fix: remove artificial model group concurrency cap

This commit is contained in:
wlx committed 2026-09-29 12:46:10 +08:00
1 parent 25114ca6b8
commit ce0a718686
13 files changed
+165 -34

No files matched your search

+6 -1
View File
@@ -46,8 +46,13 @@ services:
TRIPO_API_KEY: ${TRIPO_API_KEY:-}
GATEWAY_REQUESTS_PER_MINUTE: ${GATEWAY_REQUESTS_PER_MINUTE:-60}
GATEWAY_READS_PER_MINUTE: ${GATEWAY_READS_PER_MINUTE:-300}
GATEWAY_DEFAULT_MODEL_CONCURRENCY: ${GATEWAY_DEFAULT_MODEL_CONCURRENCY:-4}
# 0 means no MOYI-side model-group cap; the upstream provider remains
# responsible for its own concurrency and rate limits.
GATEWAY_DEFAULT_MODEL_CONCURRENCY: ${GATEWAY_DEFAULT_MODEL_CONCURRENCY:-0}
GATEWAY_DEFAULT_KEY_CONCURRENCY: ${GATEWAY_DEFAULT_KEY_CONCURRENCY:-4}
# Set true for an upstream concurrency measurement run. This bypasses
# all MOYI model/group leases while retaining per-Key protection.
GATEWAY_DISABLE_MODEL_CONCURRENCY: ${GATEWAY_DISABLE_MODEL_CONCURRENCY:-false}
GATEWAY_QUEUE_WAIT_MS: ${GATEWAY_QUEUE_WAIT_MS:-30000}
GATEWAY_QUEUE_POLL_MS: ${GATEWAY_QUEUE_POLL_MS:-100}
TASK_MAX_QUEUED_PER_MODEL: ${TASK_MAX_QUEUED_PER_MODEL:-100}