Optimize kernel neighbour table for large scale environments

Changing the default values of neighbor table (also known as ARP table)
in the kernel to avoid neighbour table overflow and thus fix
communication errors between overcloud nodes.
default kernel values support L2 network up to 1024 hosts (/22).
The patch will allow up to 4096 hosts (/20).

Change-Id: I5fabc766dd75a38cd3d835deee7e168f04dd30ce
Closes-Bug: #1690087

1 files changed, 29 insertions, 0 deletions

diff --git a/puppet/services/kernel.yaml b/puppet/services/kernel.yaml
index 2a335b67..5cbd1de4 100644
--- a/puppet/services/kernel.yaml
+++ b/puppet/services/kernel.yaml
@@ -26,6 +26,28 @@ parameters:
     default: 0
     description: Configures sysctl net.ipv6.{default/all}.disable_ipv6 keys
     type: number
+  NeighbourGcThreshold1:
+    default: 1024
+    description: Configures sysctl net.ipv4.neigh.default.gc_thresh1 value.
+                 This is the minimum number of entries to keep in the ARP
+                 cache. The garbage collector will not run if there are
+                 fewer than this number of entries in the cache.
+    type: number
+  NeighbourGcThreshold2:
+    default: 2048
+    description: Configures sysctl net.ipv4.neigh.default.gc_thresh2 value.
+                 This is the soft maximum number of entries to keep in the
+                 ARP cache. The garbage collector will  allow the number of
+                 entries to exceed this for 5 seconds before collection will
+                 be performed.
+    type: number
+  NeighbourGcThreshold3:
+    default: 4096
+    description: Configures sysctl net.ipv4.neigh.default.gc_thresh3 value.
+                 This is the hard maximum number of entries to keep in the
+                 ARP cache. The garbage collector will always run if there
+                 are more than this number of entries in the cache.
+    type: number
 
 outputs:
   role_data:
@@ -86,5 +108,12 @@ outputs:
             value: 1
           fs.suid_dumpable:
             value: 0
+          #avoid neighbour table overflow on large deployments
+          net.ipv4.neigh.default.gc_thresh1:
+            value: {get_param: NeighbourGcThreshold1}
+          net.ipv4.neigh.default.gc_thresh2:
+            value: {get_param: NeighbourGcThreshold2}
+          net.ipv4.neigh.default.gc_thresh3:
+            value: {get_param: NeighbourGcThreshold3}
       step_config: |
         include ::tripleo::profile::base::kernel