projects::arch linux on ec2
Arch Linux's minimalistic philosophy and high degree of customizability makes it a great choice for compute cloud deployment. It's light, fast, and scalable.
I am doing releases of Amazon EC2 AMIs based on Arch Linux roughly once per month.
NOTE: This project is not run, sponsored, or endorsed by my employer or the Arch Linux project.
The EC2 image build process is public, but the AMI registration portion is not. Here are the necessary tools to create an image file, but see the 2013-05-26 news post for information on how to register the images in EC2.
- ami-build-backend - These files are held on the PXE server, and fetched when the guest boots.
- ami-builder-image - This is a fork of archiso with some changes to automatically pull down my install script and do a few other things.
- ec2-packages - These are the sources for all the packages contained in the 'ec2' Pacman repository.
recent changes and news
I've added images for eu-west-2, ca-central-1, and us-east-2 as of today's release!
The paravirtual images aren't booting properly in some regions, apparently because PV-GRUB is failing to load /boot/grub/menu.lst.
Since PV is deprecated by AWS anyway, I'm going to stop making paravirtual AMIs from now on.
Whoops. I didn't notice my GPG key was going to expire on January 2nd. My bad. I've updated the keypair and submitted it to pgp.mit.edu as well as created a new ec2-keyring package. Unfortunately to install the new keyring package you need to manually unbreak your Pacman keyring by fetching the updated key directly from the keyserver:
# pacman-key -r A7B30DB9 gpg: requesting key A7B30DB9 from hkp server pool.sks-keyservers.net gpg: key A7B30DB9: "Steven Noonan <email@example.com>" 5 new user IDs gpg: key A7B30DB9: "Steven Noonan <firstname.lastname@example.org>" 42 new signatures gpg: 3 marginal(s) needed, 1 complete(s) needed, PGP trust model gpg: depth: 0 valid: 1 signed: 7 trust: 0-, 0q, 0n, 0m, 0f, 1u gpg: depth: 1 valid: 7 signed: 66 trust: 1-, 0q, 0n, 6m, 0f, 0u gpg: depth: 2 valid: 66 signed: 6 trust: 66-, 0q, 0n, 0m, 0f, 0u gpg: next trustdb check due at 2016-01-22 gpg: Total number processed: 1 gpg: new user IDs: 5 gpg: new signatures: 42 ==> Updating trust database... gpg: next trustdb check due at 2016-01-22
Once you do the above, you should be able to "pacman -Syu" as normal.
The Arch Linux AMIs have been moved to a different AWS account. This unfortunately means that all the S3 buckets had to be recreated, and when that happened the old S3 endpoints stopped working correctly. Since pacman doesn't know how to handle HTTP 301 redirects, you're going to have to manually update any existing Arch Linux instances you have. To do this, need to change your /etc/pacman.conf repo path from this:
[ec2] SigLevel = PackageRequired Server = https://s3.amazonaws.com/arch-linux-ami/repo/$arch
[ec2] SigLevel = PackageRequired Server = https://arch-linux-ami.s3.amazonaws.com/repo/$arch
Be sure to force an update of the package databases once you're done:
# pacman -Syy
I've also stopped producing AMIs for the Beijing, China (cn-north-1) region for the moment. If someone has an account in that region and wishes to produce the AMIs for me there, please get ahold of me and we'll work on making it happen.
I've added AMIs for the Beijing, China (cn-north-1) region.
