services/api/test/unit/crunch_dispatch_test.rb

   1 require 'test_helper'
   2 require 'crunch_dispatch'
   3 require 'helpers/git_test_helper'
   4
   5 class CrunchDispatchTest < ActiveSupport::TestCase
   6   include GitTestHelper
   7
   8   test 'choose cheaper nodes first' do
   9     act_as_system_user do
  10       # Replace test fixtures with a set suitable for testing dispatch
  11       Node.destroy_all
  12
  13       # Idle nodes with different prices
  14       [['compute1', 3.20, 32],
  15        ['compute2', 1.60, 16],
  16        ['compute3', 0.80, 8]].each do |hostname, price, cores|
  17         Node.create!(hostname: hostname,
  18                      info: {
  19                        'slurm_state' => 'idle',
  20                      },
  21                      properties: {
  22                        'cloud_node' => {
  23                          'price' => price,
  24                        },
  25                        'total_cpu_cores' => cores,
  26                        'total_ram_mb' => cores*1024,
  27                        'total_scratch_mb' => cores*10000,
  28                      })
  29       end
  30
  31       # Node with no price information
  32       Node.create!(hostname: 'compute4',
  33                    info: {
  34                      'slurm_state' => 'idle',
  35                    },
  36                    properties: {
  37                      'total_cpu_cores' => 8,
  38                      'total_ram_mb' => 8192,
  39                      'total_scratch_mb' => 80000,
  40                    })
  41
  42       # Cheap but busy node
  43       Node.create!(hostname: 'compute5',
  44                    info: {
  45                      'slurm_state' => 'alloc',
  46                    },
  47                    properties: {
  48                      'cloud_node' => {
  49                        'price' => 0.10,
  50                      },
  51                      'total_cpu_cores' => 32,
  52                      'total_ram_mb' => 32768,
  53                      'total_scratch_mb' => 320000,
  54                    })
  55     end
  56
  57     dispatch = CrunchDispatch.new
  58     [[1, 16384, ['compute2']],
  59      [2, 16384, ['compute2', 'compute1']],
  60      [2, 8000, ['compute4', 'compute3']],
  61     ].each do |min_nodes, min_ram, expect_nodes|
  62       job = Job.new(runtime_constraints: {
  63                       'min_nodes' => min_nodes,
  64                       'min_ram_mb_per_node' => min_ram,
  65                     })
  66       nodes = dispatch.nodes_available_for_job_now job
  67       assert_equal expect_nodes, nodes
  68     end
  69   end
  70
  71   test 'respond to TERM' do
  72     lockfile = Rails.root.join 'tmp', 'dispatch.lock'
  73     ENV['CRUNCH_DISPATCH_LOCKFILE'] = lockfile.to_s
  74     begin
  75       pid = Process.fork do
  76         begin
  77           # Abandon database connections inherited from parent
  78           # process.  Credit to
  79           # https://github.com/kstephens/rails_is_forked
  80           ActiveRecord::Base.connection_handler.connection_pools.each_value do |pool|
  81             pool.instance_eval do
  82               @reserved_connections = {}
  83               @connections = []
  84             end
  85           end
  86           ActiveRecord::Base.establish_connection
  87
  88           dispatch = CrunchDispatch.new
  89           dispatch.stubs(:did_recently).returns true
  90           dispatch.run []
