ALEKSANDR GRIBAKIN
INTROCRAFTSTACKMILESTONESCONTACT
Aleksandr Gribakin
☕
-- FPS
INITIALIZING...

< ALEKSANDR GRIBAKIN />

Lead Full Stack & DevOps Engineer
15 Years in Code
Fifteen years building and scaling distributed systems, across the industry's move from bare-metal servers and monoliths to cloud-native, Kubernetes and GitOps. I take on the work teams postpone: technical debt, CI/CD built from nothing, and keeping things up.
Kubernetes Terraform Docker GitLab CI PostgreSQL Go Node.js

Impact By The Numbers

Career Timeline

Projects & Impact

Migration Bare-Metal to Hybrid Cloud Terraform, Ansible, AWS, GCP, Linux △ 30-40% cost cut / staged cutover Staged migration off physical servers into hybrid cloud. Inventory first, then parity, then traffic — never all three at once. Platform Kubernetes Platform Kubernetes, Helm, ArgoCD, Terraform, Prometheus △ 99.9% uptime / self-serve namespaces A cluster product teams can use without learning Kubernetes. Namespaces, quotas, ingress and monitoring arrive with the namespace. IaC Infrastructure as Code Baseline Terraform, Ansible, GitLab CI, Vault, Linux △ 100% reproducible / 3 environments Every environment described in code and rebuilt from zero on demand. No snowflake servers, no undocumented hand edits. Tooling Local and Production Parity Docker, Docker Compose, Terraform, Make, Linux △ One image / 4 environments The same images and the same configuration shape everywhere. Kills the "works on my machine" class of bug outright. Delivery Standardised Delivery Pipelines GitLab CI, Jenkins, Docker, Helm, GitHub Actions △ 15 min release / dozens of services One pipeline template across dozens of services. Release time from days to about fifteen minutes. Deployment Zero-Downtime Deployment Kubernetes, ArgoCD, NGINX, Helm, Prometheus △ Zero downtime / 90s rollback Rolling and blue-green releases with health-gated promotion. A bad release rolls back before most users notice. GitOps GitOps Delivery Model ArgoCD, Git, Helm, Kustomize, Kubernetes △ Git as source of truth / auto-reconcile The repository is the desired state. Anything applied by hand gets reverted by the reconciler, on purpose. Architecture Monolith to Services PHP, Symfony, Go, gRPC, RabbitMQ △ Incremental cutover / no freeze Strangler-fig extraction with the monolith serving traffic throughout. Seams chosen by data ownership, not by folder layout. API API Layer and Contracts Node.js, TypeScript, OpenAPI, gRPC, NGINX △ Generated clients / versioned contracts REST outward, gRPC inward, one generated contract for both. Clients break at build time instead of in production. Queues Async Job Processing Python, FastAPI, Celery, RabbitMQ, Redis △ Idempotent retries / DLQ triage Queue-backed background work with idempotent handlers, retries with backoff and a dead-letter queue that gets read. Database PostgreSQL Under Load PostgreSQL, PgBouncer, Patroni, Prometheus, Linux △ Read scale-out / partitioned writes Replication, partitioning and connection pooling. Most wins came from reading query plans, not from adding hardware. Caching Multi-Level Caching Redis, NGINX, CDN, Node.js, PostgreSQL △ 4 tiers / explicit invalidation CDN, edge, application and query caches with explicit invalidation. Every layer answers what it is allowed to be wrong about. Search Search and Document Store Elasticsearch, MongoDB, Python, Kafka, Redis △ Live reindex / change-stream fed Full-text search fed by a change stream, with reindexing that runs against live traffic instead of a maintenance window. Web SPA with Server-Side Rendering React, TypeScript, Vite, Node.js, NGINX △ SSR first paint / typed end to end Rendered on the server for first paint and crawlers, hydrated on the client for everything after. Design System Component Library Vue.js, TypeScript, Webpack, Storybook, CSS △ 40+ components / WCAG AA A shared component set with tokens, documented states and accessibility built in rather than retrofitted. Monitoring Observability Stack Prometheus, Grafana, Loki, Elasticsearch, Alertmanager △ Correlated signals / SLO-based alerts Metrics, logs and traces correlated by request id. Dashboards answer questions instead of displaying numbers. Process Incident Response Grafana, Alertmanager, Ansible, Git, Linux △ Owned rotations / current runbooks On-call with owned services, runbooks that are actually current, and postmortems that produce changes rather than blame.

Milestones & Recognition

☕ Buy me a coffee
USDT TRC-20 wallet QR code
TWTCH2ZhyKvzZU1ph6h1d4vTHpHyJDkN5i
USDT TRC-20

Async Job Processing

Queues / Backend — Python, FastAPI, Celery, RabbitMQ, Redis

Reference architecture. Client code is under NDA, so this page describes the design and the reasoning behind it rather than a specific deployment.

Background processing for everything that should not happen inside a request: imports, exports, notifications, report generation. Handlers are idempotent because at-least-once delivery means a job will eventually run twice, and pretending otherwise just relocates the bug.

Retries use exponential backoff with a cap, and anything exhausting its retries lands in a dead-letter queue with the original payload and failure reason. The dead-letter queue has an owner and a dashboard, because an unread one is just a slower way to lose data.

Key Metrics

Features