Overview
InMobi operates one of the world’s largest independent mobile advertising platforms. Their customer workloads — running on Azure — needed to be migrated to Google Cloud Platform without disrupting real-time ad processing at massive scale.
I was brought in as Lead DevOps Engineer and program lead, owning both technical execution and stakeholder governance from start to handover.
Context
InMobi’s platform processes advertising decisioning in real time. The migration had to maintain strict latency SLAs and handle traffic spikes with no tolerance for prolonged outages. There was also a hard delivery timeline tied to contract commitments.
Problem Statement
Migrate mission-critical, high-throughput Azure workloads to GCP while:
- Maintaining performance under 4.5M+ requests per second
- Meeting hard delivery timelines
- Ensuring IAM, security, and compliance alignment
- Coordinating across multiple application and infrastructure teams
Constraints
- No extended maintenance windows — near-zero downtime was non-negotiable
- Multiple application teams with varying readiness levels
- Strict InfoSec requirements for network isolation and IAM governance
- Audit-ready artifacts required at handover
Architecture Design
The target GCP architecture was designed around three principles: scalability, isolation, and observability.
Key architectural decisions:
- Compute: GKE for containerized workloads, GCE for stateful services not yet containerized
- Networking: VPC Service Controls to enforce data perimeter; dedicated subnets per workload tier
- IAM: Least-privilege service accounts; workload identity for Kubernetes workloads
- CI/CD: Jenkins pipelines standardized with consistent build/deploy patterns across teams
- IaC: Full Terraform coverage — no manual resource provisioning
Technology Stack
| Layer | Technology |
|---|---|
| Compute | GKE, GCE |
| Networking | GCP VPC, VPC Service Controls, Cloud NAT |
| IAM | GCP IAM, Workload Identity |
| IaC | Terraform (modular, team-specific) |
| CI/CD | Jenkins |
| Observability | Cloud Monitoring, Cloud Logging |
| Security | VPC Service Controls, Cloud Armor, Secret Manager |
Decision Rationale
Why GCP over staying on Azure? InMobi had existing GCP investment and the migration aligned with their long-term cloud strategy. GKE’s operational maturity and GCP’s global network edge made it the right target for a high-throughput ad platform.
Why Terraform from scratch over lift-and-shift? Migrating without IaC would have created unmanageable infrastructure drift. Building Terraform modules per team enforced consistency and made rollback tractable.
Why Jenkins over moving to Cloud Build? Teams were already familiar with Jenkins. Introducing a new CI system mid-migration would have added risk. Standardization on Jenkins was a deliberate risk reduction.
Challenges
Scale coordination: Coordinating migration sequencing across multiple application teams — each with different readiness levels, different criticality, and different codeowners — was the hardest operational challenge.
Cutover sequencing: Dependencies between services meant cutovers had to be sequenced precisely. Cutting over a downstream service before its upstream was ready on GCP would cause cascading failures.
IAM complexity: InMobi’s Azure IAM model didn’t map cleanly to GCP. Translating service identities and access policies required careful design to avoid both over-permissioning and functional breakage.
Outcomes
- Successful migration of all in-scope workloads with near-zero customer impact
- Platform handling 4.5M+ requests per second on GCP post-migration
- 100% component accountability with audit-ready documentation
- Improved infrastructure consistency through Terraform standardization
- Complete handover with SOPs, architecture documentation, and runbooks
Business Impact
The migration enabled InMobi to consolidate cloud spending, leverage GCP’s global network for lower advertising latency, and align infrastructure with their long-term platform strategy.
Lessons Learned
Migrations are org problems first. The technical architecture was relatively straightforward. The hard work was sequencing cutovers across teams with competing priorities, building trust with InfoSec stakeholders, and maintaining clear communication with executive sponsors during a program with zero tolerance for slippage.
Own the narrative, not just the tickets. Weekly executive updates, clear risk tracking, and proactive escalation management kept the program on track. Technical excellence alone doesn’t deliver a program — governance does.