Merge pull request #10403 from justinsb/ebs-for-master-data-2

cluster/aws/config-default.sh

@@ -37,6 +37,9 @@ IAM_PROFILE_MINION="kubernetes-minion"
 
 LOG="/dev/null"
 
+MASTER_DISK_TYPE="${MASTER_DISK_TYPE:-gp2}"
+MASTER_DISK_SIZE=${MASTER_DISK_SIZE:-20}
+
 MASTER_NAME="${INSTANCE_PREFIX}-master"
 MASTER_TAG="${INSTANCE_PREFIX}-master"
 MINION_TAG="${INSTANCE_PREFIX}-minion"

cluster/aws/config-test.sh

@@ -33,6 +33,9 @@ IAM_PROFILE_MINION="kubernetes-minion"
 
 LOG="/dev/null"
 
+MASTER_DISK_TYPE="${MASTER_DISK_TYPE:-gp2}"
+MASTER_DISK_SIZE=${MASTER_DISK_SIZE:-20}
+
 MASTER_NAME="${INSTANCE_PREFIX}-master"
 MASTER_TAG="${INSTANCE_PREFIX}-master"
 MINION_TAG="${INSTANCE_PREFIX}-minion"

cluster/aws/templates/format-disks.sh

@@ -66,6 +66,7 @@ else
 
   # Remove any existing mounts
   for block_device in ${block_devices}; do
+    echo "Unmounting ${block_device}"
     /bin/umount ${block_device}
     sed -i -e "\|^${block_device}|d" /etc/fstab
   done
@@ -80,26 +81,28 @@ else
       echo "Found multiple ephemeral block devices, formatting with btrfs as RAID-0"
       mkfs.btrfs -f --data raid0 ${block_devices[@]}
     fi
-    echo "${block_devices[0]}  /mnt  btrfs  noatime  0 0" >> /etc/fstab
-    mount /mnt
+    echo "${block_devices[0]}  /mnt/ephemeral  btrfs  noatime  0 0" >> /etc/fstab
+    mkdir -p /mnt/ephemeral
+    mount /mnt/ephemeral
 
-    mkdir -p /mnt/kubernetes
+    mkdir -p /mnt/ephemeral/kubernetes
 
-    move_docker="/mnt"
-    move_kubelet="/mnt/kubernetes"
+    move_docker="/mnt/ephemeral"
+    move_kubelet="/mnt/ephemeral/kubernetes"
   elif [[ ${docker_storage} == "aufs-nolvm" ]]; then
     if [[ ${#block_devices[@]} != 1 ]]; then
       echo "aufs-nolvm selected, but multiple ephemeral devices were found; only the first will be available"
     fi
 
     mkfs -t ext4 ${block_devices[0]}
-    echo "${block_devices[0]}  /mnt  ext4     noatime  0 0" >> /etc/fstab
-    mount /mnt
+    echo "${block_devices[0]}  /mnt/ephemeral  ext4     noatime  0 0" >> /etc/fstab
+    mkdir -p /mnt/ephemeral
+    mount /mnt/ephemeral
 
-    mkdir -p /mnt/kubernetes
+    mkdir -p /mnt/ephemeral/kubernetes
 
-    move_docker="/mnt"
-    move_kubelet="/mnt/kubernetes"
+    move_docker="/mnt/ephemeral"
+    move_kubelet="/mnt/ephemeral/kubernetes"
   elif [[ ${docker_storage} == "devicemapper" || ${docker_storage} == "aufs" ]]; then
     # We always use LVM, even with one device
     # In devicemapper mode, Docker can use LVM directly
@@ -144,21 +147,21 @@ else
       fi
 
       mkfs -t ext4 /dev/vg-ephemeral/docker
-      mkdir -p /mnt/docker
-      echo "/dev/vg-ephemeral/docker  /mnt/docker  ext4  noatime  0 0" >> /etc/fstab
-      mount /mnt/docker
-      move_docker="/mnt"
+      mkdir -p /mnt/ephemeral/docker
+      echo "/dev/vg-ephemeral/docker  /mnt/ephemeral/docker  ext4  noatime  0 0" >> /etc/fstab
+      mount /mnt/ephemeral/docker
+      move_docker="/mnt/ephemeral"
     fi
 
