changelog / post
Mixed-mode rolling promotions
Rolling promotions now let you set max surge and max unavailable together. Control how many extra replicas Baseten can provision and how much serving capacity can temporarily decrease during the same rollout. This gives you more flexibility when spare GPU capacity is limited.
Configure both limits in your environment’s Promotion settings, through the Management API using:
1curl --request PATCH \
2 --url "https://api.baseten.co/v1/models/<MODEL_ID>/environments/production" \
3 --header "Authorization: Bearer $BASETEN_API_KEY" \
4 --header "Content-Type: application/json" \
5 --data '{
6 "promotion_settings": {
7 "rolling_deploy": true,
8 "rolling_deploy_config": {
9 "max_surge_percent": 25,
10 "max_unavailable_percent": 10
11 }
12 }
13 }'For more information, see our docs.