SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

How a non-concurrent index on 40M rows silently killed writes for 11 minutes

0
·2 views

A routine database migration at an unnamed company caused an 11-minute write outage on a Wednesday morning when a non-concurrent CREATE INDEX on a 40-million-row orders table collided with a long-running analytics query, triggering Postgres's lock-queuing behavior. Because migrations ran inside the application's startup path, pods could not serve traffic during the incident, and rolling back the deploy offered no safe escape from the partially built index. The team had no choice but to wait out the lock. In response, the engineering team decoupled migrations from application deploys, making schema changes a separate pipeline stage run before code is released. They also adopted lock and statement timeouts, mandatory use of CREATE INDEX CONCURRENTLY, and an expand-and-contract policy ensuring all schema changes remain backwards compatible with the currently running code.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Student Reviews Tencent EdgeOne Makers as Practical Tool for Secure Web Deployment

A Computer Systems and Networking student shared their experience using Tencent EdgeOne Makers, accessed through the DevHandal 2026 Batch 2 program, for deploying academic web projects. The platform streamlined the deployment process by automatically issuing SSL/TLS certificates, routing traffic through global edge nodes, and providing built-in Web Application Firewall and DDoS protection. The student noted significant improvements in page load speeds when multiple users accessed the application simultaneously, thanks to distributed delivery of static assets. Edge Functions on the platform also allowed the student to experiment with serverless concepts such as URL redirection and token validation without burdening the origin server. Additionally, the program offered access to up to 5 million LLM API tokens, enabling students to explore AI integrations in web projects at no upfront infrastructure cost.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Mistral AI Launches Regional Inference Endpoints, Open Model Support, and European Compute Plan

Mistral AI announced on August 11, 2026, a multi-part expansion of its AI platform covering regional inference, third-party model support, and a long-term European compute initiative. The company introduced dedicated endpoints for EU and US inference — api.eu.mistral.ai and api.us.mistral.ai — allowing customers to route data processing by region, though at an additional cost. Mistral also plans to host third-party open-weight models on its platform alongside its own, beginning with GLM-5.2 from Z AI, enabling model flexibility without requiring separate infrastructure stacks. A new SLA-backed Priority Tier has been introduced for mission-critical workloads that require guaranteed capacity commitments. Additionally, Mistral outlined intentions to build a European compute coalition through multi-year enterprise commitments, though this represents a forward-looking plan rather than immediately available capacity.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why Your Guitar Sounds Off Even After Perfect Tuning: Equal Temperament Explained

Digital tuners rely on equal temperament, which divides an octave into twelve precisely equal 100-cent steps, allowing instruments to play in any key without retuning. The human ear, however, perceives intervals as pure when their frequencies form simple whole-number ratios, such as 3:2 for a perfect fifth or 5:4 for a major third. These pure ratios do not align with equal-tempered pitches, creating measurable gaps — most notably, the major third is 13.7 cents sharp and the minor seventh nearly 18 cents sharp. While the ear can detect deviations as small as 5 cents, perfect fifths and fourths deviate by only about 2 cents and largely go unnoticed, whereas thirds and sixths produce the faint audible beating many listeners sense in held chords. Equal temperament is therefore a deliberate mathematical compromise that distributes tuning error evenly across all keys, sacrificing pure intervals for universal playability.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

TLS Everywhere Claims Often Hide Plaintext Gaps Inside Kubernetes Clusters

Many systems that claim full TLS encryption only secure traffic at the edge, leaving internal communications — such as ingress-to-pod, pod-to-pod, and application-to-database connections — unencrypted by default. In Kubernetes environments, ingress controllers like NGINX and Traefik terminate TLS at the boundary and forward plain HTTP to backend pods unless explicitly reconfigured for re-encryption. Database connections on platforms like RDS also default to plaintext unless connection strings explicitly include parameters such as sslmode=require or Encrypt=True. A further risk lies in cluster infrastructure certificates, which expire annually by default in kubeadm setups with no automated renewal, potentially causing a full cluster outage if missed. Closing these gaps can be achieved through targeted measures like backend re-encryption annotations and automated certificate management tools, without necessarily deploying a full service mesh.

0 comments Read more at DEV Community