Merge branch…

lib/gitlab/health_checks/fs_shards_check.rb

@@ -2,6 +2,9 @@ module Gitlab
   module HealthChecks
     class FsShardsCheck
       extend BaseAbstractCheck
+      RANDOM_STRING = SecureRandom.hex(1000).freeze
+      COMMAND_TIMEOUT = '1'.freeze
+      TIMEOUT_EXECUTABLE = 'timeout'.freeze
 
       class << self
         def readiness
@@ -41,8 +44,6 @@ module Gitlab
 
         private
 
-        RANDOM_STRING = SecureRandom.hex(1000).freeze
-
         def operation_metrics(ok_metric, latency_metric, operation, **labels)
           with_timing operation do |result, elapsed|
             [
@@ -63,8 +64,8 @@ module Gitlab
           @storage_paths ||= Gitlab.config.repositories.storages
         end
 
-        def with_timeout(args)
-          %w{timeout 1}.concat(args)
+        def exec_with_timeout(cmd_args, *args, &block)
+          Gitlab::Popen.popen([TIMEOUT_EXECUTABLE, COMMAND_TIMEOUT].concat(cmd_args), *args, &block)
         end
 
         def tmp_file_path(storage_name)
@@ -78,7 +79,7 @@ module Gitlab
         def storage_stat_test(storage_name)
           stat_path = File.join(path(storage_name), '.')
           begin
-            _, status = Gitlab::Popen.popen(with_timeout(%W{ stat #{stat_path} }))
+            _, status = exec_with_timeout(%W{ stat #{stat_path} })
             status == 0
           rescue Errno::ENOENT
             File.exist?(stat_path) && File::Stat.new(stat_path).readable?
@@ -86,7 +87,7 @@ module Gitlab
         end
 
         def storage_write_test(tmp_path)
-          _, status = Gitlab::Popen.popen(with_timeout(%W{ tee #{tmp_path} })) do |stdin|
+          _, status = exec_with_timeout(%W{ tee #{tmp_path} }) do |stdin|
             stdin.write(RANDOM_STRING)
           end
           status == 0
@@ -96,7 +97,7 @@ module Gitlab
         end
 
         def storage_read_test(tmp_path)
-          _, status = Gitlab::Popen.popen(with_timeout(%W{ diff #{tmp_path} - })) do |stdin|
+          _, status = exec_with_timeout(%W{ diff #{tmp_path} - }) do |stdin|
             stdin.write(RANDOM_STRING)
           end
           status == 0
@@ -106,7 +107,7 @@ module Gitlab
         end
 
         def delete_test_file(tmp_path)
-          _, status = Gitlab::Popen.popen(with_timeout(%W{ rm -f #{tmp_path} }))
+          _, status = exec_with_timeout(%W{ rm -f #{tmp_path} })
           status == 0
         rescue Errno::ENOENT
           File.delete(tmp_path) rescue Errno::ENOENT

spec/lib/gitlab/health_checks/fs_shards_check_spec.rb

 require 'spec_helper'
 
 describe Gitlab::HealthChecks::FsShardsCheck do
+  def command_exists?(command)
+    _, status = Gitlab::Popen.popen(%W{ #{command} 1 echo })
+    status == 0
+  rescue Errno::ENOENT
+    false
+  end
+
+  def timeout_command
+    @timeout_command ||=
+      if command_exists?('timeout')
+        'timeout'
+      elsif command_exists?('gtimeout')
+        'gtimeout'
+      else
+        ''
+      end
+  end
+
   let(:metric_class) { Gitlab::HealthChecks::Metric }
   let(:result_class) { Gitlab::HealthChecks::Result }
   let(:repository_storages) { [:default] }
@@ -15,6 +33,7 @@ describe Gitlab::HealthChecks::FsShardsCheck do
   before do
     allow(described_class).to receive(:repository_storages) { repository_storages }
     allow(described_class).to receive(:storages_paths) { storages_paths }
+    stub_const('Gitlab::HealthChecks::FsShardsCheck::TIMEOUT_EXECUTABLE', timeout_command)
   end
 
   after do
@@ -78,40 +97,76 @@ describe Gitlab::HealthChecks::FsShardsCheck do
           }.with_indifferent_access
         end
 
