Abstract
We run our storage and query backend inside our customers' own VPCs. They own the infrastructure. We own the pager. I'm the person who helps installs it, so I've watched this architecture meet a few hundred different production Kubernetes environments and I've kept notes on what went wrong.
This talk is about the operational reality of running a data plane you can't log into. How you roll out a version when every deployment is a different node type, kernel version, CNI, and security policy. How you debug a storage issue using only the telemetry the customer's environment will let you export. What breaks when the customer's admission controller, air-gapped registry, or IPv6-only network meets your Helm chart. And the incidents where none of that worked and we had to get someone on a screen share at 2am.
I'll cover the cases where this model is genuinely better than shipping data to a SaaS backend, and the cases where I'd tell you not to try it.
If you're considering a split control plane / data plane design, or you're an operator being asked to host a vendor's software next to your workloads, you'll leave knowing what you're signing up for.
$2,995, Conference (3 days). Current pricing ends November 10th. 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.
Part of the track
Sponsored Solution Track IIIFrom the same track
Tuesday 17 November
10:35 Pacific LM Session Sponsored session Details coming soon! 11:45 Pacific LM Session Notes From 300 Installs Inside Someone Else's Cloud Adam Hicks Senior Solutions Architect @Groundcover We run our storage and query backend inside our customers' own VPCs. They own the infrastructure. We own the pager. I'm the person who helps installs it, so I've watched this architecture meet a few hundred different production Kubernetes environments and I've kept notes on what went wrong. 13:35 Pacific LM Sponsored Why Your Document AI Fails in Production — And How to Fix the Infrastructure Underneath It Pavitter Josan Solutions Engineer @Nutrient The session explains why document AI that works in demos can fail in production: document structure can be lost during ingestion, rendering can break on edge cases, and extraction can vary across runs. When agents make routing or approval decisions, teams also need governance. 14:45 Pacific LM Session Trust, but Verify: Delivering Quality with Agentic Workflows Adrian Hatch Solutions Engineer @Antithesis In spite of all the recent panic, nothing has really changed about software development. We’re worried about agents being unreliable. Newsflash: human devs are too. We’re worried about agents producing more code than we can read, but did you ever really do all your reviews within 24 hours? 15:55 Pacific LM Sponsored Your Agents Have Amnesia: Lessons from Building a Memory Layer for Multi-Agent Systems Heather Downing Developer advocate @YugabyteDB Multi-agent systems have a data problem nobody warns you about. Agent B sees Agent A's final output and nothing else; the reasoning, the dead ends, and whatever Agent A figured out three steps in all get thrown away.