Merge pull request #2975 from ktsaou/master

monitor /proc/net/sockstat

3 files changed, 60 insertions, 0 deletions

diff --git a/conf.d/health.d/tcp_conn.conf b/conf.d/health.d/tcp_conn.conf
new file mode 100644
index 0000000000..7aa9a98001
--- /dev/null
+++ b/conf.d/health.d/tcp_conn.conf
@@ -0,0 +1,19 @@
+
+#
+# ${tcp_max_connections} may be nan or -1 if the system
+# supports dynamic threshold for TCP connections.
+# In this case, the alarm will always be zero.
+#
+
+   alarm: tcp_connections
+      on: ipv4.tcpsock
+      os: linux
+   hosts: *
+    calc: (${tcp_max_connections} > 0) ? ( ${connections} * 100 / ${tcp_max_connections} ) : 0
+   units: %
+   every: 10s
+    warn: $this > (($status >= $WARNING ) ? ( 60 ) : ( 80 ))
+    crit: $this > (($status >= $CRITICAL) ? ( 80 ) : ( 90 ))
+   delay: up 0 down 5m multiplier 1.5 max 1h
+    info: the percentage of IPv4 TCP connections over the max allowed
+      to: sysadmin
diff --git a/conf.d/health.d/tcp_mem.conf b/conf.d/health.d/tcp_mem.conf
new file mode 100644
index 0000000000..6927d57652
--- /dev/null
+++ b/conf.d/health.d/tcp_mem.conf
@@ -0,0 +1,20 @@
+#
+# check
+# http://blog.tsunanet.net/2011/03/out-of-socket-memory.html
+#
+# We give a warning when TCP is under memory pressure
+# and a critical when TCP is 90% of its upper memory limit
+#
+
+   alarm: tcp_memory
+      on: ipv4.sockstat_tcp_mem
+      os: linux
+   hosts: *
+    calc: ${mem} * 100 / ${tcp_mem_high}
+   units: %
+   every: 10s
+    warn: ${mem} > (($status >= $WARNING  ) ? ( ${tcp_mem_pressure} * 0.8 ) : ( ${tcp_mem_pressure}   ))
+    crit: ${mem} > (($status >= $CRITICAL ) ? ( ${tcp_mem_pressure}       ) : ( ${tcp_mem_high} * 0.9 ))
+   delay: up 0 down 5m multiplier 1.5 max 1h
+    info: the amount of TCP memory as a percentage of its max memory limit
+      to: sysadmin
diff --git a/conf.d/health.d/tcp_orphans.conf b/conf.d/health.d/tcp_orphans.conf
new file mode 100644
index 0000000000..280d6590fc
--- /dev/null
+++ b/conf.d/health.d/tcp_orphans.conf
@@ -0,0 +1,21 @@
+
+#
+# check
+# http://blog.tsunanet.net/2011/03/out-of-socket-memory.html
+#
+# The kernel may penalize orphans by 2x or even 4x
+# so we alarm warning at 25% and critical at 50%
+#
+
+   alarm: tcp_orphans
+      on: ipv4.sockstat_tcp_sockets
+      os: linux
+   hosts: *
+    calc: ${orphan} * 100 / ${tcp_max_orphans}
+   units: %
+   every: 10s
+    warn: $this > (($status >= $WARNING ) ? ( 20 ) : ( 25 ))
+    crit: $this > (($status >= $CRITICAL) ? ( 25 ) : ( 50 ))
+   delay: up 0 down 5m multiplier 1.5 max 1h
+    info: the percentage of orphan IPv4 TCP sockets over the max allowed (this may lead to too-many-orphans errors)
+      to: sysadmin