Runtime attributes

GATK-SV is implemented as a set of WDLs designed to run on Google Cloud Platform. Computations are broken up into a set of tasks that are carried out in a particular order on cloud virtual machines (VMs), each of which possesses a limited set of resources. These include the following primary components:

CPU cores
Random access memory (RAM)
Disk storage

GATK-SV attempts to request the appropriate amount of resources for each task and in some cases employs mathematical models to predict optimal requirements that minimize cost. However, if the actual computations performed require more than the requested resources on the VM, the task may fail.

info

Most tasks in GATK-SV are tuned for cohorts with ~150 samples. Most tasks should scale automatically, but running larger cohorts may lead to slowdowns or errors due to insufficient resource allocation.

In addition to VM resources, Terra and Cromwell both support automatic retries of tasks in case of ephemeral errors, as well as requesting preemptible VMs at a significant discount.

Setting runtime attributes

GATK-SV exposes optional parameters for manually tuning VM resource allocation, automatic retries, and the use of preemptible instances. These parameters are all a custom struct called RuntimeAttr, which is defined in Structs.wdl as:

struct RuntimeAttr {
    Float? mem_gb
    Int? cpu_cores
    Int? disk_gb
    Int? boot_disk_gb
    Int? preemptible_tries
    Int? max_retries
}

Users encountering errors due to insufficient VM resources or wishing to adjust automatic and preemptible retries may modify the corresponding parameters. Users should inspect the WDL source to determine whether a given RuntimeAttr parameter corresponds to a specific task.

Setting runtime attributes​

Further reading​

Setting runtime attributes

Further reading