databricks-cli/bundle/config/validate/single_node_cluster_test.go

567 lines
15 KiB
Go
Raw Normal View History

Add validation for single node clusters (#1909) ## Changes This PR adds a warning validating that the configuration for a single node cluster is valid for interactive, job, job-task, and pipeline clusters. Note: We skip the validation if a cluster policy is configured because the policy is likely to configure `spark_conf` / `custom_tags` itself. Note: Terrform originally only had validation for interactive, job, and job-task clusters. This PR adding the validation for pipeline clusters as well is new. This PR follows the same logic as we used to have in Terraform. The validation was removed from Terraform because we had no way to demote the error to a warning: https://github.com/databricks/terraform-provider-databricks/pull/4222 ### Background Single-node clusters require `spark_conf` and `custom_tags` to be correctly set in the cluster definition for them to function optimally. The cluster will be created even if incorrectly configured, but its performance will not be great. For example, if both `spark_conf` and `custom_tags` are not set and `num_workers` is 0, then only the driver process will be launched on the cluster compute instance thus leading to sub-optimal utilization of available compute resources and no parallelization across worker processes when processing a spark query. ### Issue This PR addresses some issues reported in https://github.com/databricks/cli/issues/1546 ## Tests Unit tests and manually. Example output of the warning: ``` ➜ bundle-playground git:(master) ✗ cli bundle validate Warning: Single node cluster is not correctly configured at resources.pipelines.bar.clusters[0] in databricks.yml:29:11 num_workers should be 0 only for single-node clusters. To create a valid single node cluster please ensure that the following properties are correctly set in the cluster specification: spark_conf: spark.databricks.cluster.profile: singleNode spark.master: local[*] custom_tags: ResourceClass: SingleNode Name: foobar Target: default Workspace: User: shreyas.goenka@databricks.com Path: /Workspace/Users/shreyas.goenka@databricks.com/.bundle/foobar/default Found 1 warning ```
2024-11-22 15:48:09 +00:00
package validate
import (
"context"
"testing"
"github.com/databricks/cli/bundle"
2024-11-29 01:20:41 +00:00
"github.com/databricks/cli/bundle/bundletest"
Add validation for single node clusters (#1909) ## Changes This PR adds a warning validating that the configuration for a single node cluster is valid for interactive, job, job-task, and pipeline clusters. Note: We skip the validation if a cluster policy is configured because the policy is likely to configure `spark_conf` / `custom_tags` itself. Note: Terrform originally only had validation for interactive, job, and job-task clusters. This PR adding the validation for pipeline clusters as well is new. This PR follows the same logic as we used to have in Terraform. The validation was removed from Terraform because we had no way to demote the error to a warning: https://github.com/databricks/terraform-provider-databricks/pull/4222 ### Background Single-node clusters require `spark_conf` and `custom_tags` to be correctly set in the cluster definition for them to function optimally. The cluster will be created even if incorrectly configured, but its performance will not be great. For example, if both `spark_conf` and `custom_tags` are not set and `num_workers` is 0, then only the driver process will be launched on the cluster compute instance thus leading to sub-optimal utilization of available compute resources and no parallelization across worker processes when processing a spark query. ### Issue This PR addresses some issues reported in https://github.com/databricks/cli/issues/1546 ## Tests Unit tests and manually. Example output of the warning: ``` ➜ bundle-playground git:(master) ✗ cli bundle validate Warning: Single node cluster is not correctly configured at resources.pipelines.bar.clusters[0] in databricks.yml:29:11 num_workers should be 0 only for single-node clusters. To create a valid single node cluster please ensure that the following properties are correctly set in the cluster specification: spark_conf: spark.databricks.cluster.profile: singleNode spark.master: local[*] custom_tags: ResourceClass: SingleNode Name: foobar Target: default Workspace: User: shreyas.goenka@databricks.com Path: /Workspace/Users/shreyas.goenka@databricks.com/.bundle/foobar/default Found 1 warning ```
