LCORE-1724: try spot first, fall back to on-demand with region rotation

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: are-ces

.tekton/integration-tests/pipeline/lightspeed-stack-rhelai-test.yaml

@@ -118,65 +118,59 @@ spec:
               echo "[mapt] Spot: $(params.spot)"
               echo "[mapt] =================================="
 
-              SPOT_ARGS=""
-              if [[ "$(params.spot)" == "true" ]]; then
-                SPOT_ARGS="--spot --spot-eviction-tolerance highest"
-              fi
-
-              if [[ "$(params.spot)" == "true" ]]; then
-                export AWS_DEFAULT_REGION="us-east-1"
-                echo "[mapt] Using spot instances (searching all regions)..."
-                mapt aws rhel-ai create \
-                    --project-name "mapt-rhel-ai-${RUN_ID}" \
-                    --backed-url "s3://${BUCKET}/mapt/rhel-ai/${RUN_ID}" \
+              MAPT_COMMON_ARGS="--project-name mapt-rhel-ai-${RUN_ID} \
+                    --backed-url s3://${BUCKET}/mapt/rhel-ai/${RUN_ID} \
                     --conn-details-output /opt/host-info \
-                    --compute-sizes "$(params.instance-type)" \
-                    --version "$(params.rhelai-version)" \
-                    ${SPOT_ARGS} \
+                    --compute-sizes $(params.instance-type) \
+                    --version $(params.rhelai-version) \
                     --auto-start \
-                    --model "$(params.model)" \
-                    --hf-token "${HF_TOKEN}" \
-                    --api-key "${VLLM_API_KEY}" \
+                    --model $(params.model) \
+                    --hf-token ${HF_TOKEN} \
+                    --api-key ${VLLM_API_KEY} \
                     --expose-ports 8000 \
-                    --vllm-extra-args "--max-model-len 131072 --enable-auto-tool-choice --tool-call-parser llama3_json --chat-template /opt/app-root/template/tool_chat_template_llama3.1_json.jinja" \
-                    --tags "project=lightspeed-core,environment=konflux-ci"
+                    --vllm-extra-args '--max-model-len 131072 --enable-auto-tool-choice --tool-call-parser llama3_json --chat-template /opt/app-root/template/tool_chat_template_llama3.1_json.jinja' \
+                    --tags project=lightspeed-core,environment=konflux-ci"
+
+              CREATED=0
+
+              # Try spot first
+              echo "[mapt] Trying spot instances..."
+              export AWS_DEFAULT_REGION="us-east-1"
+              if mapt aws rhel-ai create $MAPT_COMMON_ARGS --spot --spot-eviction-tolerance highest; then
+                CREATED=1
               else
+                echo "[mapt] Spot failed, cleaning up..."
+                mapt aws rhel-ai destroy \
+                    --project-name "mapt-rhel-ai-${RUN_ID}" \
+                    --backed-url "s3://${BUCKET}/mapt/rhel-ai/${RUN_ID}" \
+                    --force-destroy 2>/dev/null || true
+              fi
+
+              # Fall back to on-demand with region rotation
+              if [ "$CREATED" -ne 1 ]; then
                 REGIONS="us-east-1 us-east-2 us-west-2 eu-west-1 eu-central-1 ap-northeast-1"
                 TIMEOUT=600
-                CREATED=0
 
                 for REGION in $REGIONS; do
                   echo "[mapt] Trying on-demand in ${REGION}..."
                   export AWS_DEFAULT_REGION="$REGION"
 
-                  if timeout $TIMEOUT mapt aws rhel-ai create \
-                      --project-name "mapt-rhel-ai-${RUN_ID}" \
-                      --backed-url "s3://${BUCKET}/mapt/rhel-ai/${RUN_ID}" \
-                      --conn-details-output /opt/host-info \
-                      --compute-sizes "$(params.instance-type)" \
-                      --version "$(params.rhelai-version)" \
-                      --auto-start \
-                      --model "$(params.model)" \
-                      --hf-token "${HF_TOKEN}" \
-                      --api-key "${VLLM_API_KEY}" \
-                      --expose-ports 8000 \
-                      --vllm-extra-args "--max-model-len 131072 --enable-auto-tool-choice --tool-call-parser llama3_json --chat-template /opt/app-root/template/tool_chat_template_llama3.1_json.jinja" \
-                      --tags "project=lightspeed-core,environment=konflux-ci"; then
+                  if timeout $TIMEOUT mapt aws rhel-ai create $MAPT_COMMON_ARGS; then
                     CREATED=1
                     break
                   fi
 
-                  echo "[mapt] Failed in ${REGION}, cleaning up and trying next..."
+                  echo "[mapt] Failed in ${REGION}, cleaning up..."
                   mapt aws rhel-ai destroy \
                       --project-name "mapt-rhel-ai-${RUN_ID}" \
                       --backed-url "s3://${BUCKET}/mapt/rhel-ai/${RUN_ID}" \
                       --force-destroy 2>/dev/null || true
                 done
+              fi
 
-                if [ "$CREATED" -ne 1 ]; then
-                  echo "[mapt] ERROR: Failed to create instance in any region"
-                  exit 1
-                fi
+              if [ "$CREATED" -ne 1 ]; then
+                echo "[mapt] ERROR: Failed to create instance (spot + on-demand)"
+                exit 1
               fi
 
               echo "[mapt] Instance created and vLLM started."