Restriction

Last Updated: Oct 31, 2017

  • The number of resources referenced by one job cannot exceed 256. A table or an archive is considered as one unit.
  • The total number of bytes of resources referenced by one job cannot exceed 512 MB.
  • The number of inputs of one job cannot exceed 1,024. (The number of input tables cannot exceed 64). The number of outputs of one job cannot exceed 256.
  • Labels specified for multiple outputs cannot be null or empty strings. The label length cannot exceed 256, and the label can contain only upper-case letters, lower-case letters, digits, underlines (_), pound signs (#), periods (.), and hyphens (-).
  • The number of custom counters of one job cannot exceed 64. The group name and counter name cannot contain pound signs (#), and the sum of their lengths cannot exceed 100.
  • 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].
  • The number of times for repeatedly reading a resource by one Worker cannot exceed 64.
  • The split size is 64 MB by default, which can be set. 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 details about the restrictions, see Java sandbox.
Thank you! We've received your feedback.