2024-11-22 15:48:09 +00:00
"github.com/databricks/cli/bundle/config"
"github.com/databricks/cli/bundle/config/resources"
"github.com/databricks/cli/libs/diag"
"github.com/databricks/cli/libs/dyn"
"github.com/databricks/databricks-sdk-go/service/compute"
"github.com/databricks/databricks-sdk-go/service/jobs"
"github.com/databricks/databricks-sdk-go/service/pipelines"
"github.com/stretchr/testify/assert"
)
func failCases() []struct {
name string
sparkConf map[string]string
customTags map[string]string
} {
return []struct {
name string
sparkConf map[string]string
customTags map[string]string
}{
{
name: "no tags or conf",
},
{
name: "no tags",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "singleNode",
"spark.master": "local[*]",
},
},
{
name: "no conf",
customTags: map[string]string{"ResourceClass": "SingleNode"},
},
{
name: "invalid spark cluster profile",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "invalid",
"spark.master": "local[*]",
},
customTags: map[string]string{"ResourceClass": "SingleNode"},
},
{
name: "invalid spark.master",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "singleNode",
"spark.master": "invalid",
},
customTags: map[string]string{"ResourceClass": "SingleNode"},
},
{
name: "invalid tags",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "singleNode",
"spark.master": "local[*]",
},
customTags: map[string]string{"ResourceClass": "invalid"},
},
{
name: "missing ResourceClass tag",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "singleNode",
"spark.master": "local[*]",
},
customTags: map[string]string{"what": "ever"},
},
{
name: "missing spark.master",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "singleNode",
},
customTags: map[string]string{"ResourceClass": "SingleNode"},
},
{
name: "missing spark.databricks.cluster.profile",
sparkConf: map[string]string{
"spark.master": "local[*]",
},
customTags: map[string]string{"ResourceClass": "SingleNode"},
},
}
}
func TestValidateSingleNodeClusterFailForInteractiveClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range failCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Clusters: map[string]*resources.Cluster{
"foo": {
ClusterSpec: &compute.ClusterSpec{
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
},
},
},
},
},
}
bundletest.SetLocation(b, "resources.clusters.foo", []dyn.Location{{File: "a.yml", Line: 1, Column: 1}})
// We can't set num_workers to 0 explicitly in the typed configuration.
// Do it on the dyn.Value directly.
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.clusters.foo.num_workers", dyn.V(0))
})
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Equal(t, diag.Diagnostics{
{
Severity: diag.Warning,
Summary: singleNodeWarningSummary,
Detail: singleNodeWarningDetail,
Locations: []dyn.Location{{File: "a.yml", Line: 1, Column: 1}},
Paths: []dyn.Path{dyn.NewPath(dyn.Key("resources"), dyn.Key("clusters"), dyn.Key("foo"))},
},
}, diags)
})
}
}
func TestValidateSingleNodeClusterFailForJobClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range failCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Jobs: map[string]*resources.Job{
"foo": {
JobSettings: &jobs.JobSettings{
JobClusters: []jobs.JobCluster{
{
NewCluster: compute.ClusterSpec{
ClusterName: "my_cluster",
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
},
},
},
},
},
},
},
},
}
bundletest.SetLocation(b, "resources.jobs.foo.job_clusters[0].new_cluster", []dyn.Location{{File: "b.yml", Line: 1, Column: 1}})
// We can't set num_workers to 0 explicitly in the typed configuration.
// Do it on the dyn.Value directly.
