Bug #12026: [crunchstat-summary] gives invalid min_cores recommendation - Arvados

Actions

Copy link

Bug #12026

closed

[crunchstat-summary] gives invalid min_cores recommendation

Added by Tom Morris over 6 years ago. Updated almost 5 years ago.

Status:

Resolved

Priority:

Normal

Assigned To:

Tom Morris

Category:

Target version:

2019-03-27 Sprint

Story points:

Release:

Arvados v1.4 - Q1/Q2 2019

Release relationship:

Auto

Description

This recommendation doesn't make sense from a couple of points of view:

### Summary for cwl-runner (su92l-8i9sb-ll9w11siojvlro6)
category        metric  task_max        task_max_rate   job_total
cpu     cpus    1       -       -
# Max CPU time spent by a single task: 1045.15s
# Max CPU usage in a single interval: 99.80%
# Overall CPU usage: 18.81%
#!! cwl-runner su92l-8i9sb-ll9w11siojvlro6 max CPU usage was 100% -- try runtime_constraints "min_cores_per_node":1

It shouldn't recommend a "change" to the same value as is current and it shouldn't recommend a change for max cpu < 100%

Subtasks 1 (0 open — 1 closed)

Actions

Copy link

Updated by Tom Morris over 6 years ago

Subject changed from crunchstat-summary gives invalid min_cores recommendation to [crunchstat-summary] gives invalid min_cores recommendation
Description updated (diff)

Actions

Copy link

Updated by Tom Morris over 6 years ago

I think the problem here was triggered by the fact that there was no cores run-time constraint specified. The job got scheduled on a single core machine (which seems like the right default), but crunchstat-summary contains this code:

        used_cores = max(1, int(math.ceil(cpu_max_rate)))
        asked_cores = self.existing_constraints.get(constraint_key)
        if asked_cores is None or used_cores < asked_cores:
        yield (
                '#!! {} max CPU usage was {}% -- '
                'try runtime_constraints "{}":{}'

I think perhaps asked_cores should have a default value of 1 if it isn't specified.

Actions

Copy link