Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: declare the Hub's /api/engines retry in the log budget (#13682)

Daniel Hana413834cb1b7
diff --git a/studio/backend/loggers/handlers.py b/studio/backend/loggers/handlers.py
index dbd5a409455..5331f2fd3cc 100644
--- a/studio/backend/loggers/handlers.py
+++ b/studio/backend/loggers/handlers.py
@@ -121,6 +121,9 @@ _QUIET_SUCCESS_PATHS = {
"/api/inference/images/generate-progress",
"/api/inference/video/generate-progress",
"/api/llama/update-status",
+ # Engine status, re-read every 5s by the Hub while the vLLM GPU probe runs and by the open Settings engine list.
+ # The UI shows each answer, install jobs log their own progress, and a failed read (non-2xx) still logs.
+ "/api/engines",
"/api/export/logs",
"/api/export/status",
# The Settings > Logs viewer polls are suppressed in _SELF_READ_PATHS below.
diff --git a/studio/backend/tests/log_budget/session.py b/studio/backend/tests/log_budget/session.py
index e10e00a8415..0b9cc6b8567 100644
--- a/studio/backend/tests/log_budget/session.py
+++ b/studio/backend/tests/log_budget/session.py
@@ -54,6 +54,11 @@ IDLE_POLLS: dict[str, tuple[float, str]] = {
"/api/providers/registry": (10.0, "declared"),
"/api/settings/personalization": (10.0, "declared"),
"/api/system": (10.0, "declared"),
+ # The globally mounted download manager re-reads it every 30s in the background (useEngines(enabled, true)).
+ # Faster bursts of the same path: 3s while Settings > Inference engines is open, 1s during an install, and the
+ # Hub's 5s vLLM availability retry while the GPU probe still answers "Checking...". A path sits in one scenario
+ # only, and every 2xx read is dropped by _QUIET_SUCCESS_PATHS, so the faster cadences cost no lines either.
+ "/api/engines": (30.0, "declared"),
}
# Polls that only exist while an operation is in flight: a download running, a generation