Why capacity and latency collide
Concurrent players determine traffic, but registered players, regions, game modes, and retained seasons determine capacity. Dropping inactive players from memory reduces cost by making the next return or event-driven reactivation more expensive.
A disk-backed serving layer needs to handle those returning players while active sessions continue. Measure bursts of profile hydration, writes from progression events, and ranking reads together; smooth average traffic understates a launch or season reset.
DESIGN FOR THIS DATA PATH
Where Lavik fits
Use Lavik for application-prepared player context and independently qualified ranking state where retained payload drives Redis capacity. Preserve the authoritative game-state or event system and a replay/rebuild path. Separate low-latency simulation and critical account balances from the read-model evaluation.
Strongest evaluation fit: Growing retained player context with a durable authoritative source and clear regional or seasonal boundaries.
Design decisions that determine the outcome
Separate serving context from authoritative progression
Cache or materialize profiles with explicit source versions. Keep entitlement, inventory, and balance invariants in their authoritative workflow until equivalent failure semantics are demonstrated. A read-model rebuild must not re-award rewards or double-apply events.
Partition the audience deliberately
Include game, region, and season in the key design when they change semantics. Bound profile/document size and ranking windows. Measure small hot cohorts separately from broad returning-player reads; one global board can behave very differently from many regional boards.
Design the return and rollover paths
Throttle reconstruction after a restart or event launch. Define old-season retention, expiry, and replay order. Verify the actual session and sorted-set command sequences rather than inferring framework compatibility from RESP support.
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
> HSET profile:game1:eu:demo42 avatar ranger source_version 18 region eu
3
> HMGET profile:game1:eu:demo42 avatar source_version missing_field
["ranger","18",null]
> EXPIRE profile:game1:eu:demo42 1800
1
> DEL profile:game1:eu:demo42
1
> HMGET profile:game1:eu:demo42 avatar source_version
[null,null]Version 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.
Event and reconnect bursts
Replay returning-player loads alongside ongoing progression writes and social/profile reads.
Player-facing tail latency and timeout rates meet the launch budget with controlled origin traffic.
Season transition
Keep prior seasons accessible while new cohorts populate and old data expires.
Rollover fits memory/SSD headroom and does not degrade current-season read/write deadlines.
Projection rebuild
Inject duplicate and reordered events and rebuild the serving state after interruption.
Profiles and ranks agree with authoritative state; rewards and increments are not applied twice.
These are proposed serving architectures, not claims of game-engine, matchmaking, anti-cheat, or commercial customer validation. Qualify clustering, recovery, and the actual command mix before production adoption.
Translate business scale into capacity
Value payload ≈ retained players × context bytes × game/region variants; size ranked members separately
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 →