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.jobs.foo.job_clusters[0].new_cluster.num_workers", dyn.V(0))
})
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Equal(t, diag.Diagnostics{
{
Severity: diag.Warning,
Summary: singleNodeWarningSummary,
Detail: singleNodeWarningDetail,
Locations: []dyn.Location{{File: "b.yml", Line: 1, Column: 1}},
Paths: []dyn.Path{dyn.MustPathFromString("resources.jobs.foo.job_clusters[0].new_cluster")},
},
}, diags)
})
}
}
func TestValidateSingleNodeClusterFailForJobTaskClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range failCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Jobs: map[string]*resources.Job{
"foo": {
JobSettings: &jobs.JobSettings{
Tasks: []jobs.Task{
{
NewCluster: &compute.ClusterSpec{
ClusterName: "my_cluster",
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
},
},
},
},
},
},
},
},
}
bundletest.SetLocation(b, "resources.jobs.foo.tasks[0].new_cluster", []dyn.Location{{File: "c.yml", Line: 1, Column: 1}})
// We can't set num_workers to 0 explicitly in the typed configuration.
// Do it on the dyn.Value directly.
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.jobs.foo.tasks[0].new_cluster.num_workers", dyn.V(0))
})
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Equal(t, diag.Diagnostics{
{
Severity: diag.Warning,
Summary: singleNodeWarningSummary,
Detail: singleNodeWarningDetail,
Locations: []dyn.Location{{File: "c.yml", Line: 1, Column: 1}},
Paths: []dyn.Path{dyn.MustPathFromString("resources.jobs.foo.tasks[0].new_cluster")},
},
}, diags)
})
}
}
func TestValidateSingleNodeClusterFailForPipelineClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range failCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Pipelines: map[string]*resources.Pipeline{
"foo": {
PipelineSpec: &pipelines.PipelineSpec{
Clusters: []pipelines.PipelineCluster{
{
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
},
},
},
},
},
},
},
}
bundletest.SetLocation(b, "resources.pipelines.foo.clusters[0]", []dyn.Location{{File: "d.yml", Line: 1, Column: 1}})
// We can't set num_workers to 0 explicitly in the typed configuration.
// Do it on the dyn.Value directly.
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.pipelines.foo.clusters[0].num_workers", dyn.V(0))
})
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Equal(t, diag.Diagnostics{
{
Severity: diag.Warning,
Summary: singleNodeWarningSummary,
Detail: singleNodeWarningDetail,
Locations: []dyn.Location{{File: "d.yml", Line: 1, Column: 1}},
Paths: []dyn.Path{dyn.MustPathFromString("resources.pipelines.foo.clusters[0]")},
},
}, diags)
})
}
}
func TestValidateSingleNodeClusterFailForJobForEachTaskCluster(t *testing.T) {
ctx := context.Background()
for _, tc := range failCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Jobs: map[string]*resources.Job{
"foo": {
JobSettings: &jobs.JobSettings{
Tasks: []jobs.Task{
{
ForEachTask: &jobs.ForEachTask{
Task: jobs.Task{
NewCluster: &compute.ClusterSpec{
ClusterName: "my_cluster",
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
},
},
},
},
},
},
},
},
},
},
}
bundletest.SetLocation(b, "resources.jobs.foo.tasks[0].for_each_task.task.new_cluster", []dyn.Location{{File: "e.yml", Line: 1, Column: 1}})
// We can't set num_workers to 0 explicitly in the typed configuration.
// Do it on the dyn.Value directly.
