Why does EC2 instance not start correctly after resizing root EBS volume?

8
votes

I was using the instructions on https://matt.berther.io/2015/02/03/how-to-resize-aws-ec2-ebs-volumes/ and http://atodorov.org/blog/2014/02/07/aws-tip-shrinking-ebs-root-volume-size/ to move to a EBS volume with less disk space. In both cases, when I attached the shrinked EBS volume(as /dev/xdva or /dev/sda1 , neither works) to an EC2 instance and start it, it stops on its own with the message 

State transition reason
Client.InstanceInitiatedShutdown: Instance initiated shutdown

Some more tinkering and I found that the new volume did not have BIOS boot partition. So I used gdisk to make one and copied the MBR from the original volume(that works and using which I can start instances) to the new volume. Now the instance does not terminate but I am not able to ssh into the newly launched instance.

What might be the reason behind this happening? How can I get more information(from logs/AWS Console etc) on why this is happening?

4
linuxamazon-web-servicesamazon-ec2amazon-ebs
what os are you using for your machine? do you have anything in the AWS EC2 Console -> Syslog ? – Mircea
Amazon Linux and the AWS EC2 console was empty. – web_ninja

4 Answers

4
votes

To shrink a GPT partioned boot EBS volume below the 8GB that standard images seem to use you can do the following: (a slight variation of the dd method from https://matt.berther.io/2015/02/03/how-to-resize-aws-ec2-ebs-volumes/ )

source disk is /dev/xvdf, target is /dev/xvdg

  1. Shrink source partition

    $ sudo e2fsck -f /dev/xvdf1
$ sudo resize2fs -M /dev/xvdf1

    Will print something like

    resize2fs 1.42.12 (29-Aug-2014)  
Resizing the filesystem on /dev/xvdf1 to 257491 (4k) blocks.  
The filesystem on /dev/xvdf1 is now 257491 (4k) blocks long.

    I converted this to MB, i.e. 257491 * 4 / 1024 ~= 1006 MB

  2. copy above size + a bit more from device to device (!), not just partition to partition, because that includes both partition table & data in the boot partition

    $ sudo dd if=/dev/xvdf of=/dev/xvdg bs=1M count=1100

  3. now use gdisk to fix the GPT partition on the new disk

    $ sudo gdisk /dev/xvdg

    You'll be greeted with roughly

    GPT fdisk (gdisk) version 0.8.10

Warning! Disk size is smaller than the main header indicates! Loading
secondary header from the last sector of the disk! You should use 'v' to
verify disk integrity, and perhaps options on the experts' menu to repair
the disk.
Caution: invalid backup GPT header, but valid main header; regenerating
backup header from main header.#

Warning! One or more CRCs don't match. You should repair the disk!

Partition table scan:
  MBR: protective
  BSD: not present
  APM: not present
  GPT: damaged

****************************************************************************
Caution: Found protective or hybrid MBR and corrupt GPT. Using GPT, but disk
verification and recovery are STRONGLY recommended.
****************************************************************************

Command (? for help):

    The following is the keyboard input within gdisk. To fix the problems, the data partition that is present in the copied partition table needs to be resized to fit on the new disk. This means it needs to be recreated smaller and it's properties need to be set to match the old partition definition. Didn't test it so it's maybe not required to relocate the backup table to the actual end of the disk but I did it anyways:

    • go to extra expert options: x
    • relocate backup data structures to the end of the disk: e
    • back to main menu: m

    Now to fixing the partition size

    • print and note some properties of partition 1 (and other non-boot partitions if they exist):
      i
      1
      Will show something like 

      Partition GUID code: 0FC63DAF-8483-4772-8E79-3D69D8477DE4 (Linux filesystem)
Partition unique GUID: DBA66894-D218-4D7E-A33E-A9EC9BF045DB
First sector: 4096 (at 2.0 MiB)
Last sector: 16777182 (at 8.0 GiB)
Partition size: 16773087 sectors (8.0 GiB)
Attribute flags: 0000000000000000
Partition name: 'Linux'

