From d166a073ad3a32af77152bf4b16206c32f9890b6 Mon Sep 17 00:00:00 2001
From: Nat Lasseter <nat.lasseter@york.ac.uk>
Date: Tue, 17 Mar 2020 10:14:18 +0000
Subject: Initial commit: broke out from longboat

---
 slurm_job_states.rb | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
 create mode 100644 slurm_job_states.rb

(limited to 'slurm_job_states.rb')

diff --git a/slurm_job_states.rb b/slurm_job_states.rb
new file mode 100644
index 0000000..20b69c8
--- /dev/null
+++ b/slurm_job_states.rb
@@ -0,0 +1,41 @@
+class SlurmJobStates
+  def initialize(collector, config)
+    @collector = collector
+    @interval = config[:raid_every]
+  end
+
+  def raid
+    start_time = (Time.now - @interval).strftime("%H:%M:%S")
+
+    # Get raw data from sacct,
+    # read jobs into an array,
+    # remove any whitespace from the ends of each string,
+    # drop the header,
+    # and split each line into state and partition
+    raw = `sacct -a -P -o State,Partition -S #{start_time}`.
+      lines.
+      map(&:strip)[1..-1].
+      map{|l|l.split("|")}
+
+    # Make a tally of each state/partition combo
+    tally = Hash.new{0}
+    raw.each do |job|
+      tally[job] += 1
+    end
+
+    # Clean up any previously reported metrics
+    # to prevent stale labelsets
+    @collector.redact!("slurm_job_states")
+
+    # Report new metrics
+    tally.each do |labelset, number|
+      @collector.report!(
+        "slurm_job_states",
+        number,
+        help: "Number of jobs in each state",
+        type: "gauge",
+        labels: {state: labelset[0], partition: labelset[1]}
+      )
+    end
+  end
+end
-- 
cgit v1.2.1