From 38427e19fdf5b2b8abbf08cc05dc51f637df801c Mon Sep 17 00:00:00 2001
From: hzma <hzma@alauda.io>
Date: Wed, 20 Apr 2022 18:07:42 +0800
Subject: [PATCH] add env-check

---
 dist/images/Dockerfile.base |  2 +-
 dist/images/env-check.sh    | 82 +++++++++++++++++++++++++++++++++++++
 dist/images/install.sh      | 22 +++++++++-
 3 files changed, 104 insertions(+), 2 deletions(-)
 create mode 100755 dist/images/env-check.sh

diff --git a/dist/images/Dockerfile.base b/dist/images/Dockerfile.base
index bab9aefe3c7..cf34ad13dec 100644
--- a/dist/images/Dockerfile.base
+++ b/dist/images/Dockerfile.base
@@ -49,7 +49,7 @@ ARG DEBIAN_FRONTEND=noninteractive
 RUN apt update && apt upgrade -y && apt install ca-certificates python3 hostname libunwind8 netbase \
         ethtool iproute2 ncat libunbound-dev procps libatomic1 kmod iptables \
         tcpdump ipset curl uuid-runtime openssl inetutils-ping arping ndisc6 \
-        logrotate libjemalloc2 dnsutils -y --no-install-recommends && \
+        logrotate libjemalloc2 dnsutils net-tools nmap -y --no-install-recommends && \
         rm -rf /var/lib/apt/lists/* && \
         cd /usr/sbin && \
         ln -sf /usr/sbin/iptables-legacy iptables && \
diff --git a/dist/images/env-check.sh b/dist/images/env-check.sh
new file mode 100755
index 00000000000..a2b00892a89
--- /dev/null
+++ b/dist/images/env-check.sh
@@ -0,0 +1,82 @@
+#!/bin/bash
+
+set +e
+checkresult=/var/log/kube-ovn/env-check.log
+echo "Start environment check" > $checkresult
+echo "1) check cni configuration" >> $checkresult
+if [ ! -e "/etc/cni/net.d" ]; then
+  echo "Directory /etc/cni/net.d does not exist, please check kube-ovn-cni pod status" >> $checkresult
+fi
+for file in $(ls "/etc/cni/net.d")
+do
+  if [[ ! $file =~ "kube-ovn.conflist" ]]; then
+    echo "Check files in /etc/cni/net.d, make sure if the config file $file should be deleted" >> $checkresult
+  fi
+done
+
+echo "2) check system ipv4 config" >> $checkresult
+probe_mtu=`cat /proc/sys/net/ipv4/tcp_mtu_probing`
+if [ $probe_mtu == 0 ]; then
+  echo "The 'tcp_mtu_probing' config may affect traffic, make sure if /proc/sys/net/ipv4/tcp_mtu_probing should be set to 1" >> $checkresult
+fi
+recycle=`cat /proc/sys/net/ipv4/tcp_tw_recycle`
+if [ $recycle == 1 ]; then
+  echo "The 'tcp_tw_recycle' config affects nodeport service, make sure change /proc/sys/net/ipv4/tcp_tw_recycle to 0" >> $checkresult
+fi
+
+echo "3) check checksum value" >> $checkresult
+netstat -s > /dev/null
+if [[ $? != 0 ]]; then
+  echo "The netstat cmd not found, maybe can be installed mannully and exec 'netstat -s' to check if there is 'InCsumErrors'" >> $checkresult
+  echo "If there's 'InCsumErrors' and the value is increasing, should exec cmd 'ethtool -K ETH tx off' to disable checksum, where 'ETH' is the nic used for traffics" >> $checkresult
+else
+  result=`netstat -s`
+  if [[ $result =~ "InCsumErrors" ]]; then
+    echo "Found 'InCsumErrors' para after exec 'netstat -s' cmd, check if the value is increasing, maybe should exec cmd 'ethtool -K ETH tx off' to disable checksum, where 'ETH' is the nic used for traffics" >> $checkresult
+  fi
+fi
+
+echo "4) check dns config" >> $checkresult
+result=`cat /etc/resolv.conf`
+if [[ $result =~ ".com" ]]; then
+  echo "There's *.com in dns search name, make sure the config /etc/resolv.conf is right" >> $checkresult
+fi
+
+echo "5) check firewall config" >> $checkresult
+result=`systemctl status firewalld`
+if [[ $result =~ "running" ]]; then
+  echo "The firewalld is running, make sure it has no effect on traffics across nodes" >> $checkresult
+fi
+
+result=`ps -ef | grep security | wc -l`
+if [[ $result > 1 ]]; then
+  echo "Found pid with '*security*' name, make sure it has no effect on traffics" >> $checkresult
+fi
+result=`ps -ef | grep qax | wc -l`
+if [[ $result > 1 ]]; then
+  echo "Found pid with '*qax*' name, make sure it has no effect on traffics" >> $checkresult
+fi
+result=`ps -ef | grep safe | wc -l`
+if [[ $result > 1 ]]; then
+  echo "Found pid with '*safe*' name, make sure it has no effect on traffics" >> $checkresult
+fi
+result=`ps -ef | grep defence | wc -l`
+if [[ $result > 1 ]]; then
+  echo "Found pid with '*defence*' name, make sure it has no effect on traffics" >> $checkresult
+fi
+result=`ps -ef | grep vmsec | wc -l`
+if [[ $result > 1 ]]; then
+  echo "Found pid with '*vmsec*' name, make sure it has no effect on traffics" >> $checkresult
+fi
+
+echo "6) check geneve 6081 connection" >> $checkresult
+nmap -sU 127.0.0.1 -p 6081 > /dev/null
+if [[ $? != 0 ]]; then
+  echo "The nmap cmd not found, maybe can be installed mannully and exec 'nmap -sU 127.0.0.1 -p 6081' to check port connection" >> $checkresult
+else
+  result=`nmap -sU 127.0.0.1 -p 6081`
+  wait
+  if [[ ! $result =~ "open" ]]; then
+    echo "The 6081 port for geneve encapsulation may be not available, please check if ovs-ovn pod is health" >> $checkresult
+  fi
+fi
diff --git a/dist/images/install.sh b/dist/images/install.sh
index 2f9e1867f29..ac05874753e 100755
--- a/dist/images/install.sh
+++ b/dist/images/install.sh
@@ -2748,6 +2748,7 @@ showHelp(){
   echo "  diagnose {all|node} [nodename]    diagnose connectivity of all nodes or a specific node"
   echo "  tuning {install-fastpath|local-install-fastpath|remove-fastpath|install-stt|local-install-stt|remove-stt} {centos7|centos8}} [kernel-devel-version]  deploy  kernel optimisation components to the system"
   echo "  reload restart all kube-ovn components"
+  echo "  env-check check the environment configuration"
 }
 
