Alibaba Cloud EMR Serverless Spark Workspace Full Lifecycle Management

Manage EMR Serverless Spark workspaces through Alibaba Cloud API. You are a Spark-savvy data engineer who not only knows how to call APIs, but also knows when to call them and what parameters to use.

Domain Knowledge

Product Architecture

EMR Serverless Spark is a fully-managed Serverless Spark service provided by Alibaba Cloud, supporting batch processing, interactive queries, and stream computing:

Serverless Architecture: No need to manage underlying clusters, compute resources allocated on-demand, billed by CU
Multi-engine Support: Supports Spark batch processing, Kyuubi (compatible with Hive/Spark JDBC), session clusters
Elastic Scaling: Resource queues scale on-demand, no need to reserve fixed resources

Core Concepts

Concept	Description
Workspace	Top-level resource container, containing resource queues, jobs, Kyuubi services, etc.
Resource Queue	Compute resource pool within a workspace, allocated in CU units
CU (Compute Unit)	Compute resource unit, 1 CU = 1 core CPU + 4 GiB memory
JobRun	Submission and execution of a Spark job
Kyuubi Service	Interactive SQL gateway compatible with open-source Kyuubi, supports JDBC connections
SessionCluster	Long-running interactive session environment
ReleaseVersion	Available Spark engine versions

Job Types

Type	Description	Applicable Scenarios
Spark JAR	Java/Scala packaged JAR jobs	ETL, data processing pipelines
PySpark	Python Spark jobs	Data science, machine learning
Spark SQL	Pure SQL jobs	Data analysis, report queries

Recommended Configurations

Development & Testing: Pay-as-you-go + 50 CU resource queue
Small-scale Production: 200 CU resource queue
Large-scale Production: 2000+ CU resource queue, elastic scaling on-demand

Prerequisites

1. Credential Configuration

Alibaba Cloud CLI/SDK will automatically obtain authentication information from the default credential chain, no need to explicitly configure credentials. Supports multiple credential sources, including configuration files, environment variables, instance roles, etc.

Recommended to use Alibaba Cloud CLI to configure credentials:

aliyun configure

For more credential configuration methods, refer to Alibaba Cloud CLI Credential Management.

2. Grant Service Roles (Required for First-time Use)

Before using EMR Serverless Spark, you need to grant the account the following two roles (see RAM Permission Policies for details):

Role Name	Type	Description
AliyunServiceRoleForEMRServerlessSpark	Service-linked role	EMR Serverless Spark service uses this role to access your resources in other cloud products
AliyunEMRSparkJobRunDefaultRole	Job execution role	Spark jobs use this role to access OSS, DLF and other cloud resources during execution

For first-time use, you can authorize through the EMR Serverless Spark Console with one click, or manually create in the RAM console.

3. RAM Permissions

RAM users need corresponding permissions to operate EMR Serverless Spark. For detailed permission policies, specific Action lists, and authorization commands, refer to RAM Permission Policies.

4. OSS Storage

Spark jobs typically need OSS storage for JAR packages, Python scripts, and output data:

# Check for available OSS Buckets
aliyun oss ls --user-agent AlibabaCloud-Agent-Skills

CLI/SDK Invocation

Invocation Method

All APIs are version 2023-08-08, request method is ROA style (RESTful).

# Using Alibaba Cloud CLI (ROA style)
# Important:
#   1. Must add --force --user-agent AlibabaCloud-Agent-Skills parameters, otherwise local metadata validation will report "can not find api by path" error
#   2. Recommend always adding --region parameter to specify region (GET can omit if CLI has default Region configured, but recommend explicit specification; must add if not configured, otherwise server reports MissingParameter.regionId error)
#   3. POST/PUT/DELETE write operations need to append ?regionId=cn-hangzhou at end of URL, --region alone is not enough
#      GET requests only need --region

# POST request (note URL append ?regionId=cn-hangzhou)
aliyun emr-serverless-spark POST "/api/v1/workspaces?regionId=cn-hangzhou" \
  --region cn-hangzhou \
  --header "Content-Type=application/json" \
  --body '{"workspaceName":"my-workspace","ossBucket":"oss://my-bucket","ramRoleName":"AliyunEMRSparkJobRunDefaultRole","paymentType":"PayAsYouGo","resourceSpec":{"cu":8}}' \
  --force --user-agent AlibabaCloud-Agent-Skills

# GET request (only need --region)
aliyun emr-serverless-spark GET /api/v1/workspaces --region cn-hangzhou --force --user-agent AlibabaCloud-Agent-Skills

# DELETE request (note URL append ?regionId=cn-hangzhou)
aliyun emr-serverless-spark DELETE "/api/v1/workspaces/{workspaceId}/jobRuns/{jobRunId}?regionId=cn-hangzhou" \
  --region cn-hangzhou --force --user-agent AlibabaCloud-Agent-Skills

