[observability] Observability Coverage Report - 2026-08-07 #51237
Closed
Replies: 1 comment
|
This discussion has been marked as outdated by Daily Observability Report for AWF Firewall and MCP Gateway. A newer discussion is available at Discussion #51479. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Executive Summary
Analyzed 20 representative workflow runs from the last 7 days using the repository-wide
logsfetch output saved under/tmp/gh-aw/aw-mcp/logs/, then capped the audit to 20 runs per workflow-report instructions. All 20 sampled runs were firewall-enabled and all 20 showed MCP Gateway startup inrun_summary.json, but recursive artifact discovery found zero retainedaccess.logfiles and zero retainedgateway.jsonlorrpc-messages.jsonlfiles.This is a systemic observability retention gap rather than a single-workflow defect. The saved bundles still contain
summary.jsonin 20/20 runs andagent-stdio.login 20/20 runs, and those derived artifacts show 1426 firewall requests across 33 unique domains with 106 blocked requests (7.4%). That means execution-time telemetry existed, but the canonical raw logs needed for debugging were not preserved in the run artifacts.Key Alerts and Anomalies
🔴 Critical Issues:
run-<id>/found noaccess.log, including the canonicalsandbox/firewall/logs/squid-logs/access.logand legacysandbox/firewall/logs/access.loglocations.run-<id>/found neithergateway.jsonlnorrpc-messages.jsonl, so there is no canonical per-call trace left for postmortem debugging.aw_info.jsonwas absent from all sampled bundles; only 2/20 sampled runs retained anawf-config.jsonfile, so configuration provenance is inconsistent even when derived firewall summaries are present.safe_output.jsonlwas absent from all sampled bundles. Safe-output job steps completed, but the structured safe-output log was not retained as an artifact.agent_output.jsonwas present in 19/20 sampled runs, so supplementary telemetry coverage is better than raw firewall/MCP coverage but still not complete.Coverage Summary
access.log)gateway.jsonlorrpc-messages.jsonl)📋 Detailed Run Analysis
Firewall-Enabled Runs
Entries/Allowed/Blocked are derived from
summary.jsonbecause rawaccess.logwas missing in every sampled run.Missing Firewall Logs (access.log)
MCP-Enabled Runs
Missing MCP Telemetry (no gateway.jsonl or rpc-messages.jsonl)
🔍 Telemetry Quality Analysis
Firewall Log Quality
access.logfiles found anywhere under the 20 sampledrun-<id>/directories.Gateway Log Quality
gateway.jsonland norpc-messages.jsonlin any sampled run directory.run_summary.jsonconfirms MCP Gateway startup in all sampled runs.Additional Artifact Coverage
agent-stdio.log: 20/20agent_output.json: 19/20summary.json: 20/20run_summary.json: 20/20awf-config.json: 2/20 (runs §31223940604 and §31223955453)Healthy Runs Summary
Recommended Actions
access.login the agent artifact bundle for every firewall-enabled run, including workflow_call-prefixed paths, and fail the observability upload step when the file is absent.mcp-logs/gateway.jsonlormcp-logs/rpc-messages.jsonlfor every MCP-enabled run and add a post-agent assertion that marks the run degraded when both are missing.aw_info.jsonorawf-config.json, plussafe_output.jsonl, so raw telemetry can be correlated with configuration and safe-output activity.📊 Historical Trends
References:
All reactions