how to specify job specific configuration

miniwdl-ext / miniwdl-aws

Extends miniwdl to run workflows on AWS Batch & EFS

MIT License

21 stars 8 forks source link

how to specify job specific configuration #7

Open nh13 opened 2 years ago

nh13 commented 2 years ago

I am looking to set batch job specific information prior to submission, for example:

job definition (to allow for custom cloud init if needed)
job queue
instance types (may be achieved through a queue) for tasks that need GPU or FPGA
disks to mount, and/or root/attached storage size

jdidion commented 2 years ago

Came here to ask for this same feature, specifically for instance types.

The WDL spec suggests that backend/specific configuration should be specified as follows:

WDL 1.x

runtime {
  aws: {
    instanceType: "c5.16xlarge"
  }
}

WDL development

hints {
  aws: {
    instanceType: "c5.16xlarge"
  }
}

SebastianGuelfi commented 1 year ago

Does anyone know how to configure the disk space in runtime? It seems "disks" is ignored by miniwdl and my workflow fails because "No space left on device" error.

mlin commented 1 year ago

@SebastianGuelfi

are you able to distinguish whether it's the task working directory (typically a shared NFS) vs. the container root filesystem (including /tmp; typically on local scratch disks or EBS volumes) that are filling up?
how are your Batch compute environments getting provisioned?