Merge pull request #15 from NVIDIA/test-model

Topograph model import for use with toposim

README.md

@@ -99,16 +99,32 @@ systemctl disable topograph.service
 systemctl daemon-reload
 ```
 
-#### Testing the Service
-To verify the service is running correctly, you can use the following commands:
+#### Verifying Health
+To verify the service is healthy, you can use the following command:
 
 ```bash
 curl http://localhost:49021/healthz
+```
+
+#### Using Toposim
+To test the service on a simulated cluster, first add the following line to `/etc/topograph/topograph-config.yaml` so that any topology requests are forwarded to toposim.
+```bash
+forward_service_url: dns:localhost:49025
+```
+Then run the topograph service as normal.
 
-id=$(curl -s -X POST -H "Content-Type: application/json" -d '{"provider":{"name":"test"},"engine":{"name":"test"}}' http://localhost:49021/v1/generate)
+You must then start the toposim service as such, setting the path to the test model that you want to use in simulation:
+```bash
+/usr/local/bin/topograph -m /usr/local/bin/tests/models/<cluster-model>.yaml
+```
+
+You can then verify the topology results via simulation by querying topograph using the `test` provider and engine, and specifying the test model path as a parameter to the provider, as such:
+```bash
+id=$(curl -s -X POST -H "Content-Type: application/json" -d '{"provider":{"name":"test", "params":{"model_path":"/usr/local/bin/topograph/tests/models/<cluster-model>.yaml"}},"engine":{"name":"test"}}' http://localhost:49021/v1/generate)
 
