[AIR] [Release] Don't use head node for XGBoost Benchmark (ray-projec…

…t#29091) Signed-off-by: Amog Kamsetty [email protected] Seems like some recent changes on head node have caused performance regression for the xgboost benchmark. We change the compute config to only use worker nodes for compute instead. Closes ray-project#28974
rickyyx · Oct 6, 2022 · 8fd6a5b · 8fd6a5b
1 parent 184b405
commit 8fd6a5b
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 5 deletions.
diff --git a/release/air_tests/air_benchmarks/xgboost_compute_tpl.yaml b/release/air_tests/air_benchmarks/xgboost_compute_tpl.yaml
@@ -1,17 +1,20 @@
 cloud_id: {{env["ANYSCALE_CLOUD_ID"]}}
 region: us-west-2
 
-max_workers: 9
+max_workers: 10
 
 head_node_type:
  name: head_node
- instance_type: m5.4xlarge
+ instance_type: m5.2xlarge
+ resources:
+ cpu: 0
+
 
 worker_node_types:
  - name: worker_node
  instance_type: m5.4xlarge
- max_workers: 9
- min_workers: 9
+ max_workers: 10
+ min_workers: 10
  use_spot: false
 
 aws:

diff --git a/release/release_tests.yaml b/release/release_tests.yaml
@@ -196,7 +196,7 @@
  script: python workloads/xgboost_benchmark.py
 
  wait_for_nodes:
- num_nodes: 10
+ num_nodes: 11
 
  type: sdk_command
  file_manager: job