    • now delete
      d
      1
      and recreate the partition
      n
      1
      Enter the required parameters. All defaults worked for me here (= press enter), when in doubt refer to partition information from above

      • First sector = 4096
      • Last sector = whatever is the actual end of the new disk - take the default here
      • type = 8300 (Linux)

    • The new partition's default name did not match the old one. So change it to the original One
      c
      1
      Linux (see Partition name from above)

    • Next thing to change is the partition's GUID
      x
      c
      1
      DBA66894-D218-4D7E-A33E-A9EC9BF045DB (see Partition unique GUID, not the partition guid code above that)

    • That should be it. Back to main menu & print state
      m
      i
      1
      Will now print 

      Partition GUID code: 0FC63DAF-8483-4772-8E79-3D69D8477DE4 (Linux filesystem)
Partition unique GUID: DBA66894-D218-4D7E-A33E-A9EC9BF045DB
First sector: 4096 (at 2.0 MiB)
Last sector: 8388574 (at 4.0 GiB)
Partition size: 8384479 sectors (4.0 GiB)
Attribute flags: 0000000000000000
Partition name: 'Linux'

      The only change should be the Partition size.

    • write to disk and exit
      w
      y

  4. grow filesystem to match entire (smaller) disk. The fist step shrunk it down to the minimal size it can fit

    $ sudo resize2fs -p /dev/xvdg1

  5. We're done. Detach volume & snapshot it.

  6. Optional step. Choosing proper Kernel ID for the AMI.

If you are dealing with PVM image and encounter following mount error in instance logs

Kernel panic - not syncing: VFS: Unable to mount root

when your instance doesn't pass startup checks, you may probably be required to perform this additional step.

The solution to this error would be to choose proper Kernel ID for your PVM image during image creation from your snapshot. The full list of Kernel IDs (AKIs) can be obtained here.

Do choose proper AKI for your image, they are restricted by regions and architectures!

2
votes

The problem was with the BIOS boot partition. I was able to solve this by first initializing an instance with a smaller EBS volume. Then detaching the volume and attaching it to an instance whihc will be used to copyt the contents fromt he larger volume o the smaller volume. That created a BIOS boot partition which actually works. Simply creating a new one and copying the boot partition does not work.

Now following the steps outlined in any of the two links will help one shrink the volume of root EBS.

1
votes

Today, using UBUNTU doesn't work any other solution here. However, I found it:

  1. For caution: snapshot the large volume (backup)
  2. CREATE an instance IDENTICAL as possible such as LARGE volume works well. BUT with a SMALLER volume (desired size)
  3. Detach his new volume and ATTACH the large volume (as /dev/sda1) and START instance
  4. ATTACH the smaller new volume as /dev/sdf
  5. LOG IN in new instance. Mount smaller volume on /mnt: sudo mount -t ext4 /dev/xvdf1 /mnt
  6. DELETE everything on /mnt. Ignore WARNNG/error since /mnt can’t be deleted :) with sudo rm -rf /mnt
  7. Copy entire / to smaller volume: sudo cp -ax / /mnt
  8. Exit from instance and Stop it in AWS console
  9. Detach BOTH volumes. Now, re-attach the smaller volume, IMPORTANT, as /dev/sda1
  10. Start instance. LOG IN instance and confirm everything is ok with smaller volume
  11. Delete large volume, delete large snapshot, create a new snapshot of smaller volume. END.
0
votes

The above procedures are not complete, missing steps:

  1. Copy disk UUID
  2. Install grub boot loader
  3. Copy label

A more complete procedure can be found here:

https://medium.com/@m.yunan.helmy/decrease-the-size-of-ebs-volume-in-your-ec2-instance-ea326e951bce

This procedure is faster and more simple (no dd/resize2fs only rsync).

Tested with newer Nvme AWS disks.

Post any questions if you need help