Merge branch 'develop'

Conflicts:
	platform.rb

Change-Id: Ic2e08e594d29a585691341c8667ac0b64933a505

8 files changed, 48 insertions, 3 deletions

diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg
index 28f333b0..95ddd2ca 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg
@@ -6,11 +6,19 @@
  I 127.0.0.1 localhost:5984 .* ok
  # https://leap.se/code/issues/5246
  I Shutting down group server
+ # ignore bigcouch conflict errors, mainly coming from tapicero creating new users
+ I Error in process.*{{nocatch,conflict}
  # ignore "Uncaught error in HTTP request: {exit, normal}" error
  # it's suppressed in later versions of bigcouch anhow
  # see https://leap.se/code/issues/5226
  I Uncaught error in HTTP request: {exit,normal}
  I Uncaught error in HTTP request: {exit,
+ # Ignore rexi_EXIT bigcouch error (Bug #6512) 
+ I Error in process <[0-9.]+> on node .* with exit value: {{rexi_EXIT,{(killed|noproc|shutdown),\[{couch_db,collect_results
+ # Ignore "Generic server terminating" bigcouch message (Feature #6544)
+ I Generic server <.*> terminating
+ I {error_report,<.*>,
+ I {error_info,
  C Uncaught error in HTTP request: {error,
  C Response abnormally terminated: {nodedown,
  C rexi_DOWN,noproc
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg
index 623d1e46..3af5045b 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg
@@ -2,4 +2,5 @@
  C WSGI application error
  C Error
  C error
- W Timing out client:
+# Removed this line because we determined it was better to ignore it (#6566)
+# W Timing out client: 
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/openvpn.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/openvpn.cfg
index d58e876d..ac17c0ca 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/openvpn.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/openvpn.cfg
@@ -2,6 +2,12 @@
 # suddenly hangup before properly establishing
 # a tls connection
  I ovpn-.*TLS Error: Unroutable control packet received from
- I ovpn-.*TLS Error: TLS key negotiation failed to occur within 60 seconds (check your network connectivity)
+ I ovpn-.*TLS Error: TLS key negotiation failed to occur within 60 seconds \(check your network connectivity\)
  I ovpn-.*TLS Error: TLS handshake failed
+ I ovpn-.*TLS Error: TLS object -> incoming plaintext read error
+ I ovpn-.*Fatal TLS error \(check_tls_errors_co\), restarting
+ I ovpn-.*TLS_ERROR: BIO read tls_read_plaintext error: error:140890B2:SSL routines:SSL3_GET_CLIENT_CERTIFICATE:no certificate
+
+ I ovpn-.*SIGUSR1\[soft,tls-error\] received, client-instance restarting
+ I ovpn-.*VERIFY ERROR: depth=0, error=certificate has expired
 
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/tapicero.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/tapicero.cfg
index 93ce0311..e5721eea 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/tapicero.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/tapicero.cfg
@@ -1,3 +1,5 @@
+# Ignore transient Tapicero errors when creating a db (#6511)
+ I tapicero.*(Creating database|Checking security of|Writing security to|Uploading design doc to) user-.* failed (\(trying again soon\)|(twice )?due to): (RestClient::Resource Not Found|RestClient::InternalServerError): (404 Resource Not Found|500 Internal Server Error)
  C tapicero.*RestClient::InternalServerError:
 # possible race condition between multiple tapicero
 # instances, so we ignore it
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog_tail.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog_tail.cfg
index 450b9e90..71395c50 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/syslog_tail.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog_tail.cfg
@@ -1,8 +1,14 @@
 # some general patterns
+ I Error: Driver 'pcspkr' is already registered, aborting...
+# ignore postfix errors on lost connection (Bug #6476)
+ I postfix/smtpd.*SSL_accept error from.*lost connection
+# ignore postfix too many errors after DATA (#6545)
+ I postfix/smtpd.*too many errors after DATA from
  C panic
  C Oops
- I Error: Driver 'pcspkr' is already registered, aborting...
  C Error
+# ignore ipv6 icmp errors for now (Bug #6540)
+ I kernel: .*icmpv6_send: no reply to icmp error
  C error
  W generic protection rip
  W .*Unrecovered read error - auto reallocate failed
diff --git a/puppet/modules/site_check_mk/files/extra_host_conf.mk b/puppet/modules/site_check_mk/files/extra_host_conf.mk
new file mode 100644
index 00000000..2c96f97a
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/extra_host_conf.mk
@@ -0,0 +1,6 @@
+# retry 3 times before setting a host into a hard state
+# and send out notification
+extra_host_conf["max_check_attempts"] = [ 
+  ("4", ALL_HOSTS ) 
+]
+
diff --git a/puppet/modules/site_check_mk/files/extra_service_conf.mk b/puppet/modules/site_check_mk/files/extra_service_conf.mk
new file mode 100644
index 00000000..03d1ea76
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/extra_service_conf.mk
@@ -0,0 +1,13 @@
+# retry 3 times before setting a service into a hard state
+# and send out notification
+extra_service_conf["max_check_attempts"] = [ 
+  ("4", ALL_HOSTS , ALL_SERVICES ) 
+]
+
+# run check_mk_agent every 2 minutes if it terminates
+# successfully.
+# see https://leap.se/code/issues/6539 for the rationale
+extra_service_conf["normal_check_interval"] = [
+  ("2", ALL_HOSTS , "Check_MK" )
+]
+
diff --git a/puppet/modules/site_check_mk/files/ignored_services.mk b/puppet/modules/site_check_mk/files/ignored_services.mk
new file mode 100644
index 00000000..35dc4433
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/ignored_services.mk
@@ -0,0 +1,3 @@
+ignored_services = [
+  ( ALL_HOSTS, [ "NTP Time" ] )
+]