<!channel> any other opinions on :point_up:
# team-maintainers
j
@channel any other opinions on ☝️
a
...do we need a label? What's wrong with setting the
process.accelerator
?
or
process.queue
for other HPCs, or
process.machineType
for GCP or Azure...OK maybe we need a label
😅 1
j
That was the original cointext
Copy code
// Default executor
    executor = 'lsf'

    // Dynamic queue selection based on job requirements
    queue = {
        if (task.time > 144.h) {
            return 'long'
        } else if (task.label && task.label.toString().contains('process_low_gpu')) {
            return 'gpuexpress'
        } else if (task.label && task.label.toString().contains('gpu')) {
            return 'gpu'
        } else if (task.time <= 12.h && task.cpus <= 8) {
            return 'express'
        } else {
            return 'premium'
        }
    }
a
Copy code
withQueue
Is that a thing?!
j
OH I didn't see that
a
Well I see no harm in adding
label 'gpu'
, which can then be combined with
label process_low
etc
But I don't see why we should make it more complicated with
process_low_gpu
,
j
but I think the point according to their queue logic is also the number of GPUs to request?
process.accelerator takes integers no?
a
Copy code
queue = {
    if (task.label && task.label.contains("process_low") && task.label.contains("gpu")) {
        return 'gpuexpress'
    } else if (task.label && task.label.contains("gpu")) {
        return 'gpu'
    } else {
        return 'non_gpu'
    }
}
☝️ just treat label process_gpu as a boolean, then set the logic for your particular system
👍 1
j
You're being outvoted atm though
a
process.accelerator takes integers no?
it behaves differently on different executors
👍 1
According to the vote everyone voted for number 2?
j
OH FFS
😆 1
I cannot read this morning
a
You know Slack has a poll system right?
j
Which sends me through like 4 windows nowadays
🤮 1
a
(I've never used it)
j
Beacuyse they want you to PAY fori t
shocked 1
OK apparnetly this is already solved: https://github.com/nf-core/tools/pull/3272/files
a
Let's thinkkkk through each option: On premise (HPC):
Copy code
queue = {
    if (task.label && task.label.contains("process_low") && task.label.contains("gpu")) {
        return 'gpuexpress'
    } else if (task.label && task.label.contains("gpu")) {
        return 'gpu'
    } else {
        return 'non_gpu'
    }
}
👍 1
AWS and GCP:
Copy code
process {
    accelerator = {
    if (task.label && task.label.contains("process_low") && task.label.contains("process_gpu")) {
        return 1
    } else if (task.label && task.label.contains("gpu")) {
        return 4
    } else {
        return null
    }
}
👍 1
Azure:
Copy code
process {
    machineType = {
    if (task.label && task.label.contains("gpu")) {
        return "Standard_NC*_v3"
    } else {
        return "Standard_E*d_v5"
    }
}
Probably less tidy than using
process_low_gpu
and `process_high_gpu`:
Copy code
process {
    withName: process_low_gpu {
        accelerator = 1
    }
    withName: process_high_gpu {
        accelerator = 4
    }
}