing rapidly Kinesis Data Streams Lambda Data ProducerData ProducerData producer Downstream system
- 42. Kinesis Data Streams IteratorAge is growing rapidly • Solutions • If the Lambda is erroring • Configure an SQS Queue or SNS Topic for failed batches • Configure MaximumRetryAttempts, BisectBatchOnFunctionError, and MaximumRecordAgeInSeconds • Update the Lambda function to log records causing errors and return successfully • If the Lambda is throttling? • Increase per function limit/reservation, or raise the account level limit
- 43. Kinesis Data Streams IteratorAge is growing rapidly • Solutions • If there is a large increase in KDS Metrics IncommingRecords or IncommingBytes • If this is temporary, you may be able to wait it out. Watch IteratorAge to make sure it doesn’t climb too high • Increase the stream data retention, this can be increased up to 7 days • Increase the Parallelization Factor • Increase the number of shards in the stream • Increase the memory assigned to the Lambda function or otherwise optimize the function’s performance
- 44. Kinesis Data Streams ReadProvisionedThroughputExceeded The 5 read/sec. or 2 MiB/sec. limit is being hit • Use enhanced fanout or remove one or more subscribers • Remember that Kinesis Data Firehose and Kinesis Data Analytics are subscribers as well!
- 45. DynamoDB Streams IteratorAge is growing rapidly • Initial questions • How many Lambda functions subscribed to the stream? • Does the Lambda function show any errors or throttles? • Does the Lambda function show an increase in duration? • Is there a large increase in the DynamoDB table write (WCU) metrics • Is there a large increase in the DynamoDB stream metrics
- 46. DynamoDB Stream IteratorAge is growing rapidly • Solutions • If there is a large increase in writes on the DDB Table: • If this is temporary, you may be able to wait it out. Watch IteratorAge to make sure it doesn’t climb too high • Unlike KDS you can NOT increase the data retention time, so you need to take action more quickly • Increase the memory assigned to the Lambda function or otherwise optimize the function’s performance • Increase the Parallelization Factor • If there are more than two Lambda functions subscribed to the stream, consider adding a Kinesis Data Stream for increasing the fan-out
- 47. DynamoDB stream fanout Clients Produce r Clients Produce r Clients DynamoDB DynamoDB stream Lambda Kinesis Data Streams
- 48. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 49. Performance • What matters to your application? • End-to-end latency • Overall cost • Kinesis Data Streams enhanced fanout (EFO) • DynamoDB Streams • Small messages in Kinesis Data Streams • Aggregation/de-aggregation libraries • Compression • Low throughput streams—batch window to the rescue
- 50. Lambda supports Kinesis Data Streams Enhanced Fan-Out and HTTP/2 for faster streaming Enhanced fan-out allows customers to scale the number of functions reading from a stream in parallel while maintaining performance HTTP/2 data retrieval API improves data delivery speed between data producers and Lambda functions by more than 65% Amazon Kinesis Data Streams
- 51. Kinesis Data Streams: Enhanced Fan-Out When to use standard consumers: • Total number of consuming applications is low (< 3) • Consumers are not latency-sensitive • Newer error handling options are needed* • Minimize cost When to use Enhanced Fan-Out consumers: • Multiple consumer applications for the same Kinesis Data Stream • Default limit of 5 registered consuming applications. More can be supported with a service limit increase request • Low-latency requirements for data processing • Messages are typically delivered to a consumer in less than 70 ms
- 52. Optimizing Small Messages in Kinesis • Kinesis Data Streams per shard write limits • 1MiB/sec or 1,000 messages/sec • With high volumes of small messages you reach the 1,000 messages/sec limit easily • This leads to lower throughput per shard and higher costs Aggregation is the answer!
- 53. Aggregation / de-aggregation options • Producer side • Kinesis Producer Library(KPL)(https://github.com/awslabs/amazon-kinesis-producer) • Kinesis Aggregation Library(https://github.com/awslabs/kinesis-aggregation) • Consumer side within Lambda • Kinesis Aggregation Library(https://github.com/awslabs/kinesis-aggregation) • Java, Node.js, and Python versions available • Another option if your data has a consistent format is Avro
- 54. Low-throughput streams Lambda triggered with very small batches Leads to higher cost per message For archiving workloads the resulting payload is too small FunctionKinesis data stream Shard Shard Shard Shard
- 55. Batch window • Additional knob to tune the stream trigger • Set a time to wait before triggering. Max five minutes, set in seconds. • Batch size is still respected and will trigger on full batches before the batch window is up • Works for both Kinesis Data Streams and DynamoDB Streams triggers FunctionKinesis Stream Shard Shard Shard Shard
- 56. Conclusion • Be clear on the goals of your streaming system • Understand how your system scales • Prepare for failures, make use of the new rror handling options • Test individual components as well as end to end
- 57. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Free, on-demand courses on serverless, including Visit the Learning Library at https://aws.training Additional digital and classroom trainings cover modern application development and computing Learn serverless with AWS Training and Certification Resources created by the experts at AWS to help you learn modern application development • Introduction to Serverless Development • Getting into the Serverless Mindset • AWS Lambda Foundations • Amazon API Gateway for Serverless Applications • Amazon DynamoDB for Serverless Architectures
- 58. Thank you! © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 59. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.