Why capacity and latency collide
A SKU is rarely one cached record. Markets, sellers, currencies, fulfillment regions, experiments, and content variants create many serving representations. Evicting the long tail saves DRAM by moving latency and database load onto the first shopper who needs it.
SSD-backed reads must still fit product-page and browse deadlines during a promotion. A disk tier that looks acceptable at average traffic can expose latency when shoppers suddenly spread across a wider catalog and refresh writes compete for I/O.
DESIGN FOR THIS DATA PATH
Where Lavik fits
Place application-prepared catalog and offer snapshots in Lavik, with authoritative data and refresh logic kept in the commerce platform. NVMe SSD capacity makes a broader retained read model worth evaluating. The benefit should appear as lower serving-capacity cost at the required latency and freshness, with fewer expensive origin lookups.
Strongest evaluation fit: Large, rebuildable commerce read models where catalog coverage and variant count drive memory spend.
Design decisions that determine the outcome
Keep the purchasing authority clear
Use cached price and availability as serving snapshots with explicit versions and freshness rules. Checkout must perform the platform's authoritative price, inventory, and reservation checks. A fast key-value lookup does not implement stock reservation or a payment ledger.
Make key geometry follow the storefront
Include tenant/seller, market, locale, and snapshot version where they change the response. Bound MGET fan-out for a product grid. Store prepared representations when that reduces expensive joins, while measuring duplicated payload against the saved request work.
Make refresh idempotent
Handle duplicate and out-of-order events in the CDC consumer. Use application versions or reconciliations to prevent stale materialization, and spread expirations to avoid synchronized refill. Lavik does not supply an automatic commerce CDC connector.
A command example executed on Lavik
Docker check passedA functional check using example data, an isolated instance, and actual replies. It verifies the command sequence shown; it is not a performance or end-to-end correctness test of the industry workload.
Inspect actual requests and replies
> SET offer:t1:eu:sku42:v5 "{\"currency\":\"EUR\",\"display_price\":9900,\"version\":5}" EX 300
"OK"
> SET offer:t1:eu:sku43:v5 "{\"currency\":\"EUR\",\"display_price\":7900,\"version\":5}" EX 300
"OK"
> MGET offer:t1:eu:sku42:v5 offer:t1:eu:sku43:v5 offer:t1:eu:missing:v5
["{\"currency\":\"EUR\",\"display_price\":9900,\"version\":5}","{\"currency\":\"EUR\",\"display_price\":7900,\"version\":5}",null]
> DEL offer:t1:eu:sku42:v5
1Version and verification scope
lavik 0.1.0-beta.1 · Minimal package · aarch64 · 2026-09-21
Offline container with a read-only root, temporary data, and a reset between scenarios. Examples use an authenticated local connection. TTL ranges and exact arguments are preserved in the execution receipt.
Execution receipt ↗What should decide the migration?
Write down the application budgets before replaying traffic. Validate these criteria in your own system; published benchmarks are a starting point.
Promotion traffic
Replay hot-item spikes and broad catalog discovery while refreshing offers.
Browse/product-page tail latency and origin traffic satisfy the event plan.
Freshness under update lag
Inject duplicate, delayed, and reordered catalog/price events into your consumer.
Visible versions do not regress, and stale snapshots follow the storefront's explicit policy.
Multi-market capacity
Load realistic seller, market, locale, and version combinations; measure payload duplication.
Compare complete serving cost at equal catalog coverage, freshness, and latency targets.
This design does not replace search, transactional inventory, or payment systems. Category and industry examples are reference architectures for evaluation, not customer deployments.
Translate business scale into capacity
Value payload ≈ SKUs × serving variants × average snapshot bytes × retained versions
lower value-capacity cost
When DRAM costs 20 times as much per GiB as NVMe SSD, the same value payload costs one twentieth as much for media capacity: 95% less. Include index memory, CPU, replicas, storage amplification, and recovery headroom in the complete deployment.
Calculate with your capacity prices →