  91         ensure
  92           Process.exit!
  93         end
  94       end
  95       assert_with_timeout 5, "Dispatch did not lock #{lockfile}" do
  96         !can_lock(lockfile)
  97       end
  98     ensure
  99       Process.kill("TERM", pid)
 100     end
 101     assert_with_timeout 20, "Dispatch did not unlock #{lockfile}" do
 102       can_lock(lockfile)
 103     end
 104   end
 105
 106   test 'override --cgroup-root with CRUNCH_CGROUP_ROOT' do
 107     ENV['CRUNCH_CGROUP_ROOT'] = '/path/to/cgroup'
 108     Rails.configuration.crunch_job_wrapper = :none
 109     act_as_system_user do
 110       j = Job.create(repository: 'active/foo',
 111                      script: 'hash',
 112                      script_version: '4fe459abe02d9b365932b8f5dc419439ab4e2577',
 113                      script_parameters: {})
 114       ok = false
 115       Open3.expects(:popen3).at_least_once.with do |*args|
 116         if args.index(j.uuid)
 117           ok = ((i = args.index '--cgroup-root') and
 118                 (args[i+1] == '/path/to/cgroup'))
 119         end
 120         true
 121       end.raises(StandardError.new('all is well'))
 122       dispatch = CrunchDispatch.new
 123       dispatch.parse_argv ['--jobs']
 124       dispatch.refresh_todo
 125       dispatch.start_jobs
 126       assert ok
 127     end
 128   end
 129
 130   def assert_with_timeout timeout, message
 131     t = 0
 132     while (t += 0.1) < timeout
 133       if yield
 134         return
 135       end
 136       sleep 0.1
 137     end
 138     assert false, message + " (waited #{timeout} seconds)"
 139   end
 140
 141   def can_lock lockfile
 142     lockfile.open(File::RDWR|File::CREAT, 0644) do |f|
 143       return f.flock(File::LOCK_EX|File::LOCK_NB)
 144     end
 145   end
 146
 147   test 'rate limit of partial line segments' do
 148     act_as_system_user do
 149       Rails.configuration.crunch_log_partial_line_throttle_period = 1
 150
 151       job = {}
 152       job[:bytes_logged] = 0
 153       job[:log_throttle_bytes_so_far] = 0
 154       job[:log_throttle_lines_so_far] = 0
 155       job[:log_throttle_bytes_skipped] = 0
 156       job[:log_throttle_is_open] = true
 157       job[:log_throttle_partial_line_last_at] = Time.new(0)
 158       job[:log_throttle_first_partial_line] = true
 159
 160       dispatch = CrunchDispatch.new
 161
 162       line = "first log line"
 163       limit = dispatch.rate_limit(job, line)
 164       assert_equal true, limit
 165       assert_equal "first log line", line
 166       assert_equal 1, job[:log_throttle_lines_so_far]
 167
 168       # first partial line segment is skipped and counted towards skipped lines
 169       now = Time.now.strftime('%Y-%m-%d-%H:%M:%S')
 170       line = "#{now} localhost 100 0 stderr [...] this is first partial line segment [...]"
 171       limit = dispatch.rate_limit(job, line)
 172       assert_equal true, limit
 173       assert_includes line, "Rate-limiting partial segments of long lines", line
 174       assert_equal 2, job[:log_throttle_lines_so_far]
 175
 176       # next partial line segment within throttle interval is skipped but not counted towards skipped lines
 177       line = "#{now} localhost 100 0 stderr [...] second partial line segment within the interval [...]"
 178       limit = dispatch.rate_limit(job, line)
 179       assert_equal false, limit
 180       assert_equal 2, job[:log_throttle_lines_so_far]
 181
 182       # next partial line after interval is counted towards skipped lines
 183       sleep(1)
 184       line = "#{now} localhost 100 0 stderr [...] third partial line segment after the interval [...]"
 185       limit = dispatch.rate_limit(job, line)
 186       assert_equal false, limit
 187       assert_equal 3, job[:log_throttle_lines_so_far]
 188
 189       # this is not a valid line segment
 190       line = "#{now} localhost 100 0 stderr [...] does not end with [...] and is not a partial segment"
 191       limit = dispatch.rate_limit(job, line)
 192       assert_equal true, limit
 193       assert_equal "#{now} localhost 100 0 stderr [...] does not end with [...] and is not a partial segment", line
 194       assert_equal 4, job[:log_throttle_lines_so_far]
 195
 196       # this also is not a valid line segment
 197       line = "#{now} localhost 100 0 stderr does not start correctly but ends with [...]"
 198       limit = dispatch.rate_limit(job, line)
 199       assert_equal true, limit
 200       assert_equal "#{now} localhost 100 0 stderr does not start correctly but ends with [...]", line
 201       assert_equal 5, job[:log_throttle_lines_so_far]
 202     end
 203   end
 204 end