[smg] Work around sglang's notorious orphan process problem (#16756)

This commit is contained in:
Simo Lin
2026-01-08 14:41:03 -08:00
committed by GitHub
parent 55a8dd0095
commit ee71e773e1
2 changed files with 30 additions and 7 deletions
@@ -61,7 +61,9 @@ HEALTH_CHECK_INTERVAL = 2 # Check every 2s (was 5s)
# Model loading configuration
INITIAL_GRACE_PERIOD = 30 # Wait before first health check (model loading time)
LAUNCH_STAGGER_DELAY = 5 # Delay between launching multiple workers
LAUNCH_STAGGER_DELAY = (
10 # Delay between launching multiple workers (avoid I/O contention)
)
# Retry configuration
MAX_RETRY_ATTEMPTS = (