From 7b35c62b49f0a50a5fbe5aff0759c78326130785 Mon Sep 17 00:00:00 2001 From: Killian Murphy Date: Fri, 30 Oct 2020 16:13:04 +0000 Subject: Add beginning of QoS pending prototype raider --- .rubocop.yml | 5 +++++ pending_on_qos.rb | 44 ++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 49 insertions(+) create mode 100644 .rubocop.yml create mode 100644 pending_on_qos.rb diff --git a/.rubocop.yml b/.rubocop.yml new file mode 100644 index 0000000..48211dc --- /dev/null +++ b/.rubocop.yml @@ -0,0 +1,5 @@ +Layout/LineLength: + Max: 80 + +AllCops: + NewCops: enable diff --git a/pending_on_qos.rb b/pending_on_qos.rb new file mode 100644 index 0000000..7f7b3df --- /dev/null +++ b/pending_on_qos.rb @@ -0,0 +1,44 @@ +# frozen_string_literal: true + +# For each Viking partition, report the number of jobs pending due to QoS +# reasons. +class PendingOnQos + def initialize(collector, config) + @collector = collector + @partition_thresholds = { + nodes: 604_800, + week: 1_209_600, + month: 2_419_200, + himem: 604_800, + himem_week: 1_209_600, + gpu: 604_800, + interactive: 900, + test: 900, + preempt: 2_419_200 + } + end + + def raid + @partition_thresholds.each do |partition, threshold| + start_time = (Time.now - @partition_thresholds[:partition]) + .strftime('%Y-%m-:%d') + + squeue_cmd = [ + 'squeue', + '--format="%A,%R,%V"', + '--noheader', + '--parsable2', + "--partition=#{partition}", + '--state=PENDING' + ].join(' ') + end + + @collector.report!( + name: 'pending_on_qos', + value: 255, + help: 'Number of jobs pending for QoS reasons', + type: 'gauge', + labels: { partition: 'nodes' } + ) + end +end -- cgit v1.2.1