Make pid_max settable on OSD nodes.

On OSD nodes during recovery you can ovewhelm the default PID max of
32k. The docs recommend setting to the largest possible value, 4194303.
By default this will not be set.

http://docs.ceph.com/docs/jewel/rados/troubleshooting/troubleshooting-osd/

Change-Id: Id7714e8d237db2137fdc32002ab7ae373fb81357
Closes-Bug: #1362825

manifests/osds.pp

@@ -27,6 +27,24 @@
 # [*defaults*] A config hash
 #   Optional. Defaults to a empty hash
 #
-class ceph::osds($args = {}, $defaults = {}) {
+# [*pid_max*] Value for pid_max. Defaults to undef. Optional.
+#   For OSD nodes it is recommended that you raise pid_max above the
+#   default value because you may hit the system max during
+#   recovery. The recommended value is the absolute max for pid_max: 4194303
+#   http://docs.ceph.com/docs/jewel/rados/troubleshooting/troubleshooting-osd/
+#
+class ceph::osds(
+  $args = {},
+  $defaults = {},
+  $pid_max = $::ceph::profile::params::pid_max,
+)
+{
   create_resources(ceph::osd, $args, $defaults)
+
+  if $pid_max {
+    $sysctl_settings = {
+      'kernel.pid_max' => { value => $pid_max },
+    }
+    ensure_resources(sysctl::value,$sysctl_settings)
+  }
 }

manifests/profile/params.pp

@@ -145,6 +145,12 @@
 #   Set to 0 to disable it.
 #   Optional. Default provided by Ceph
 #
+# [*pid_max*] Value for pid_max. Defaults to undef. Optional.
+#   For OSD nodes it is recommended that you raise pid_max above the
+#   default value because you may hit the system max during
+#   recovery. The recommended value is the absolute max for pid_max: 4194303
+#   http://docs.ceph.com/docs/jewel/rados/troubleshooting/troubleshooting-osd/
+#
 # [*rgw_keystone_version*] The api version for keystone.
 #   Possible values 'v2.0', 'v3'
 #   Optional. Default is 'v2.0'
@@ -199,6 +205,7 @@ class ceph::profile::params (
   $osd_recovery_max_single_start = undef,
   $osd_max_scrubs = undef,
   $osd_op_threads = undef,
+  $pid_max = undef,
   $rgw_keystone_version = 'v2.0',
   $rgw_keystone_admin_domain = undef,
   $rgw_keystone_admin_project = undef,

metadata.json

@@ -66,6 +66,10 @@
     {
       "name": "puppetlabs/stdlib",
       "version_requirement": ">=4.10.0 <5.0.0"
+    },
+    {
+      "name": "duritong/sysctl",
+      "version_requirement": ">=0.0.1 <1.0.0"
     }
   ]
 }

releasenotes/notes/allow_changing_pidmax_on_osd_nodes-d1a98328f666a895.yaml

@@ -0,0 +1,6 @@
+---
+features:
+  - On OSD nodes you can run out of PIDs during a recovery.
+    Therefore it is recommended that you raise pid_max. This
+    change makes that possible with an enablement flag and
+    the ability to set the value.

spec/classes/ceph_osds_spec.rb

@@ -39,14 +39,25 @@ describe 'ceph::osds' do
       is_expected.to contain_ceph__osd('/dev/sdb').with(
         :ensure  => 'present',
         :journal => '/srv/journal',
-        :cluster => 'CLUSTER'
-    )
+        :cluster => 'CLUSTER')
       is_expected.to contain_ceph__osd('/srv/data').with(
         :ensure  => 'present',
         :cluster => 'CLUSTER')
+      is_expected.not_to contain_sysctl__value('kernel.pid_max')
     }
   end
 
+  context 'sets pid_max when enabled' do
+    let :params do
+    {
+      :pid_max => 123456,
+    }
+    end
+    it do
+      is_expected.to contain_sysctl__value('kernel.pid_max').with_value(123456)
+    end
+  end
+
   describe 'Ubuntu' do
     let :facts do
       {