How the depla and game updates work without interruptions

Why zero-downtime casino releases

Any "micropause" in iGaming is lost bets, sessions and trust. Updates should occur unnoticed by the player: bets continue to be accepted, the stream does not break, the wallet and ledger remain consistent, and the metrics do not jump. The key is version discipline, contract compatibility, and step-by-step, observable calculations.

Reference principles

1. Forward/backward compatibility. New versions are required to understand old events/fields, and old clients are required to safely ignore new ones.

2. Immutable assets. Static and game resources are given with hash names; no "overwriting" of files.

3. Split write/read path. Cash transactions (hold/settle) are isolated and atomic, UI/assets change independently.

4. Observability as a contract. Release without tracing/metrics - prohibition.

5. Rollback is the same norm as release. Ready-made images, both-way migrations, rollback button without manual shamanism.

Zero-downtime architecture in practice

1) Versions and contracts

SemVer for API/events: 'MAJOR. MINOR. PATCH ', the' eventVer/contractVer 'field in each message.

Expand→Migrate→Contract for database schemes: first add fields/indexes (expand), then background migration (migrate), and only after turning off the contract.

Dual-write/Dual-read when changing critical logic (for example, calculating a bonus): for a while we write to the old and new tables, compare.

2) Assets and CDNs

Bundles/sprites/texts: 'app. a1b2c3. js`, `paytable. 98f0. png ', headers:


Cache-Control: public, max-age=31536000, immutable

Asset manifest on server/CDN. We switch the link to the new manifest - players instantly get a fresh UI, old pages continue to live with the same files (without broken links).

Tag-purge for frequently changing JSON (catalogs/banners) + 'stale-while-revalidate' for soft shift.

3) Traffic strategies

Blue-Green for critical components (wallet/ledger/bridge): keep two identical environments, switch ingress/virtual service in seconds.

Canary for API/game gateways: 1-5% of traffic → SLO/Fin Delta analysis → expand to 100%.

Feature flags for UI and mechanics: include for a segment, region or game, without releasing code.

4) Live games and RNG

Live (WebRTC/LL-HLS):

We change the player/overlays separately from the video stream (different domains/configs).
Server time and soft switching to a new signal.

RNG/Client:

The new build of the game is given as a new version of the resource. Players who have already started the round end it on the old client/rules.
В `round. settled 'fix' calcVer '- a version of the calculation engine so that controversial rounds are played "as it was."

5) Wallet and ledger - how not to break money

One writer per shard. Writer switching is a separate procedure (with locks) and only within the AZ/region.

Idempotence on all paths: 'bet. place`, `round. settle`, `payout. request`, `cashier. webhook` — с `idempotencyKey`.

PITR and shadow checks: during canary calculation, we duplicate the wiring in the "shadow," check the units (GGR/NGR) before the promotion.

Step-by-step release without downtime (reference scenario)

1. Preparation:

Release contract: that 'PATCH/MINOR/MAJOR', compatibility matrix.
'Expand 'migrations are applied in advance, index builds are applied online.
Assets loaded into CDN, manifest ready.

2. Canary Start (API/Game Gateways):

1-5% of traffic. Watching p95/99'bet. place ',' settle ',' error _ rate ',' VOID 'growth, payment imbalance.
Compare financials with control group (delta

3. Extension/Switch:

Increase the percentage to 25/50/100 or switch Blue-Green on the money core.
We turn on the features with flags (locales/games/regions).

4. 'migrate'migration:

Background jabs carry data/progress, dual-write is enabled.
Telemetry validates a metric match.

5. Optimization and 'contract':

Disable legasi-reading, remove "shadows" in the next MINOR/MAJOR.
Update the directories of schemes/events, close the deprecate.

6. Documentation and retro:

Post-mortem/retro even without incident: what to improve in SLO, alerts, checklists.

Observability and SLO at release

SLI/SLO:

`bet. place p95 '(target ≤ 150-250 ms),' error _ rate '(<0. 3%), `round. settle p95` (≤2 с), `payout. submit p95 '(≤800 ms).
Live QoS: `webrtc_rtt_ms`, `dropped_frames`, `aborted_rounds`.
Version tags: 'buildId', 'semver', 'contractVer', 'calcVer' in logs and trace.
Fin deltas: GGR/NGR/hold comparison by old/new branch segments.

Rollbacks without pain

Blue-Green: instant route return to "blue."

Canary: reduce traffic to 0%, disable features with a flag.

Assets: the old manifest remains immutable, players on the old pages do not break.

Data: if there was dual-write, when rolling back, we read the "old" source; there were no destructive migrations before confirm.

Organization and Processes

Change windows with SRE protection: release slots for peaks/sporting events are not touched.

Runbooks: checklists for ingress switches, database roles, feature flag, contact chains.

Dark-launch: turn on everything except visibility in the UI, run the "hidden" load.

Frequent errors (anti-patterns)

Overwriting assets without versioning → broken clients and pink squares.

Breaking event changes/APIs "quietly" → the dump of provider and dashboard integrations.

Schema + logic migrations in one step without dual-write → financial discrepancies.

Lack of idempotency → double debits in retrays.

A single switch at once 100% without canaries and metrics.

Mixing UI release and calculated kernel in one deployment.

No rollback plan or rollback requires "manual" SQL.

Zero-downtime release checklist

Contracts and data

SemVer + 'contractVer/eventVer/calcVer' are written and documented.
'expand 'migrations applied beforehand; 'migrate' in the background; 'contract'in the next cycle.
Dual-write/Dual-read where the finlogy changes.

Infrastructure

CDN: immutable assets, manifest, tag-purge, 'stale-while-revalidate'.
Blue-Green for the core of money; Canary for API/game gateways.
Feature-flags for UI/mechanic; flags are managed without deploy.

Observability

Trails with 'buildId/semver/calcVer'; dashboards SLO and fin delta.
Alerts to growth 'VOID', 'error _ rate', degradation of live-QoS.

Rollback and safety

rollback (ingress/routing) button, old manifest available.
PITR and shadow postings to check the ledger.
The rollback test has been worked out on the stage and in the small production segment.

Processes

Runbooks switches; consistent change windows.
Dark-launch/canary; retro after release.

Zero-downtime in iGaming is a system practice: versions and contracts, immutable assets and CDN, blue-green/canary, non-idle migrations, idempotent money and strict observability. Following this checklist, you update games and the platform so that the player does not notice anything - except that everything has become faster and more stable.