Remove code to parse env vars into kaniko arguments

api/turing/config/config.go

@@ -430,7 +430,12 @@ type MLPConfig struct {
 }
 
 type MlflowConfig struct {
-	TrackingURL         string `validate:"required_if=ArtifactServiceType gcs ArtifactServiceType s3"`
+	TrackingURL string `validate:"required_if=ArtifactServiceType gcs ArtifactServiceType s3"`
+	// Note that the Kaniko image builder needs to be configured correctly to have the necessary credentials to download
+	// the artifacts from the blob storage tool depending on the artifact service type selected (gcs/s3). For gcs, the
+	// credentials can be provided via a k8s service account or a secret but for s3, the credentials can be provided via
+	// additional arguments in the config KanikoAdditionalArgs e.g.
+	// --build-arg=[AWS_ACCESS_KEY_ID/AWS_SECRET_ACCESS_KEY/AWS_DEFAULT_REGION/AWS_ENDPOINT_URL]=xxx
 	ArtifactServiceType string `validate:"required,oneof=nop gcs s3"`
 }
 

api/turing/imagebuilder/imagebuilder.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"fmt"
 	"net/http"
-	"os"
 	"sort"
 	"strings"
 	"time"
@@ -298,16 +297,6 @@ func (ib *imageBuilder) createKanikoJob(
 		"--single-snapshot",
 	}
 
-	if ib.artifactServiceType == "s3" {
-		kanikoArgs = append(
-			kanikoArgs,
-			fmt.Sprintf("--build-arg=%s=%s", "AWS_ACCESS_KEY_ID", os.Getenv("AWS_ACCESS_KEY_ID")),
-			fmt.Sprintf("--build-arg=%s=%s", "AWS_SECRET_ACCESS_KEY", os.Getenv("AWS_SECRET_ACCESS_KEY")),
-			fmt.Sprintf("--build-arg=%s=%s", "AWS_DEFAULT_REGION", os.Getenv("AWS_DEFAULT_REGION")),
-			fmt.Sprintf("--build-arg=%s=%s", "AWS_ENDPOINT_URL", os.Getenv("AWS_ENDPOINT_URL")),
-		)
-	}
-
 	annotations := make(map[string]string)
 	if !ib.imageBuildingConfig.SafeToEvict {
 		// The image-building jobs are timing out. We found that one of the root causes is the node pool got scaled down