Release: stable util-band autoscaler + live GPU meters #162

2026-06-30T19:16:25-04:00

bvandeusen commented

2026-06-30 19:16:25 -04:00

Agent-only. Fixes two operator-reported issues:

Worker count flopping every cycle (GPU spikes). Replaces the throughput hill-climb (which probed +1 / reverted −1 every tick on noisy windows) with a GPU-utilization-band controller that HOLDS while util is in a healthy band, growing only on clear spare capacity and shrinking under saturation / memory pressure. EWMA-smoothed util + spaced decisions → steady load.
GPU util/VRAM bars only updating on manual refresh. Adds a dedicated /gpu endpoint (local nvidia-smi, no curator round-trip) polled every 1.5s, and drops the curator queue-status timeout 15s→5s so /status stays snappy.

Ships in the agent image.

🤖 Generated with Claude Code

Agent-only. Fixes two operator-reported issues: - **Worker count flopping every cycle (GPU spikes).** Replaces the throughput hill-climb (which probed +1 / reverted −1 every tick on noisy windows) with a GPU-utilization-band controller that HOLDS while util is in a healthy band, growing only on clear spare capacity and shrinking under saturation / memory pressure. EWMA-smoothed util + spaced decisions → steady load. - **GPU util/VRAM bars only updating on manual refresh.** Adds a dedicated `/gpu` endpoint (local nvidia-smi, no curator round-trip) polled every 1.5s, and drops the curator queue-status timeout 15s→5s so `/status` stays snappy. Ships in the agent image. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

bvandeusen added 1 commit 2026-06-30 19:16:25 -04:00

fix(agent): stable util-band autoscaler + live GPU meters

CI / lint (push) Successful in 4s

Details

CI / frontend-build (push) Successful in 19s

Details

CI / backend-lint-and-test (push) Successful in 27s

Details

CI / integration (push) Successful in 3m40s

Details

3b34230fbd

Two operator-reported issues with the GPU agent:

1. Worker count flopped almost every cycle, spiking the GPU. The hill-climb
   probed +1, judged it over a too-short noisy throughput window, saw no clear
   gain and reverted -1 — every tick. Replace it with a GPU-utilization-band
   controller: HOLD while smoothed util sits in a healthy band, grow only on
   clear spare capacity (util below the low mark + VRAM headroom), shrink under
   saturation or memory pressure. Util is EWMA-smoothed and decisions are spaced
   (DECIDE_EVERY samples), so a noisy nvidia-smi reading can't move the pool.
   Load stays consistent instead of probe/reverting.

2. GPU util/VRAM bars only updated on manual refresh. They rode the /status
   poll, which blocks on the curator queue call (slow when curator is busy), so
   the meters froze between refreshes. Give them a dedicated /gpu endpoint
   (local nvidia-smi only, no curator round-trip) polled every 1.5s, and drop
   the curator queue-status timeout 15s -> 5s so /status itself stays snappy.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa

bvandeusen merged commit 3c1e76bd44 into main

2026-06-30 19:16:29 -04:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: bvandeusen/FabledCurator#162