Skip to main content
George Santana / krakenserv.dev

01 — The Problem

Small engineering teams need dependency-aware pipelines, retries, and live logs for backups and data tasks, yet deploying heavy orchestrators with Redis, Postgres, and multi-container stacks adds disproportionate operational overhead.

02 — Architecture & Flow

A single zero-CGo static Go binary bundling a 5-field cron engine, topological DAG resolver, WAL-mode SQLite persistence (modernc.org/sqlite), and a dark web dashboard embedded directly via go:embed.

01 YAML Load & Cycle Check

Kahn topological sort validates dependency graphs and rejects cyclic workflows upfront.

02 Cron & Ad-hoc Scheduling

Triggers workflows on schedule or CLI demand with deterministic step state transitions.

03 Setpgid Process Isolation

Subprocesses execute in isolated groups with exponential backoff and clean signal teardown.

04 SSE UI & Alert Dispatch

WAL SQLite audit trail streamed over SSE and routed asynchronously to Discord or Telegram.

03 — Technical Solution

Implemented hot-reloading YAML workflow definitions, upfront cycle detection, exponential-backoff retries, dedicated POSIX process-group isolation (Setpgid) to prevent zombie subprocesses, and live Server-Sent Events (SSE) log streaming.

04 — Production Results

  • Zero-dependency single-binary deployment managed cleanly under systemd without Redis or Docker.
  • Clean subprocess teardown (SIGTERM followed by SIGKILL) preventing orphaned processes.
  • Offline CLI fallback enabling ad-hoc pipeline execution and log inspection even when the daemon is stopped.

Architecture Inside a Single Binary

Zero operational drag on Linux nodes

GlowKron is built around one constraint: an automation runner should never be harder to keep alive than the jobs it schedules.

  • Upfront graph verification: Before registering or triggering any workflow (glowkron validate), the step graph passes through Kahn’s algorithm to catch circular dependencies before a single command runs.
  • Daemonless CLI fallback: CLI commands read and write the WAL SQLite store directly, allowing operators to inspect historical runs or trigger emergency pipelines in-process even if the background service is offline.
  • Race-checked concurrency: Developed with go test -race to validate concurrent step scheduling, log streaming, and clean signal shutdowns.

Interactive engineering demo

LIVE INTERACTIVE DEMO // REACT ISLAND

Test the engineering logic live

Topological workflow execution (Kahn's algorithm) with process-group isolation and exponential retry backoff.

01 · rootIDLE

pg_dump -Fc prod_db

02 · after: dumpIDLE

gzip -9 backup.dump

03 · after: compressIDLE

s3 cp backup.dump.gz

Execution log stream

[idle] Cycle check passed (3 topological steps ready).

05 — CONTACT & SCOPE

Shall we talk about your next system?

Share a short brief of what you want to launch or fix. You will get a clear technical approach, realistic timeline, and next steps.

Project type

Estimated budget

Target timeline

Spam-protected with Zod boundary validation and Cloudflare Turnstile.

Direct technical reply within 24 business hours.

GitHub ↗ LinkedIn ↗