MongoDB

How to Export FTDC From MongoDB Atlas

This technical how-to shows how to extract FTDC (Full Time Diagnostic Data Capture) from MongoDB Atlas using an undocumented Atlas Admin API (v1.0) that creates a log collection job, polls for success, and downloads a bundle containing per-member diagnostic.data directories including the live metrics.interim file. The author lists the three required inputs (programmatic API key, project ID, replica set name), provides example curl calls, and describes bundle metadata (expiration ~30 days). Atlas sets diagnosticDataCollectionDirectorySizeMB to 400MB, which typically yields 2–5 days of FTDC under real load (longer on idle clusters). The post also warns that the built-in clusterMonitor role grants read access to the oplog and demonstrates a more restrictive diagnostics-only custom role. The author references tooling (Big Hole, keyhole) for reading FTDC locally.

Reduced OPEX 99.5% by Rewriting App in Go

A technical case study describing how the author rewrote a Ruby on Rails-based document search SaaS into a single Go binary (the Project Apario reader) and drastically cut operational expenses. The original PhoenixVault stack incurred roughly $7,000/month in OPEX (ElasticSearch, MongoDB Atlas, OVH hosting, plus other services). Replacing external managed services and Rails with a self-contained Go application running on OVH bare-metal reduced costs to about $33/month (~99.5% reduction). The post documents implementation details — local JSON/.db directories, built-in CORS/CSP, self-signed auto-rotating TLS (auto-tls), rate limiting, IP-ban middleware, multiple search algorithms, and configuration flags — and links to the open-source Project Apario repositories. Publication date: 2026-07-27.

idemkit: Python idempotency library v0.1 released

idemkit is an open-source Python library (v0.1, Apache-2.0) that provides a unified approach to making arbitrary operations safe to retry (HTTP requests, queue consumers, function calls). It implements an atomic claim, an expiring lease based on the storage clock, and a fencing token to prevent late writes from crashed workers, aiming for effectively-once semantics (not exactly-once). The project exposes multiple surfaces (FastAPI/Flask/Django middleware, queue consumer integration, and an @idempotent decorator) and supports several backends including in-memory, Redis, Postgres, MongoDB, and DynamoDB. The package is available via pip and the source code is hosted on GitHub. The article was posted by Violetta Pidvolotska on 2026-07-26.