[ArtStudio] Honest GPU queue position — GpuCoordinator lease-queue emit → worker → SignalR → timeline #366

New Issue

2026-07-04T22:37:57Z

spikerj commented

2026-07-04 22:37:57 +00:00

Split from #364 (closed); epic #346. The stage timeline's queuePosition input has been plumbed since P5 but never fed, because IGpuLeaseService.AcquireAsync blocks inside the RPC with no queue visibility.

Scope (end-to-end):

GpuCoordinator: while a lease request waits, publish periodic queue-position updates (position + waiting count) keyed by the requesting service/job — either events on the existing gpu.lease.* direct-exchange family or a new lightweight topic; in-memory ledger stays authoritative (single replica unchanged)
SharedGpuLeaseService (or a callback seam on AcquireAsync): surface position updates to the caller while waiting
ArtPipeProcessor: forward position changes as art.asset.stage.progress.<stage> events (state Queued, new queuePosition field on ArtAssetStageProgressEvent) — throttled like progress
Relay/SignalR: field passes through ReceiveArtAssetStageProgress untouched
Angular: SignalRArtAssetProgressSource maps queuePosition into the existing timeline input; polling fallback may show position from the stage-run doc if the worker also persists it (optional)

Acceptance: with two submissions racing for the GPU, the second student's timeline shows "waiting for GPU — #2 in line" (or equivalent i18n), updating as the queue drains; staff-priority behavior unchanged; no events emitted when a lease is acquired immediately.

Split from #364 (closed); epic #346. The stage timeline's `queuePosition` input has been plumbed since P5 but never fed, because `IGpuLeaseService.AcquireAsync` blocks inside the RPC with no queue visibility. **Scope (end-to-end):** - [ ] GpuCoordinator: while a lease request waits, publish periodic queue-position updates (position + waiting count) keyed by the requesting service/job — either events on the existing `gpu.lease.*` direct-exchange family or a new lightweight topic; in-memory ledger stays authoritative (single replica unchanged) - [ ] `SharedGpuLeaseService` (or a callback seam on AcquireAsync): surface position updates to the caller while waiting - [ ] ArtPipeProcessor: forward position changes as `art.asset.stage.progress.<stage>` events (state Queued, new `queuePosition` field on `ArtAssetStageProgressEvent`) — throttled like progress - [ ] Relay/SignalR: field passes through `ReceiveArtAssetStageProgress` untouched - [ ] Angular: `SignalRArtAssetProgressSource` maps `queuePosition` into the existing timeline input; polling fallback may show position from the stage-run doc if the worker also persists it (optional) **Acceptance:** with two submissions racing for the GPU, the second student's timeline shows "waiting for GPU — #2 in line" (or equivalent i18n), updating as the queue drains; staff-priority behavior unchanged; no events emitted when a lease is acquired immediately.

spikerj added the enhancement label 2026-07-04 22:37:57 +00:00

spikerj referenced this issue

2026-07-04 22:45:09 +00:00

[ArtStudio/Infra] Unified elastic GPU capacity — one coordinator model that scales from a single 24 GB card to multi-node, and runs identically on the M5 dev Mac #367

spikerj commented

2026-07-04 23:12:03 +00:00

Resolved: backend PR #73 + angular PR #122, both merged to master. GpuCoordinator emits change-only queue-status (plus 5 s heartbeat) on a new additive gpu.lease.queue-status exchange; IGpuLeaseService gained an optional callback via a default interface method (all other GPU tenants zero-diff); ArtPipeProcessor forwards position changes as Queued-state progress events and persists queuePosition on the stage run (SignalR and polling covered); the timeline now shows "Waiting for GPU — #N in line" live. 418/418 backend + 128/128 lib tests. Live two-submission race noted for the next full-stack session (unit tests pin ordering semantics). Closing.

Resolved: backend PR #73 + angular PR #122, both merged to `master`. GpuCoordinator emits change-only queue-status (plus 5 s heartbeat) on a new additive `gpu.lease.queue-status` exchange; `IGpuLeaseService` gained an optional callback via a default interface method (all other GPU tenants zero-diff); ArtPipeProcessor forwards position changes as Queued-state progress events and persists `queuePosition` on the stage run (SignalR *and* polling covered); the timeline now shows "Waiting for GPU — #N in line" live. 418/418 backend + 128/128 lib tests. Live two-submission race noted for the next full-stack session (unit tests pin ordering semantics). Closing.

spikerj closed this issue

2026-07-04 23:12:04 +00:00

spikerj referenced this issue

2026-07-04 23:22:04 +00:00

[Epic] ProArt 3D Art Studio — student asset pipeline (concept → model → texture → rig → animate → game-ready GLB) #346

spikerj referenced this issue

2026-07-09 01:40:06 +00:00

backend: expand integration coverage — art pipe messaging, safety RPC, GPU leases, quiz path #465

spikerj referenced this issue

2026-07-17 16:44:35 +00:00

fix(data): cast null to nullable JsonSerializerOptions in EF value converters (CS8600, 66 findings) #666

spikerj referenced this issue