-        it { is_expected.to include(metric_class.new(:filesystem_accessible, 0, shard: :default)) }
-        it { is_expected.to include(metric_class.new(:filesystem_readable, 0, shard: :default)) }
-        it { is_expected.to include(metric_class.new(:filesystem_writable, 0, shard: :default)) }
+        it { is_expected.to all(have_attributes(labels: { shard: :default })) }
+
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_accessible, value: 0)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_readable, value: 0)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_writable, value: 0)) }
 
-        it { is_expected.to include(have_attributes(name: :filesystem_access_latency, value: be >= 0, labels: { shard: :default })) }
-        it { is_expected.to include(have_attributes(name: :filesystem_read_latency, value: be >= 0, labels: { shard: :default })) }
-        it { is_expected.to include(have_attributes(name: :filesystem_write_latency, value: be >= 0, labels: { shard: :default })) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_access_latency, value: be >= 0)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_read_latency, value: be >= 0)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_write_latency, value: be >= 0)) }
       end
 
       context 'storage points to directory that has both read and write rights' do
         before do
           FileUtils.chmod_R(0755, tmp_dir)
         end
+        it { is_expected.to all(have_attributes(labels: { shard: :default })) }
 
-        it { is_expected.to include(metric_class.new(:filesystem_accessible, 1, shard: :default)) }
-        it { is_expected.to include(metric_class.new(:filesystem_readable, 1, shard: :default)) }
-        it { is_expected.to include(metric_class.new(:filesystem_writable, 1, shard: :default)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_accessible, value: 1)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_readable, value: 1)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_writable, value: 1)) }
 
-        it { is_expected.to include(have_attributes(name: :filesystem_access_latency, value: be >= 0, labels: { shard: :default })) }
-        it { is_expected.to include(have_attributes(name: :filesystem_read_latency, value: be >= 0, labels: { shard: :default })) }
-        it { is_expected.to include(have_attributes(name: :filesystem_write_latency, value: be >= 0, labels: { shard: :default })) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_access_latency, value: be >= 0)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_read_latency, value: be >= 0)) }
+        it { is_expected.to include(an_object_having_attributes(name: :filesystem_write_latency, value: be >= 0)) }
+      end
+    end
+  end
+
+  context 'when timeout kills fs checks' do
+    before do
+      stub_const('Gitlab::HealthChecks::FsShardsCheck::COMMAND_TIMEOUT', '1')
+
+      allow(described_class).to receive(:exec_with_timeout).and_wrap_original { |m| m.call(%w(sleep 60)) }
+      FileUtils.chmod_R(0755, tmp_dir)
+    end
+
+    describe '#readiness' do
+      subject { described_class.readiness }
+
+      it { is_expected.to include(result_class.new(false, 'cannot stat storage', shard: :default)) }
+    end
+
+    describe '#metrics' do
+      subject { described_class.metrics }
+
+      it 'provides metrics' do
+        expect(subject).to all(have_attributes(labels: { shard: :default }))
+
+        expect(subject).to include(an_object_having_attributes(name: :filesystem_accessible, value: 0))
+        expect(subject).to include(an_object_having_attributes(name: :filesystem_readable, value: 0))
+        expect(subject).to include(an_object_having_attributes(name: :filesystem_writable, value: 0))
+
+        expect(subject).to include(an_object_having_attributes(name: :filesystem_access_latency, value: be >= 0))
+        expect(subject).to include(an_object_having_attributes(name: :filesystem_read_latency, value: be >= 0))
+        expect(subject).to include(an_object_having_attributes(name: :filesystem_write_latency, value: be >= 0))
       end
     end
   end
 
   context 'when popen always finds required binaries' do
     before do
-      allow(Gitlab::Popen).to receive(:popen).and_wrap_original do |method, *args, &block|
+      allow(described_class).to receive(:exec_with_timeout).and_wrap_original do |method, *args, &block|
         begin
           method.call(*args, &block)
-        rescue RuntimeError
+        rescue RuntimeError, Errno::ENOENT
           raise 'expected not to happen'
         end
       end
+
+      stub_const('Gitlab::HealthChecks::FsShardsCheck::COMMAND_TIMEOUT', '10')
     end
 
     it_behaves_like 'filesystem checks'