Case study · Flipkart

OTA CodePush Platform Migration

Re-architecting a high-traffic OTA delivery pipeline from Azure to GCP with Redis-backed caching and phased zero-downtime cutover.

Software Development Engineer II · 2025

0.2%

Release success rate

—

Critical downtime events

0M+

Monthly bundle fetches

The problem

—Legacy Azure CodePush paths showed rising P95 bundle latency as release cadence and cohort size grew quarter over quarter.
—Operational cost on Azure compute + egress climbed with traffic spikes during festival sale events.
—Mobile clients needed predictable sub-second metadata resolution — cache misses directly impacted update adoption.

The approach

—Designed GCP Cloud Functions + Redis architecture with metadata hot paths served from memory.
—Ran shadow traffic and canary cohorts before full cutover — no critical downtime during migration window.
—Built observability dashboards for cache hit ratio, function cold starts, and rollback triggers.

System flow

High-level architecture

React Native Clients

CDN Edge

GCP Functions

Redis Cache

Bundle Storage

Performance metrics

Metric	Before	After	Change
Bundle delivery P95	820ms	510ms	-38%
Cache hit ratio	61%	89%	+28 pts
Failed release rate	2.1%	0.4%	-81%

Cost & efficiency

Area	Before	After	Notes
Monthly infra (OTA path)	₹4.8L	₹3.1L	~35% reduction post-rightsizing
Egress per 1M bundles	₹12.4K	₹7.9K	CDN + regional caching strategy
Incident MTTR	47 min	18 min	Improved observability & runbooks

Implementation snapshot

cache-first bundle resolution

async resolveBundle(appId, platform, version) {
  const cacheKey = `ota:${appId}:${platform}:${version}`;
  const cached = await redis.get(cacheKey);
  if (cached) return JSON.parse(cached);
  const manifest = await fetchFromStorage(appId, version);
  await redis.setex(cacheKey, 3600, JSON.stringify(manifest));
  return manifest;
}

Rollout phases

Assess & map

Traffic profiling, Azure cost baseline, failure-mode analysis.

GCP + Redis

Functions, cache layer, CI/CD parity across staging/prod.

Phased cutover

Shadow → 5% → 25% → 100% cohort migration.

Tune & operate

Cache warming, on-call runbooks, cost reviews.

Tech stack

React NativeNext.jsTypeScriptGCP Cloud FunctionsRedisAzure (legacy)CI/CDJest

Want to dive deeper into this kind of work? Let's talk.