     # Remaining 5% is for kubernetes data
     # TODO: Should this be a thin pool?  e.g. would we ever want to snapshot this data?
     lvcreate -l 100%FREE -n kubernetes vg-ephemeral
     mkfs -t ext4 /dev/vg-ephemeral/kubernetes
-    mkdir -p /mnt/kubernetes
-    echo "/dev/vg-ephemeral/kubernetes  /mnt/kubernetes  ext4  noatime  0 0" >> /etc/fstab
-    mount /mnt/kubernetes
+    mkdir -p /mnt/ephemeral/kubernetes
+    echo "/dev/vg-ephemeral/kubernetes  /mnt/ephemeral/kubernetes  ext4  noatime  0 0" >> /etc/fstab
+    mount /mnt/ephemeral/kubernetes
 
-    move_kubelet="/mnt/kubernetes"
+    move_kubelet="/mnt/ephemeral/kubernetes"
   else
     echo "Ignoring unknown DOCKER_STORAGE: ${docker_storage}"
   fi

cluster/aws/templates/setup-master-pd.sh

+#!/bin/bash
+
+# Copyright 2015 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Format and mount the disk, create directories on it for all of the master's
+# persistent data, and link them to where they're used.
+
+echo "Waiting for master pd to be attached"
+attempt=0
+while true; do
+  echo Attempt "$(($attempt+1))" to check for /dev/xvdb
+  if [[ -e /dev/xvdb ]]; then
+    echo "Found /dev/xvdb"
+    break
+  fi
+  attempt=$(($attempt+1))
+  sleep 1
+done
+
+# Mount Master Persistent Disk
+echo "Mounting master-pd"
+mkdir -p /mnt/master-pd
+mkfs -t ext4 /dev/xvdb
+echo "/dev/xvdb  /mnt/master-pd  ext4  noatime  0 0" >> /etc/fstab
+mount /mnt/master-pd
+
+# Contains all the data stored in etcd
+mkdir -m 700 -p /mnt/master-pd/var/etcd
+# Contains the dynamically generated apiserver auth certs and keys
+mkdir -p /mnt/master-pd/srv/kubernetes
+# Contains the cluster's initial config parameters and auth tokens
+mkdir -p /mnt/master-pd/srv/salt-overlay
+# Directory for kube-apiserver to store SSH key (if necessary)
+mkdir -p /mnt/master-pd/srv/sshproxy
+
+ln -s -f /mnt/master-pd/var/etcd /var/etcd
+ln -s -f /mnt/master-pd/srv/kubernetes /srv/kubernetes
+ln -s -f /mnt/master-pd/srv/sshproxy /srv/sshproxy
+ln -s -f /mnt/master-pd/srv/salt-overlay /srv/salt-overlay
+
+# This is a bit of a hack to get around the fact that salt has to run after the
+# PD and mounted directory are already set up. We can't give ownership of the
+# directory to etcd until the etcd user and group exist, but they don't exist
+# until salt runs if we don't create them here. We could alternatively make the
+# permissions on the directory more permissive, but this seems less bad.
+if ! id etcd &>/dev/null; then
+  useradd -s /sbin/nologin -d /var/etcd etcd
+fi
+chown -R etcd /mnt/master-pd/var/etcd
+chgrp -R etcd /mnt/master-pd/var/etcd

cluster/aws/util.sh

@@ -27,6 +27,9 @@ ALLOCATE_NODE_CIDRS=true
 NODE_INSTANCE_PREFIX="${INSTANCE_PREFIX}-minion"
 ASG_NAME="${NODE_INSTANCE_PREFIX}-group"
 
+# We could allow the master disk volume id to be specified in future
+MASTER_DISK_ID=
+
 case "${KUBE_OS_DISTRIBUTION}" in
   ubuntu|wheezy|coreos)
     source "${KUBE_ROOT}/cluster/aws/${KUBE_OS_DISTRIBUTION}/util.sh"
@@ -55,7 +58,7 @@ MINION_SG_NAME="kubernetes-minion-${CLUSTER_ID}"
 # Be sure to map all the ephemeral drives.  We can specify more than we actually have.
 # TODO: Actually mount the correct number (especially if we have more), though this is non-trivial, and
 #  only affects the big storage instance types, which aren't a typical use case right now.
-BLOCK_DEVICE_MAPPINGS="[{\"DeviceName\": \"/dev/sdb\",\"VirtualName\":\"ephemeral0\"},{\"DeviceName\": \"/dev/sdc\",\"VirtualName\":\"ephemeral1\"},{\"DeviceName\": \"/dev/sdd\",\"VirtualName\":\"ephemeral2\"},{\"DeviceName\": \"/dev/sde\",\"VirtualName\":\"ephemeral3\"}]"
+BLOCK_DEVICE_MAPPINGS="[{\"DeviceName\": \"/dev/sdc\",\"VirtualName\":\"ephemeral0\"},{\"DeviceName\": \"/dev/sdd\",\"VirtualName\":\"ephemeral1\"},{\"DeviceName\": \"/dev/sde\",\"VirtualName\":\"ephemeral2\"},{\"DeviceName\": \"/dev/sdf\",\"VirtualName\":\"ephemeral3\"}]"
 
 function json_val {
     python -c 'import json,sys;obj=json.load(sys.stdin);print obj'$1''
@@ -361,6 +364,34 @@ function authorize-security-group-ingress {
   fi
 }
 
