fix: update default aux container limits and instance types (#8959)

determined-ai · Mar 6, 2024 · dcaa893 · dcaa893
1 parent c7e5d43
commit dcaa893
Show file tree

Hide file tree

Showing 8 changed files with 15 additions and 15 deletions.
diff --git a/harness/determined/deploy/aws/templates/efs.yaml b/harness/determined/deploy/aws/templates/efs.yaml
@@ -61,7 +61,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -111,7 +111,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to launch on agents in the default auxiliary resource pool.
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/aws/templates/fsx.yaml b/harness/determined/deploy/aws/templates/fsx.yaml
@@ -61,7 +61,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -111,7 +111,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to launch on agents in the default auxiliary resource pool.
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/aws/templates/govcloud.yaml b/harness/determined/deploy/aws/templates/govcloud.yaml
@@ -22,7 +22,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -77,7 +77,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to keep running on agents in the CPU resource pool
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/aws/templates/lore.yaml b/harness/determined/deploy/aws/templates/lore.yaml
@@ -61,7 +61,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -111,7 +111,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to launch on agents in the default auxiliary resource pool.
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/aws/templates/secure.yaml b/harness/determined/deploy/aws/templates/secure.yaml
@@ -77,7 +77,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -132,7 +132,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to launch on agents in the default auxiliary resource pool.
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/aws/templates/simple-rds.yaml b/harness/determined/deploy/aws/templates/simple-rds.yaml
@@ -48,7 +48,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -116,7 +116,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to launch on agents in the default auxiliary resource pool.
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/aws/templates/simple.yaml b/harness/determined/deploy/aws/templates/simple.yaml
@@ -48,7 +48,7 @@ Parameters:
   AuxAgentInstanceType:
     Type: String
     Description: Instance Type of agents in the auxiliary resource pool
-    Default: t2.xlarge
+    Default: m5.xlarge
 
   ComputeAgentInstanceType:
     Type: String
@@ -103,7 +103,7 @@ Parameters:
   MaxAuxContainersPerAgent:
     Type: Number
     Description: Maximum number of CPU containers to launch on agents in the default auxiliary resource pool.
-    Default: 100
+    Default: 16
 
   MaxIdleAgentPeriod:
     Type: String

diff --git a/harness/determined/deploy/gcp/constants.py b/harness/determined/deploy/gcp/constants.py
@@ -8,7 +8,7 @@ class defaults:
     GPU_NUM = 4
     GPU_TYPE = "nvidia-tesla-t4"
     MASTER_INSTANCE_TYPE = "n1-standard-2"
-    MAX_AUX_CONTAINERS_PER_AGENT = 100
+    MAX_AUX_CONTAINERS_PER_AGENT = 16
     MAX_IDLE_AGENT_PERIOD = "10m"
     MAX_AGENT_STARTING_PERIOD = "20m"
     OPERATION_TIMEOUT_PERIOD = "5m"