Built an observability layer that treats CI/CD pipelines as a queryable system and brings job results across repositories, services, environments, runs, runners, and commits into one diagnostic interface.
Platform-agnostic task instrumentation for build, test, deploy, scan, and release jobs, CI metadata normalization for GitHub Actions and GitLab CI, OpenTelemetry Collector Gateway, Grafana dashboards, optional JUnit test spans, logs, traces, runner metrics, and commit-aware regression detection.
Failed build, test, deploy, scan, and release jobs were trapped inside individual CI runs, logs, artifacts, and provider-specific pages.
A Grafana observability surface connects task results, Loki logs, Tempo traces, Prometheus metrics, runner resource signals, and commit metadata across runs and repositories.
The system instruments CI/CD tasks with a platform-agnostic Python wrapper, enriches execution telemetry with normalized CI metadata, and routes logs, traces, and metrics through OpenTelemetry Collectors into Grafana, Loki, Tempo, and Prometheus. The dashboard gives teams one diagnostic interface for failed builds, tests, deployments, scans, migrations, release jobs, and custom automation tasks.
Selected views appear where they provide concrete proof of architecture, workflow, or reporting behavior.
Modern delivery pipelines contain builds, tests, security scans, infrastructure checks, deployments, database migrations, releases, and custom scripts. Native CI interfaces are effective for inspecting one run, but weaker when the question spans jobs, repositories, services, environments, runners, commits, and time ranges.
A Python command wrapper runs inside the CI job and executes the original build, test, deploy, scan, migration, or release command as a subprocess. It captures stdout and stderr, emits task spans and status records, optionally parses JUnit XML into per-test-case spans, and exports telemetry over OTLP.
CI tasks run as independent short-lived processes, so trace correlation cannot rely on shared in-memory state. The implementation derives trace and span identifiers from CI metadata and normalizes GitHub Actions and GitLab CI variables into a shared context.
The system turns build, test, deploy, scan, migration and release jobs into observable tasks that can be queried across providers, repositories, runs and commits.
The dashboard provides cascading selectors for namespace, service name, job target, task target, and time range. It scopes from a full delivery landscape down to one task and keeps high-signal diagnostic panels visible.
Failed tasks can be inspected through raw Loki logs and Tempo traces without leaving the diagnostic workflow. When JUnit XML is available, the trace includes per-test-case spans so failed test evidence is correlated with the parent task.
The prototype was evaluated with Microsoft eShop on GitHub Actions and Google Online Boutique on GitLab CI to test cross-provider and service-based workflows.
The system turns CI/CD from a collection of isolated pipeline pages into a queryable diagnostic surface. The business value is faster failure investigation, clearer regression ownership, and less manual navigation through CI runs, logs, artifacts, and provider-specific screens.
The gallery collects additional anonymized screenshots without hiding the technical case study inside a slider.
