From bb68f740cd85f5c14c1122172ff758f59f7b915f Mon Sep 17 00:00:00 2001
From: Justin Santa Barbara <justin@fathomdb.com>
Date: Tue, 15 Mar 2016 20:26:46 -0400
Subject: [PATCH] AWS kube-up: collect logs from e2e

By implementing detect-node-names and a few other helpers, we are able
to collect the logs from nodes.
---
 cluster/aws/util.sh | 37 +++++++++++++++++++++++++++++--------
 cluster/log-dump.sh | 14 ++++++++++----
 2 files changed, 39 insertions(+), 12 deletions(-)

diff --git a/cluster/aws/util.sh b/cluster/aws/util.sh
index 95905abfe35..adb7d7d04a8 100755
--- a/cluster/aws/util.sh
+++ b/cluster/aws/util.sh
@@ -240,7 +240,13 @@ function query-running-minions () {
            --query ${query}
 }
 
-function find-running-minions () {
+function detect-node-names () {
+  # If this is called directly, VPC_ID might not be set
+  # (this is case from cluster/log-dump.sh)
+  if [[ -z "${VPC_ID:-}" ]]; then
+    VPC_ID=$(get_vpc_id)
+  fi
+
   NODE_IDS=()
   NODE_NAMES=()
   for id in $(query-running-minions "Reservations[].Instances[].InstanceId"); do
@@ -251,8 +257,14 @@ function find-running-minions () {
   done
 }
 
+# Called to detect the project on GCE
+# Not needed on AWS
+function detect-project() {
+  :
+}
+
 function detect-nodes () {
-  find-running-minions
+  detect-node-names
 
   # This is inefficient, but we want NODE_NAMES / NODE_IDS to be ordered the same as KUBE_NODE_IP_ADDRESSES
   KUBE_NODE_IP_ADDRESSES=()
@@ -1225,7 +1237,7 @@ function wait-minions {
     max_attempts=90
   fi
   while true; do
-    find-running-minions > $LOG
+    detect-node-names > $LOG
     if [[ ${#NODE_IDS[@]} == ${NUM_NODES} ]]; then
       echo -e " ${color_green}${#NODE_IDS[@]} minions started; ready${color_norm}"
       break
@@ -1552,24 +1564,33 @@ function test-teardown {
 }
 
 
-# SSH to a node by name ($1) and run a command ($2).
-function ssh-to-node {
+# Gets the hostname (or IP) that we should SSH to for the given nodename
+# For the master, we use the nodename, for the nodes we use their instanceids
+function get_ssh_hostname {
   local node="$1"
-  local cmd="$2"
 
   if [[ "${node}" == "${MASTER_NAME}" ]]; then
     node=$(get_instanceid_from_name ${MASTER_NAME})
     if [[ -z "${node-}" ]]; then
-      echo "Could not detect Kubernetes master node.  Make sure you've launched a cluster with 'kube-up.sh'"
+      echo "Could not detect Kubernetes master node.  Make sure you've launched a cluster with 'kube-up.sh'" 1>&2
       exit 1
     fi
   fi
 
   local ip=$(get_instance_public_ip ${node})
   if [[ -z "$ip" ]]; then
-    echo "Could not detect IP for ${node}."
+    echo "Could not detect IP for ${node}." 1>&2
     exit 1
   fi
+  echo ${ip}
+}
+
+# SSH to a node by name ($1) and run a command ($2).
+function ssh-to-node {
+  local node="$1"
+  local cmd="$2"
+
+  local ip=$(get_ssh_hostname ${node})
 
   for try in $(seq 1 5); do
     if ssh -oLogLevel=quiet -oConnectTimeout=30 -oStrictHostKeyChecking=no -i "${AWS_SSH_KEY}" ${SSH_USER}@${ip} "echo test > /dev/null"; then
diff --git a/cluster/log-dump.sh b/cluster/log-dump.sh
index b751021dab2..ea59d2f455d 100755
--- a/cluster/log-dump.sh
+++ b/cluster/log-dump.sh
@@ -45,7 +45,7 @@ function copy-logs-from-node() {
     local -r scp_files="{$(echo ${files[*]} | tr ' ' ',')}"
 
     if [[ "${KUBERNETES_PROVIDER}" == "aws" ]]; then
-        local ip=$(get_instance_public_ip "${node}")
+        local ip=$(get_ssh_hostname "${node}")
         scp -i "${AWS_SSH_KEY}" "${SSH_USER}@${ip}:${scp_files}" "${dir}" > /dev/null || true
     else
         gcloud compute copy-files --project "${PROJECT}" --zone "${ZONE}" "${node}:${scp_files}" "${dir}" > /dev/null || true
@@ -62,10 +62,14 @@ function save-logs() {
     if [[ "${KUBERNETES_PROVIDER}" == "gce" ]]; then
         files="${files} ${gce_logfiles}"
     fi
+    if [[ "${KUBERNETES_PROVIDER}" == "aws" ]]; then
+        files="${files} ${aws_logfiles}"
+    fi
     if ssh-to-node "${node_name}" "sudo systemctl status kubelet.service" &> /dev/null; then
         ssh-to-node "${node_name}" "sudo journalctl --output=cat -u kubelet.service" > "${dir}/kubelet.log" || true
+        ssh-to-node "${node_name}" "sudo journalctl --output=cat -u docker.service" > "${dir}/docker.log" || true
     else
-        files="${files} ${supervisord_logfiles}"
+        files="${files} ${initd_logfiles} ${supervisord_logfiles}"
     fi
     copy-logs-from-node "${node_name}" "${dir}" "${files}"
 }
@@ -75,8 +79,10 @@ readonly node_ssh_supported_providers="gce gke aws"
 
 readonly master_logfiles="kube-apiserver kube-scheduler kube-controller-manager etcd"
 readonly node_logfiles="kube-proxy"
+readonly aws_logfiles="cloud-init-output"
 readonly gce_logfiles="startupscript"
-readonly common_logfiles="kern docker"
+readonly common_logfiles="kern"
+readonly initd_logfiles="docker"
 readonly supervisord_logfiles="kubelet supervisor/supervisord supervisor/kubelet-stdout supervisor/kubelet-stderr"
 
 # Limit the number of concurrent node connections so that we don't run out of
@@ -85,7 +91,7 @@ readonly max_scp_processes=25
 
 if [[ ! "${master_ssh_supported_providers}" =~ "${KUBERNETES_PROVIDER}" ]]; then
     echo "Master SSH not supported for ${KUBERNETES_PROVIDER}"
-elif ! $(detect-master &> /dev/null); then
+elif ! (detect-master &> /dev/null); then
     echo "Master not detected. Is the cluster up?"
 else
     readonly master_dir="${report_dir}/${MASTER_NAME}"