Abstract
Autoscaling stateless services is a solved problem. In a sharded datastore, a resize is not a capacity change — it's moving ownership of key ranges between nodes while the cluster serves live traffic, and rebalancing takes long enough that reacting to a traffic peak is already too late. Every partitioned datastore has this problem, not just caches.
This talk is the story of building autoscaling for Snap's caching fleet: hundreds of Valkey and Memorystore clusters, ~20,000 instances across AWS and GCP, serving 250M+ QPS. What worked: a single decision engine combining reactive scaling, scheduled pre-scaling, and emergency circuit breakers, with safety rules that block any action that could hurt a cluster. Unlike one-off right-sizing campaigns whose savings decay the moment traffic shifts, capacity now tracks demand as patterns change, with no engineer in the loop. And along the way: the tuning challenges and edge cases that only show up at fleet scale.
What every attendee should leave with:
- If you shard, you have a resharding path. If it only runs when a human resizes a cluster once a year, it has bugs you haven't found.
- Audit it before an autoscaler finds them for you.
- Scale-out is scheduling; scale-in is data safety — a node can only be removed after its data has been migrated away.
- Know how long a resize takes before tuning anything else. That number decides whether you can react or must predict.
- One writer per cluster — and the system must know when to do nothing.
A pure end-user story: what breaks when nobody can hand-tune anything, ever.
$2,835, Conference (3 days). Current pricing ends September 8th. All pass options.
QCon San Francisco 2026 is a three day conference for senior software engineers, architects and team leads. An international program committee of working engineers selects every session. Patterns and practices, not products and pitches.
From the same track
Tuesday 17 November
10:35 Ballroom BC Session Orderly Keys, Wild Values: Adaptive Compression for Distributed Key-Value Storage Joseph Lynch, Ayushi Singh At Netflix scale - billions of requests per day and petabytes of key-value data - even small inefficiencies in storage and network paths become expensive. 11:45 Ballroom BC Session When Your Users Are Agents: Lessons from Running Serverless Postgres Gwen Shapira Co-Founder and CPO @Nile, Previously Engineering Leader @Confluent, PMC Member @Kafka, & Committer Apache Sqoop Most systems are designed with human users in mind: predictable session lengths, expected traffic spikes, and query patterns shaped by application code. But what happens to your core infrastructure when some of your most active users are autonomous AI agents? 13:35 Ballroom BC Session Adaptive Systems in Production: What Recommendation Systems Can Teach Us About Agents Mallika Rao Senior Engineering Manager @Zocdoc, Previously @Netflix, @Twitter and @Walmart As organizations race to build AI agents, many teams are encountering challenges that feel new: evaluation uncertainty, feedback loops, behavioral drift, exploration versus exploitation, and maintaining user trust in systems that continuously adapt. But these challenges are not new. 14:45 Ballroom BC Session How to Build Online Systems with Object Storage Almog Gavra Co-Founder @Responsive.dev - Building Object-Native Databases, Previously @Confluent and @LinkedIn “Diskless” systems that delegate durability to object storage are everywhere, and for three good reasons: 15:55 Seacliff D Unconference Unconference: Distributed Systems in Production 17:05 Ballroom BC Session Autoscaling 800 Valkey Clusters: Distributed Systems Lessons from Scaling Stateful Fleets Autoscaling stateless services is a solved problem. In a sharded datastore, a resize is not a capacity change — it's moving ownership of key ranges between nodes while the cluster serves live traffic, and rebalancing takes long enough that reacting to a traffic peak is already too late.