track-request

Follow one request through a serverless stack: every lambda it invoked, every batch job it submitted, the whole tree with durations, memory and errors per node:

$ track-request /aws/lambda/api-prod 8f6e1b

/aws/lambda/api-prod 8f6e1b (1.24s, 134/512 MB, 210 lines, 0 errors, 2 subrequests)
  /aws/lambda/process-order 2a9c7d (0.98s, 201/512 MB, 118 lines, 0 errors, 1 subrequests)
    /aws/batch/job/default/7f3e5a (12.40s, ?/2048 MB, 55 lines, 0 errors, 0 subrequests)
  /aws/lambda/send-email 91dd02 (0.31s, 84/512 MB, 12 lines, 0 errors, 0 subrequests)

Queries cloudwatch insights for the request logs, finds lambda invocations and batch submissions in the aws sdk debug lines, fetches those too, recurses.

Usage

usage: track-request [-h] [-v] [-f FORMAT] [-o OUTPUT] [-d DAYS]
                     [-l LIMIT] [-m MAXDEPTH]
                     log_group_name request_id

• -f FORMAT - row template; the default shows group, id, duration, memory, lines, errors and subrequest count
• -o OUTPUT - also write the full logs to a file; a name ending in .gz compresses, the prefix log-group: splits output into one file per log group
• -d DAYS - how far back to search: a number of days, or start|end timestamps
• -l LIMIT - stop after this many log lines per request
• -m MAXDEPTH - traversal depth, 0 for unlimited

Install

A python package with a single dependency, botocore, on python 3.10 or newer:

uv tool install git+https://github.com/paunovic/track-request.git

or plain pip:

pip install git+https://github.com/paunovic/track-request.git

credentials come from the ambient aws chain - environment variables, profiles, etc.