Filter / transform / forward logs from AWS Lambda functions before they are sent to CloudWatch Logs.
Note
This is a library for developers. If you are looking for a binary executable, see AWS Lambda Log Filter.
Caution
Possible data loss if you write tons of logs and return immediately. See possible data loss below.
cargo add aws-lambda-log-proxySee examples for more details.
A real world case: AWS Lambda Log Filter.
We need a solution to realize the following features without modifying the code of your existing AWS Lambda functions:
- Reduce the volume of logs to lower the costs.
- Forward logs to other destinations (like Elasticsearch) in real time.
- But keep the EMF lines untouched so Amazon CloudWatch can still retrieve the metrics.
You need a wrapper script to start the handler process (with the _LAMBDA_TELEMETRY_LOG_FD environment variable removed) and pipe the logs to the proxy. The proxy will filter / transform / forward the logs before they are sent to CloudWatch Logs.
We use AWS Lambda Runtime Proxy to intercept AWS Lambda runtime API requests so we can suppress the invocation/next request to process logs as much as possible. You need to override the AWS_LAMBDA_RUNTIME_API environment variable for your handler function to point to the runtime proxy in the wrapper script.
Actually you could just create a shell scripts with the content env -u _LAMBDA_TELEMETRY_LOG_FD "$@" 2>&1 | grep --line-buffered xxx and use a wrapper script to realize a simple filter. However we need this lib to realize the following features:
- Suppress the
invocation/nextrequest to process logs as much as possible. This is very important if you want to stream the logs to external services in real time. - Identify the log lines with EMF so we can keep them untouched.
- Implement custom filters / transformers / forwarders in Rust.
Though we tried our best to suppress the invocation/next request to process logs as much as possible, if you write tons of logs (more than thousands of lines) and return immediately, there might be some logs not processed.
As a best practice, it is your responsibility to do a thorough benchmark test against your own use case to ensure the logs are processed as expected.