All Products
Search
Document Center

MaxCompute:Limits

Last Updated:Mar 26, 2026

MaxCompute Graph enforces hard limits on jobs, workers, resources, and components. Review these limits before submitting a job to avoid runtime errors.

Job limits

Category Limit Details
Resources per job 256 Each table and each archive counts as one unit.
Total resource size per job 512 MB Combined size of all referenced resources.
Inputs per job 1,024
Input tables per job 64
Outputs per job 256
Custom counters per job 64

Worker limits

Category Default Range
Workers per job Calculated by the framework Up to 1,000. An exception is thrown if the job requires more.
CPU per worker 200 units 50–800 units
Memory per worker 4,096 MB 256 MB–12 GB
Resource reads per worker Up to 64 times per resource

Output label constraints

Labels used for multiple outputs must meet the following requirements:

  • Cannot be null or an empty string.

  • Maximum length: 256 characters.

  • Allowed characters: letters, digits, underscores (_), number signs (#), periods (.), and hyphens (-).

Counter constraints

Counter quantity

The maximum number of custom counters per job is 64.

Counter naming rules

Constraint Rule
Counter group name Cannot contain #
Counter name Cannot contain #
Combined length of group name and counter name Must not exceed 100 characters

split_size constraints

Property Value
Default 64 MB
Minimum Greater than 0
Maximum 8,796,093,022,207 MB (equivalent to 9223372036854775807>>20)

Java sandbox scope

GraphLoader, Vertex, and Aggregator run under the Java sandbox when executed in a cluster. The main program of a Graph job is not subject to the Java sandbox.

Component Subject to Java sandbox
GraphLoader Yes
Vertex Yes
Aggregator Yes
Main program No
For more information about the Java sandbox, see Java Sandbox that run in distributed mode are limited by Java sandboxes. The main function of MapReduce is not subject to the limits of Java sandboxes.").