0

How can we get the minimum and maximum lambda execution time by querying cloudwatch? Note we only want to query for the successful execution and skip the failed ones.

This is my first attempt:

fields @message
| filter @type = "REPORT"
| stats min(@duration) as MinimumTime,
        max(@duration) as MaximumTime

Response:

response

It does generates the Min and Max execution time, but it includes the failed lambda executions.

Partially working solution: as the failed lambda executions has log both for exception and report; and successful execution only has report, the following query will remove the failed executions and list only the requestId of successful executions:

fields @message
| filter @message like "REPORT" or @message like "[ERROR]"
| filter ispresent(@requestId)
| stats count(*) as requestIdLogCount by @requestId
| filter requestIdLogCount == 1

Response:

response

Now is it possible to update the above query with some sort of subquery or something else where I can use the @requestId to filter out the log and generate report only with the successful lambda execution? Here is another query which is using the @requestId from above query

fields @message
| filter @type = "REPORT"
| filter @requestId in ["a458412a-95e7-5023-842e-1dbe8f58a876", "082db28d-8f43-5f83-9a72-ee1985861515"]
| stats min(@duration) as MinimumTime,
        max(@duration) as MaximumTime
marc_s
  • 732,580
  • 175
  • 1,330
  • 1,459
Sazzadur Rahman
  • 2,650
  • 1
  • 18
  • 34
  • Also be aware of [Lambda function metrics](https://docs.aws.amazon.com/lambda/latest/dg/monitoring-metrics.html) and the min/max statistics. – jarmod Mar 21 '23 at 22:08
  • Thanks, but still trying to figure out the log query. Actually need to create cloudwatch dashboard with this matrics. – Sazzadur Rahman Mar 21 '23 at 22:17

1 Answers1

1

The better way would be to use Lambda Metrics as suggested by @jarmond. CloudWatch metrics can easily used in CloudWatch Dashboards and are much, much faster to query (while also costing less).

Using the CloudWatch Logs query has a few caveats:

  • Only 7 (IIRC) queries at the same time by default
  • Takes some time to query/search the logs
  • Billed by MB/scanned

So the best advice would be to use the provided metrics and if that doesn't suffice, only then use log file scanning.

Since you only want the successful invocations, the lambda function itself could write the metrics at the end of the request.

To the metrics query:

CloudWatch Logs can't cross reference other logs, so there is no sub-query functionality. You can only filter upon information in the same log line without cross-referencing other log lines.

If you insist in the use case, you would need to parse the logs yourself by using streams or log filters. General idea:

  • Invoke log-specific-lambda upon successful invocation that is logged in CW logs
  • Query CW Logs for the execution time and update a metric

Alternative would be to export CW Logs to S3 and use Athena or other products to do complex queries...

Augunrik
  • 1,866
  • 1
  • 21
  • 28