1

I've noticed a number of other questions talking about lambda concurrency, but none seem to have been exactly the same as my issue so here's a new one:

I have a current lambda function using python requests that takes an SQS trigger. I have a reserved concurrency limit set, but when testing the code firing 1000 SQS triggers, only 1 or 2 instances of the function run.

The SQS trigger settings are as follows:

Activate trigger: No;
Batch size: 1000;
Batch window: 100;
Concurrent batches per shard: None;
Last processing result: None;
Maximum age of record: None;
Maximum concurrency: 800;
On-failure destination: None;
Report batch item failures: Yes;
Retry attempts: None;
Split batch on error: None;
Starting position: None;
Tumbling window duration: None;

The lambda concurrency settings are as follows:

Function concurrency:
Use reserved concurrency;
Reserved concurrency:
800;

The general configuration of the lambda function is as follows:

Memory:
128MB;
Ephemeral storage:
512MB;
Timeout:
15min0sec;
SnapStartInfo:
None;

The SQS function is set to use a Dead Letter Queue, but nothing is reaching there. As far as I can tell these messages get stuck in flight and then seem to dissapear. Lambda has ran a maximum of 6 concurrent executions when testing my script, but normally it's about 2, with 1 failed.

I send the messages to the SQS using a python script using boto3. I know these reach the SQS properly, because the received message metrics are shown to be about 1000 after running the code.

If anyone has any ideas, please share them. Am at a complete loss.

Ashley Southworth
  • 112
  • 1
  • 7
  • 2
    Did you mean to set batch size to 1000? The default is 10. Is your Lambda function actually looping over all messages in the batch and processing them? – jarmod Apr 10 '23 at 14:35
  • The errors are normally caused by requests timing out, about 9/10 times the vpc connection works properly, but then random times there's a URL time out. Maybe if I add a try except catch there so it doesnt fail...? – Ashley Southworth Apr 10 '23 at 14:36
  • 1
    Your batch size of 1000 means it is only going to call the Lambda function once per every 1000 messages. So it sounds like it is working exactly like you have it configured to work. A batch size of 1000 is extremely high, and is most likely not what you intended. Batch size is usually set to a value between 1 and 20. – Mark B Apr 10 '23 at 14:38
  • 1
    Your batch window of 100 seconds also means that the Lambda service is allowed to wait for longer (up to 100s) to accumulate messages to send in a batch before invoking your Lambda function. – jarmod Apr 10 '23 at 14:43

1 Answers1

0

Noting that @jarmod noting the batch size issue was indeed the issue! Changed the batch size to one and issue was solved!

Ashley Southworth
  • 112
  • 1
  • 7