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.jobs.foo.tasks[0].for_each_task.task.new_cluster.num_workers", dyn.V(0))
})
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Equal(t, diag.Diagnostics{
{
Severity: diag.Warning,
Summary: singleNodeWarningSummary,
Detail: singleNodeWarningDetail,
Locations: []dyn.Location{{File: "e.yml", Line: 1, Column: 1}},
Paths: []dyn.Path{dyn.MustPathFromString("resources.jobs.foo.tasks[0].for_each_task.task.new_cluster")},
},
}, diags)
})
}
}
func passCases() []struct {
name string
numWorkers *int
sparkConf map[string]string
customTags map[string]string
policyId string
} {
zero := 0
one := 1
return []struct {
name string
numWorkers *int
sparkConf map[string]string
customTags map[string]string
policyId string
}{
{
name: "single node cluster",
sparkConf: map[string]string{
"spark.databricks.cluster.profile": "singleNode",
"spark.master": "local[*]",
},
customTags: map[string]string{
"ResourceClass": "SingleNode",
},
numWorkers: &zero,
},
{
name: "num workers is not zero",
numWorkers: &one,
},
{
name: "num workers is not set",
},
{
name: "policy id is not empty",
policyId: "policy-abc",
numWorkers: &zero,
},
}
}
func TestValidateSingleNodeClusterPassInteractiveClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range passCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Clusters: map[string]*resources.Cluster{
"foo": {
ClusterSpec: &compute.ClusterSpec{
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
PolicyId: tc.policyId,
},
},
},
},
},
}
if tc.numWorkers != nil {
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.clusters.foo.num_workers", dyn.V(*tc.numWorkers))
})
}
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Empty(t, diags)
})
}
}
func TestValidateSingleNodeClusterPassJobClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range passCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Jobs: map[string]*resources.Job{
"foo": {
JobSettings: &jobs.JobSettings{
JobClusters: []jobs.JobCluster{
{
NewCluster: compute.ClusterSpec{
ClusterName: "my_cluster",
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
PolicyId: tc.policyId,
},
},
},
},
},
},
},
},
}
if tc.numWorkers != nil {
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.jobs.foo.job_clusters[0].new_cluster.num_workers", dyn.V(*tc.numWorkers))
})
}
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Empty(t, diags)
})
}
}
func TestValidateSingleNodeClusterPassJobTaskClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range passCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Jobs: map[string]*resources.Job{
"foo": {
JobSettings: &jobs.JobSettings{
Tasks: []jobs.Task{
{
NewCluster: &compute.ClusterSpec{
ClusterName: "my_cluster",
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
PolicyId: tc.policyId,
},
},
},
},
},
},
},
},
}
if tc.numWorkers != nil {
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.jobs.foo.tasks[0].new_cluster.num_workers", dyn.V(*tc.numWorkers))
})
}
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Empty(t, diags)
})
}
}
func TestValidateSingleNodeClusterPassPipelineClusters(t *testing.T) {
ctx := context.Background()
for _, tc := range passCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Pipelines: map[string]*resources.Pipeline{
"foo": {
PipelineSpec: &pipelines.PipelineSpec{
Clusters: []pipelines.PipelineCluster{
{
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
PolicyId: tc.policyId,
},
},
},
},
},
},
},
}
if tc.numWorkers != nil {
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.pipelines.foo.clusters[0].num_workers", dyn.V(*tc.numWorkers))
})
}
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Empty(t, diags)
})
}
}
func TestValidateSingleNodeClusterPassJobForEachTaskCluster(t *testing.T) {
ctx := context.Background()
for _, tc := range passCases() {
t.Run(tc.name, func(t *testing.T) {
b := &bundle.Bundle{
Config: config.Root{
Resources: config.Resources{
Jobs: map[string]*resources.Job{
"foo": {
JobSettings: &jobs.JobSettings{
Tasks: []jobs.Task{
{
ForEachTask: &jobs.ForEachTask{
Task: jobs.Task{
NewCluster: &compute.ClusterSpec{
ClusterName: "my_cluster",
SparkConf: tc.sparkConf,
CustomTags: tc.customTags,
PolicyId: tc.policyId,
},
},
},
},
},
},
},
},
},
},
}
if tc.numWorkers != nil {
bundletest.Mutate(t, b, func(v dyn.Value) (dyn.Value, error) {
return dyn.Set(v, "resources.jobs.foo.tasks[0].for_each_task.task.new_cluster.num_workers", dyn.V(*tc.numWorkers))
})
}
diags := bundle.ApplyReadOnly(ctx, bundle.ReadOnly(b), SingleNodeCluster())
assert.Empty(t, diags)
})
}
}