Five years at Microsoft on Azure SQL — Hyperscale control plane, then Serverless. Fastest promotion to Senior on the team; tech lead for a group of 6. I owned features end-to-end across India and US teams: design, rollout, on-call, billing, and customer-facing reliability.

Azure SQL Serverless powers on the order of ~3.5 million databases worldwide. Microsoft does not publish exact counts; that figure reflects adoption trends and internal scale estimates from my time on the team. The logical pause/resume path I redesigned runs at that scale — tens of thousands of allocation workflows per region, per day — as described in our SIGMOD 2024 paper.

Azure SQL DB Serverless · Mar 2022 – Sep 2025

Core workflow contributor and primary stakeholder for Serverless products — logical pause/resume infrastructure, metadata workflows, and billing systems.

Logical pause/resume

Feature owner and technical lead for a more efficient logical pause/resume design for serverless databases. At Azure scale, the earlier approach routed every pause and resume through the central control-plane service — thousands of databases pausing at once in a region competed with backups, security workflows, and everything else that service manages.

I led a redesign that put the database engine in charge of its own pause/resume state: logins reach the engine directly instead of detouring through the control plane, billing updates happen on a direct path, and the control plane handles metadata asynchronously. The result was the most efficient logical pause/resume path in production — 60% reduction in workflow latency for customers waking paused databases, lower login latency, and meaningful COGS savings on the control plane.

  • Designed, implemented, and rolled out end-to-end to 60+ Azure regions, with fallback and rollback mechanisms across services spanning the data and control planes
  • Handles tens of thousands of pause/resume workflows per region, per day at serverless scale
  • Ran login stress testing and cross-module upgrade-failure testing; built org-level dashboards tracking COGS savings, billing correctness, and rollout reliability; shipped auto-mitigation runners for stuck workflows and under/over-billing
  • First shipped inside the SQL DB Free offer — a multi-service rollout coordinated across deployment trains, delivered as the fastest-implemented feature on my team
  • Co-authored the SIGMOD 2024 paper on proactive resource pause/resume — a team paper; I was one of the authors, not the presenter
Diagram comparing earlier logical pause/resume (login routed through the control plane) with the redesign (the database engine owns pause/resume state)
The core idea: move logical pause/resume ownership into the database engine and take the control plane off the login hot path.

Utility DB for firewall rules

Architected a regional Utility DB to cache firewall rules — reducing unnecessary master activations and compute COGS.

  • Intercepted ARM API calls and redirected to Utility DB, cutting activation churn
  • Designed deactivation workflows with race-condition handling and fallback paths
  • Led threat model reviews with security teams; built alerting for data integrity and latency

Activation latency & on-call

  • Investigated user DB activation latency and master DB dependencies; proved parallel activation behavior with real-time dashboards
  • Validated that master deactivation doesn't hurt user DB latency — enabling broader rollout of cost-saving deactivation
  • On-call for Sev-2/Sev-3 incidents: resume failures, metadata corruption, broken logical DB states from compute exhaustion
  • Built observability and auto-mitigation runners; led RCAs and shipped fixes for duplicate activity logs and pause latency propagation

Billing & shipped products

  • Migrated billing logic from Storage V1 to V2; reduced overbilling risk by refining logical resume triggers
  • Found and fixed billing bugs — wrote runners around under- and over-billing logic
  • Shipped Azure SQL DB Free offer (Spinnaker) end-to-end
  • Shipped Trident Native SQL in Microsoft Fabric with Power BI integration
  • Designed features across modules; exposed APIs consumed by microservices

Azure SQL DB Hyperscale · Jun 2020 – Feb 2022

Hyperscale control plane regional expert — backend systems for write-heavy customers at scale.

  • Auto-scaling page servers for customers migrating to Hyperscale
  • File-system configuration management via automated workflows
  • Automated reliable rollouts to Hyperscale's storage engine; billing logic to prevent incorrect charges during migration
  • Management service committer; production incident mitigation and root-cause analysis

What I took away

How to own distributed backend systems at scale — design for latency, COGS, and reliability together; build observability that makes on-call survivable; and ship customer-facing products (Free tier, Fabric integration) while keeping the infrastructure underneath honest.