 curl -s "http://localhost:49021/v1/topology?uid=$id"
 ```
+Note the path specified in the topograph query should point to the same model as provided to toposim. 
 
 #### Using the Cluster Topology Generator
 

cmd/toposim/main.go

@@ -26,6 +26,7 @@ import (
 	"github.com/oklog/run"
 	"k8s.io/klog/v2"
 
+	"github.com/NVIDIA/topograph/pkg/models"
 	"github.com/NVIDIA/topograph/pkg/toposim"
 )
 
@@ -50,7 +51,7 @@ func mainInternal() error {
 		return fmt.Errorf("must specify topology model path and listening port")
 	}
 
-	model, err := toposim.NewModelFromFile(path)
+	model, err := models.NewModelFromFile(path)
 	if err != nil {
 		return err
 	}

pkg/common/const.go

@@ -35,6 +35,7 @@ const (
 	KeyTopoConfigmapNamespace = "topology_configmap_namespace"
 	KeyBlockSizes             = "block_sizes"
 	KeySkipReload             = "skip_reload"
+	KeyModelPath              = "model_path"
 
 	KeyPlugin        = "plugin"
 	ValTopologyTree  = "topology/tree"

pkg/common/types.go

@@ -76,13 +76,14 @@ type TopologyRequest struct {
 }
 
 type provider struct {
-	Name  string            `json:"name"`
-	Creds map[string]string `json:"creds"` // access credentials
+	Name   string            `json:"name"`
+	Creds  map[string]string `json:"creds"` // access credentials
+	Params map[string]string `json:"params"`
 }
 
 type engine struct {
 	Name   string            `json:"name"`
-	Params map[string]string `json:"params"` // access credentials
+	Params map[string]string `json:"params"`
 }
 
 type ComputeInstances struct {
@@ -108,6 +109,7 @@ func (p *TopologyRequest) String() string {
 	sb.WriteString("TopologyRequest:\n")
 	sb.WriteString(fmt.Sprintf("  Provider: %s\n", p.Provider.Name))
 	sb.WriteString(map2string(p.Provider.Creds, "  Credentials", true, "\n"))
+	sb.WriteString(map2string(p.Provider.Params, "  Parameters", false, "\n"))
 	sb.WriteString(fmt.Sprintf("  Engine: %s\n", p.Engine.Name))
 	sb.WriteString(map2string(p.Engine.Params, "  Parameters", false, "\n"))
 	sb.WriteString("  Nodes: ")

pkg/common/types_test.go

@@ -36,6 +36,7 @@ func TestPayload(t *testing.T) {
 			print: `TopologyRequest:
   Provider: 
   Credentials: []
+  Parameters: []
   Engine: 
   Parameters: []
   Nodes: 
@@ -58,7 +59,8 @@ func TestPayload(t *testing.T) {
 	"creds": {
       "access_key_id": "id",
       "secret_access_key": "secret"
-    }
+    },
+	"params": {}
   },
   "engine": {
     "name": "slurm",
@@ -94,6 +96,7 @@ func TestPayload(t *testing.T) {
 						"access_key_id":     "id",
 						"secret_access_key": "secret",
 					},
+					Params: map[string]string{},
 				},
 				Engine: engine{
 					Name: "slurm",
@@ -124,6 +127,7 @@ func TestPayload(t *testing.T) {
 			print: `TopologyRequest:
   Provider: aws
   Credentials: [access_key_id:*** secret_access_key:***]
+  Parameters: []
   Engine: slurm
   Parameters: [block_sizes:30,120 plugin:topology/block]
   Nodes: region1: [instance1:node1 instance2:node2 instance3:node3] region2: [instance4:node4 instance5:node5 instance6:node6] 

pkg/factory/provider.go

@@ -22,6 +22,7 @@ import (
 	"net/http"
 
 	"github.com/NVIDIA/topograph/pkg/common"
+	"github.com/NVIDIA/topograph/pkg/models"
 	"github.com/NVIDIA/topograph/pkg/providers/aws"
 	"github.com/NVIDIA/topograph/pkg/providers/baremetal"
 	"github.com/NVIDIA/topograph/pkg/providers/cw"
@@ -30,7 +31,7 @@ import (
 	"github.com/NVIDIA/topograph/pkg/translate"
 )
 
-func GetProvider(provider string) (common.Provider, *common.HTTPError) {
+func GetProvider(provider string, params map[string]string) (common.Provider, *common.HTTPError) {
 	var (
 		prv common.Provider
 		err error
@@ -48,7 +49,7 @@ func GetProvider(provider string) (common.Provider, *common.HTTPError) {
 	case common.ProviderBM:
 		prv, err = baremetal.GetProvider()
 	case common.ProviderTest:
-		prv = GetTestProvider()
+		prv, err = GetTestProvider(params)
 	default:
 		return nil, common.NewHTTPError(http.StatusBadRequest, fmt.Sprintf("unsupported provider %q", provider))
 	}
@@ -65,11 +66,25 @@ type testProvider struct {
 	instance2node map[string]string
 }
 
-func GetTestProvider() *testProvider {
+func GetTestProvider(params map[string]string) (*testProvider, error) {
 	p := &testProvider{}
-	p.tree, p.instance2node = translate.GetTreeTestSet(false)
 
-	return p
+	var modelPath string
+	if len(params) != 0 {
+		modelPath = params[common.KeyModelPath]
+	}
+
+	if len(modelPath) == 0 {
+		p.tree, p.instance2node = translate.GetTreeTestSet(false)
+	} else {
+		model, err := models.NewModelFromFile(modelPath)
+		if err != nil {
+			return nil, err
+		}
+		p.tree, p.instance2node = model.ToTree()
+	}
+
+	return p, nil
 }
 
 func (p *testProvider) GetCredentials(_ map[string]string) (interface{}, error) {

pkg/toposim/model.go → pkg/models/model.go

@@ -14,13 +14,15 @@
  * limitations under the License.
  */
 
-package toposim
+package models
 
 import (
 	"fmt"
 	"os"
 
 	"gopkg.in/yaml.v3"
+
+	"github.com/NVIDIA/topograph/pkg/common"
 )
 
 type Model struct {
@@ -137,3 +139,49 @@ func getNetworkLayers(name string, swmap map[string]string) ([]string, error) {
 		name = parent
 	}
 }
+
+func (model *Model) ToTree() (*common.Vertex, map[string]string) {
+	instance2node := make(map[string]string)
+	nodeVertexMap := make(map[string]*common.Vertex)
+	swVertexMap := make(map[string]*common.Vertex)
+	swRootMap := make(map[string]bool)
+
+	// Create all the vertices for each node
+	for k, v := range model.Nodes {
+		instance2node[k] = k
+		nodeVertexMap[k] = &common.Vertex{ID: v.Name, Name: v.Name}
+	}
+
+	// Initialize all the vertices for each switch (setting each on to be a possible root)
+	for _, sw := range model.Switches {
+		swVertexMap[sw.Name] = &common.Vertex{ID: sw.Name, Vertices: make(map[string]*common.Vertex)}
+		swRootMap[sw.Name] = true
+	}
+
+	// Connect all the switches to their sub-switches and sub-nodes
+	for _, sw := range model.Switches {
+		for _, subsw := range sw.Switches {
+			swRootMap[subsw] = false
+			swVertexMap[sw.Name].Vertices[subsw] = swVertexMap[subsw]
+		}
+		for _, cbname := range sw.CapacityBlocks {
+			for _, block := range model.CapacityBlocks {
+				if cbname == block.Name {
+					for _, node := range block.Nodes {
+						swVertexMap[sw.Name].Vertices[node] = nodeVertexMap[node]
+					}
+					break
+				}
+			}
+		}
+	}
+
+	// Connects all root vertices to the hidden root
+	root := &common.Vertex{Vertices: make(map[string]*common.Vertex)}
+	for k, v := range swRootMap {
+		if v {
+			root.Vertices[k] = swVertexMap[k]
+		}
+	}
+	return root, instance2node
+}

pkg/toposim/model_test.go → pkg/models/model_test.go

@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-package toposim
+package models
 
 import (
 	"testing"
@@ -23,7 +23,7 @@ import (
 )
 
 func TestNewModelFromFile(t *testing.T) {
-	cfg, err := NewModelFromFile("testdata/toposim.yaml")
+	cfg, err := NewModelFromFile("../../tests/models/medium-h100.yaml")
 	require.NoError(t, err)
 
 	expected := &Model{