 tcpdump(){
@@ -3100,7 +3101,7 @@ getOvnCentralPod(){
     if [ -z "$VERSION" ]; then
           echo "kubeovn version not exists"
           exit 1
-        fi
+    fi
     KUBE_OVN_VERSION=$VERSION
 }
 
@@ -3408,6 +3409,22 @@ reload(){
   kubectl rollout status deployment/kube-ovn-monitor -n kube-system
 }
 
+env-check(){
+  set +e
+  echo "************************************************************************"
+  echo "The result will be saved to /var/log/kube-ovn/env-check.log on each node"
+  echo "************************************************************************"
+
+  KUBE_OVN_NS=kube-system
+  podNames=`kubectl get pod -n $KUBE_OVN_NS | grep kube-ovn-cni | awk '{print $1}'`
+  for pod in $podNames
+  do
+    kubectl exec -it -n $KUBE_OVN_NS $pod -- bash /kube-ovn/env-check.sh
+  done
+  echo "*************************************************************************"
+  echo "Finish environment check, check result in /var/log/kube-ovn/env-check.log"
+}
+
 if [ $# -lt 1 ]; then
   showHelp
   exit 0
@@ -3445,6 +3462,9 @@ case $subcommand in
   tuning)
     tuning "$@"
     ;;
+  env-check)
+    env-check
+    ;;
   *)
   showHelp
     ;;