+# Gets master persistent volume, if exists
+# Sets MASTER_DISK_ID
+function find-master-pd {
+  local name=${MASTER_NAME}-pd
+  if [[ -z "${MASTER_DISK_ID}" ]]; then
+    MASTER_DISK_ID=`$AWS_CMD --output text describe-volumes \
+                             --filters Name=availability-zone,Values=${ZONE} \
+                                       Name=tag:Name,Values=${name} \
+                                       Name=tag:KubernetesCluster,Values=${CLUSTER_ID} \
+                             --query Volumes[].VolumeId`
+  fi
+}
+
+# Gets or creates master persistent volume
+# Sets MASTER_DISK_ID
+function ensure-master-pd {
+  local name=${MASTER_NAME}-pd
+
+  find-master-pd
+
+  if [[ -z "${MASTER_DISK_ID}" ]]; then
+    echo "Creating master disk: size ${MASTER_DISK_SIZE}GB, type ${MASTER_DISK_TYPE}"
+    MASTER_DISK_ID=`$AWS_CMD create-volume --availability-zone ${ZONE} --volume-type ${MASTER_DISK_TYPE} --size ${MASTER_DISK_SIZE} --query VolumeId --output text`
+    add-tag ${MASTER_DISK_ID} Name ${name}
+    add-tag ${MASTER_DISK_ID} KubernetesCluster ${CLUSTER_ID}
+  fi
+}
+
 # Verify prereqs
 function verify-prereqs {
   if [[ "$(which aws)" == "" ]]; then
@@ -724,6 +755,9 @@ function kube-up {
   # HTTPS to the master is allowed (for API access)
   authorize-security-group-ingress "${MASTER_SG_ID}" "--protocol tcp --port 443 --cidr 0.0.0.0/0"
 
+  # Get or create master persistent volume
+  ensure-master-pd
+
   (
     # We pipe this to the ami as a startup script in the user-data field.  Requires a compatible ami
     echo "#! /bin/bash"
@@ -756,6 +790,7 @@ function kube-up {
     echo "readonly DOCKER_STORAGE='${DOCKER_STORAGE:-}'"
     grep -v "^#" "${KUBE_ROOT}/cluster/aws/templates/common.sh"
     grep -v "^#" "${KUBE_ROOT}/cluster/aws/templates/format-disks.sh"
+    grep -v "^#" "${KUBE_ROOT}/cluster/aws/templates/setup-master-pd.sh"
     grep -v "^#" "${KUBE_ROOT}/cluster/aws/templates/create-dynamic-salt-files.sh"
     grep -v "^#" "${KUBE_ROOT}/cluster/aws/templates/download-release.sh"
     grep -v "^#" "${KUBE_ROOT}/cluster/aws/templates/salt-master.sh"
@@ -778,10 +813,9 @@ function kube-up {
   add-tag $master_id KubernetesCluster ${CLUSTER_ID}
 
   echo "Waiting for master to be ready"
-
   local attempt=0
 
-   while true; do
+  while true; do
     echo -n Attempt "$(($attempt+1))" to check for master node
     local ip=$(get_instance_public_ip ${master_id})
     if [[ -z "${ip}" ]]; then
@@ -797,8 +831,14 @@ function kube-up {
       KUBE_MASTER_IP=$(assign-elastic-ip $ip $master_id)
       echo -e " ${color_green}[master running @${KUBE_MASTER_IP}]${color_norm}"
 
-      # We are not able to add a route to the instance until that instance is in "running" state.
+      # We are not able to add a route or volume to the instance until that instance is in "running" state.
       wait-for-instance-running $master_id
+
+      # This is a race between instance start and volume attachment.  There appears to be no way to start an AWS instance with a volume attached.
+      # To work around this, we wait for volume to be ready in setup-master-pd.sh
+      echo "Attaching peristent data volume (${MASTER_DISK_ID}) to master"
+      $AWS_CMD attach-volume --volume-id ${MASTER_DISK_ID} --device /dev/sdb --instance-id ${master_id}
+
       sleep 10
       $AWS_CMD create-route --route-table-id $ROUTE_TABLE_ID --destination-cidr-block ${MASTER_IP_RANGE} --instance-id $master_id > $LOG