Have to use container_commands because when commands are run the source bundle is not fully unpacked yet.
.ebextensions/whatever.config
container_commands:
chmod:
command: chmod +x .platform/hooks/predeploy/mount-volume.sh
Predeploy hooks run after container commands but before the deployment. No need to restart your docker container even if it mounts a directory on the attached ebs volume, because beanstalk spins it up after predeploy hooks complete. You can see it in the logs.
.platform/hooks/predeploy/mount-volume.sh
#!/bin/sh
# Make sure LF line endings are used in the file, otherwise there would be an error saying "file not found".
# All platform hooks run as root user, no need for sudo.
if mountpoint /path/to/mount/point; then
# Don't need to attach and mount if it's not an initial deploy but an app version update
exit 0
fi
# If it's a new EC2 instance created by ASG because the old one terminated,
# need to wait until the volume becomes available again. Can take a few minutes.
aws ec2 wait volume-available --volume-ids vol-xxx --region us-east-1
# Before attaching the volume find out the root volume's name, so that we can later use it for filtering purposes.
# -d – to filter out partitions.
# -P – to display the result as key-value pairs.
# -o – to output only the matching part.
# lsblk strips the "/dev/" part
ROOT_VOLUME_NAME=$(lsblk -d -P | grep -o 'NAME="[a-z0-9]*"' | grep -o '[a-z0-9]*')
aws ec2 attach-volume --volume-id vol-xxx --instance-id $(curl -s http://169.254.169.254/latest/meta-data/instance-id) --device /dev/sdf --region us-east-1
# The above command is async, so we need to wait.
aws ec2 wait volume-in-use --volume-ids vol-xxx --region us-east-1
# Now lsblk should show two devices. We figure out which one is non-root by filtering out the stored root volume name.
NON_ROOT_VOLUME_NAME=$(lsblk -d -P | grep -o 'NAME="[a-z0-9]*"' | grep -o '[a-z0-9]*' | awk -v name="$ROOT_VOLUME_NAME" '$0 !~ name')
FILE_COMMAND_OUTPUT=$(file -s /dev/$NON_ROOT_VOLUME_NAME)
# Create a file system on the non-root device only if there isn't one already, so that we don't accidentally override it.
if test "$FILE_COMMAND_OUTPUT" = "/dev/$NON_ROOT_VOLUME_NAME: data"; then
mkfs -t xfs /dev/$NON_ROOT_VOLUME_NAME
fi
mkdir -p /path/to/mount/point
mount /dev/$NON_ROOT_VOLUME_NAME /path/to/mount/point
# Need to make sure that the volume gets mounted after every reboot, because by default only root volume is automatically mounted.
cp /etc/fstab /etc/fstab.orig
NON_ROOT_VOLUME_UUID=$(lsblk -d -P -o +UUID | awk -v name="$NON_ROOT_VOLUME_NAME" '$0 ~ name' | grep -o 'UUID="[-0-9a-z]*"' | grep -o '[-0-9a-z]*')
# We specify 0 to prevent the file system from being dumped, and 2 to indicate that it is a non-root device.
# If you ever boot your instance without this volume attached, the nofail mount option enables the instance to boot
# even if there are errors mounting the volume.
# Debian derivatives, including Ubuntu versions earlier than 16.04, must also add the nobootwait mount option.
echo "UUID=$NON_ROOT_VOLUME_UUID /path/to/mount/point xfs defaults,nofail 0 2" | tee -a /etc/fstab
Pretty sure that things that I do with grep and awk could be done in a more concise manner. I'm not great at Linux.
Instance profile should include these permissions:
{
"Version": "2012-10-17",
"Statement": [
{
"Effect": "Allow",
"Action": [
"ec2:AttachVolume",
"ec2:DetachVolume",
"ec2:DescribeVolumes"
],
"Resource": "*"
}
]
}
You have to ensure that you deploy ebs volume in the same AZ as beanstalk and that you use SingleInstance deployment. Then if your instance crashes, ASG will terminate it, create another one, and attach the volume to the new instance keeping all the data.
ignoreErrorsyou cantest: "[ ! -b /dev/sdh ]". – Battista