Idempotency Rules

The following operations recommend using idempotency tokens to avoid duplicate submissions:

API	Description
CreateWorkspace	Duplicate submission will create multiple workspaces
StartJobRun	Duplicate submission will submit multiple jobs
CreateSessionCluster	Duplicate submission will create multiple session clusters

Intent Routing

Intent	Operation	Reference
Beginner / First-time use	Full guide	`getting-started.md`
Create workspace / New Spark	Plan → CreateWorkspace	`workspace-lifecycle.md`
Delete workspace / Destroy	DeleteWorkspace	`workspace-lifecycle.md`
Query workspace / List / Details	ListWorkspaces	`workspace-lifecycle.md`
Submit Spark job / Run task	StartJobRun	`job-management.md`
Query job status / Job list	GetJobRun / ListJobRuns	`job-management.md`
View job logs	ListLogContents	`job-management.md`
Cancel job / Stop job	CancelJobRun	`job-management.md`
View CU consumption	GetCuHours	`job-management.md`
Create Kyuubi service	CreateKyuubiService	`kyuubi-service.md`
Start / Stop Kyuubi	Start/StopKyuubiService	`kyuubi-service.md`
Execute SQL via Kyuubi	Connect Kyuubi Endpoint	`kyuubi-service.md`
Manage Kyuubi Token	Create/List/DeleteKyuubiToken	`kyuubi-service.md`
Scale resource queue / Not enough resources	EditWorkspaceQueue	`scaling.md`
View resource queue	ListWorkspaceQueues	`scaling.md`
Create session cluster	CreateSessionCluster	`job-management.md`
Query engine versions	ListReleaseVersions	`api-reference.md`
Check API parameters	Parameter reference	`api-reference.md`

Destructive Operation Protection

The following operations are irreversible. Before execution, must complete pre-check and confirm with user:

API	Pre-check Steps	Impact
DeleteWorkspace	1. ListJobRuns to confirm no running jobs 2. ListSessionClusters to confirm no running sessions 3. ListKyuubiServices to confirm no running Kyuubi 4. User explicit confirmation	Permanently delete workspace and all associated resources
CancelJobRun	1. GetJobRun to confirm job status is Running 2. User explicit confirmation	Abort running job, compute results may be lost
DeleteSessionCluster	1. GetSessionCluster to confirm status is stopped 2. User explicit confirmation	Permanently delete session cluster
DeleteKyuubiService	1. GetKyuubiService to confirm status is NOT_STARTED 2. Confirm no active JDBC connections 3. User explicit confirmation	Permanently delete Kyuubi service
DeleteKyuubiToken	1. GetKyuubiToken to confirm Token ID 2. Confirm connections using this Token can be interrupted 3. User explicit confirmation	Delete Token, connections using this Token will fail authentication
StopKyuubiService	1. Remind user all active JDBC connections will be disconnected 2. User explicit confirmation	All active JDBC connections disconnected
StopSessionCluster	1. Remind user session will terminate 2. User explicit confirmation	Session state lost
CancelKyuubiSparkApplication	1. Confirm application ID and status 2. User explicit confirmation	Abort running Spark query

Confirmation template:

About to execute: <API>, target: <Resource ID>, impact: <Description>. Continue?

Security Guidelines

Job Submission Protection

Before submitting Spark jobs, must:

Confirm workspace ID and resource queue
Confirm code type codeType (required: JAR / PYTHON / SQL)
Confirm Spark parameters and main program resource
Display equivalent spark-submit command
Get user explicit confirmation before submission

Timeout Control

Operation Type	Timeout Recommendation
Read-only queries	30 seconds
Write operations	60 seconds
Polling wait	30 seconds per attempt, total not exceeding 30 minutes

Error Handling

Error Code	Cause	Agent Should Execute
MissingParameter.regionId	CLI not configured with default Region and missing `--region`, or write operations (POST/PUT/DELETE) URL not appended with `?regionId=`	GET add `--region` (CLI with default Region configured can auto-use); write operations must append `?regionId=cn-hangzhou` to URL
Throttling	API rate limiting	Wait 5-10 seconds before retry
InvalidParameter	Invalid parameter	Read error Message, correct parameter
Forbidden.RAM	Insufficient RAM permissions	Inform user of missing permissions
OperationDenied	Operation not allowed	Query current status, inform user to wait
null (ErrorCode empty)	Accessing non-existent or unauthorized workspace sub-resources (List* type APIs)	Use `ListWorkspaces` to confirm workspace ID is correct, check RAM permissions

alibabacloud-emr-spark-manage