The limits of MaxCompute Graph are as follows:
  • Each job can reference up to 256 resources. A table or an archive is considered as one unit (that is, one resource).
  • The total number of bytes of resources referenced by one job cannot exceed 512 MB. Each job can reference up to 512 MB of bytes of resources.
  • The number of inputs of one job cannot exceed 1,024. and the number of input tables cannot exceed 64. The number of outputs of one job cannot exceed 256.
  • A label can be up to 256 characters in length and can contain letters, numbers, and special characters including underscores (_), pound signs (#), periods (.), and hyphens (-). Labels specified for multiple outputs cannot be null or empty strings.
  • Each job can have up to 64 custom counters. The group name and counter name can be up to 100 characters in length. The names cannot contain pound signs (#).
  • The number of Workers of one job is calculated by the framework. The maximum number is 1,000. If this threshold value is exceeded, an exception is thrown.
  • One Worker occupies 200 resources of the CPU by default. The range is [50, 800].
  • One Worker occupies 4096 MB of the memory by default. The range is [256 MB, 12 GB].
  • A threshold for a Worker to read a resource repeatedly is 64.
  • The split size can be set, however,as64 MB is the by default size.. The range is 0 < split_size <= (9223372036854775807 >>  20).
  • In the MaxCompute Graph program, GraphLoader/Vertex/Aggregator running in a cluster is restricted by the Java sandbox. (The main program of Graph jobs is not restricted.)  For more information about the restrictions, see  Java sandbox.