We're now up to Linux 3.15.2. I've removed xen-fbfront from the initramfs, because the module was causing 30-second boot delays:
[ 2.050081] tsc: Refined TSC clocksource calibration: 2793.267 MHz [ 6.370066] xenbus_probe_frontend: Waiting for devices to initialise: 25s...20s... [ 12.390306] random: nonblocking pool is initialized [ 16.370070] 15s...10s...5s...0s... [ 31.371241] xenbus_probe_frontend: Timeout connecting to device: device/vfb/0 (local state 3, remote state 1)
The module is not required for an instance to boot correctly, so it can be removed from the initramfs. If you are running an AMI older than the 2014.06.27 release and would like to improve your instance's boot time, you can prune the module yourself:
[root@ip-10-0-155-25 ~]# grep fbfront /etc/mkinitcpio.conf MODULES="button ipmi-msghandler ipmi-poweroff virtio virtio-blk virtio-net virtio-pci virtio-ring xen-blkfront xen-fbfront xen-netfront xen-pcifront xen-privcmd hv_storvsc hv_balloon hv_vmbus hv_utils hv_netvsc ixgbevf" [root@ip-10-0-155-25 ~]# sed -ri 's/xen-fbfront //g' /etc/mkinitcpio.conf [root@ip-10-0-155-25 ~]# mkinitcpio -p linux-ec2 ==> Building image from preset: /etc/mkinitcpio.d/linux-ec2.preset: 'default' -> -k /boot/vmlinuz-linux-ec2 -c /etc/mkinitcpio.conf -g /boot/initramfs-linux-ec2.img -S autodetect ==> Starting build: 3.15.2-1-ec2 -> Running build hook: [base] -> Running build hook: [udev] -> Running build hook: [modconf] -> Running build hook: [block] -> Running build hook: [filesystems] -> Running build hook: [growfs] -> Running build hook: [keyboard] -> Running build hook: [fsck] ==> Generating module dependencies ==> Creating gzip initcpio image: /boot/initramfs-linux-ec2.img ==> Image generation successful [root@ip-10-0-155-25 ~]#
After removing xen-fbfront from mkinitcpio.conf's MODULES section, subsequent reboots will be 30 seconds shorter. Before:
[root@ip-10-0-155-25 ~]# systemd-analyze Startup finished in 32.530s (kernel) + 3.997s (userspace) = 36.528s
[root@ip-10-0-155-25 ~]# systemd-analyze Startup finished in 2.345s (kernel) + 2.092s (userspace) = 4.438s
The AMI now uses systemd's networkd, timesyncd, and resolved services. This makes the AMI have a significantly smaller footprint. Right now our biggest non-core packages are CUDA (in the GPU AMI) and cloud-init, which has a large dependency chain. I'd like to slim things even further, but I'll need to investigate how to do so.
We're up to Linux 3.13.7 for the ec2 kernel and 3.10.34 for the ec2-lts kernel. I didn't make a news post earlier, but kernels are now built with 'debug' and 'strip' options, which will create split-out debug information packages (i.e. linux-ec2-debug, linux-ec2-lts-debug). This is useful for tools like perf, oprofile, and systemtap. Note that the -debug packages are compressed with 'lrzip'. New AMI builds have lrzip preinstalled, but if you're running an instance based on one of the older AMIs, you will need to install lrzip before you can make use of the -debug packages.
New AMIs are being built right now and contain a couple changes:
- EBS root volumes are now automatically resized to fill the block device. You can take advantage of this feature by launching an instance with a root volume size larger than the snapshot.
- The resolv.conf file permissions are now 0644, allowing non-root users to resolve hostnames.
Geoff H. and David B. both reported an issue with the current AMI release. The /etc/resolv.conf permissions are set to 0600 rather than 0644, which means that non-root users cannot resolve hostnames to IP addresses. This is an unintentional regression, most likely caused by a default 'umask' change in some package. dhclient will create a new resolv.conf and copy it over any existing file, which preserves the target file's permissions. But if no such file exists, then the permissions of the source file are copied. Previously, this worked fine because the file was generated with 0644 permissions, but now it's being generated with 0600. I've implemented a fix for future AMI builds. In the meantime, if non-root users need to perform DNS requests in your instances, be sure to do 'chmod 0644 /etc/resolv.conf'.
A new ec2-pacman-mirrors package is available, and will provide your instances with optimal Arch Linux mirrors for your EC2 region. The upgrade path is as follows:
- Edit /etc/pacman.conf, change 'ec2' mirror URL to https://s3.amazonaws.com/arch-linux-ami/repo/$arch
- Run 'pacman -Sy ec2-pacman-mirrors'
New AMIs will be published very shortly which use the new mirror list and point to the new EC2 package repository.
I've added some links to this page, which are the complete set of files needed to do an EC2 image build. This does not include the AMI registration process, however. The tools Amazon provides for HVM AMI registration are still under NDA at the moment, and the bits necessary to do that are included in my AMI registration tools. So I can't make those public right now. The process itself can be replicated relatively easily, though:
- Build your VM image using the build-backend and builder-image repos above. PXE is what I use, but you could just as easily make it into an ISO or something. If you intend to do an S3-backed AMI, you will need to make the image no larger than 10GB (I use 8GB).
- Trim the image down (I do a 'mount -o loop,discard' on the image, then 'fstrim' the mount point, making the image into a sparse file).
- Tarball the image (tar cSzf, S to preserve the sparseness).
- Upload the tarball to S3.
- In each region, launch an instance and attach an empty 8GB EBS volume to them.
- On each of those instances, download the tarball and extract with 'tar xSf'.
- Use 'ddpt', an enhanced dd which pays attention to the sparseness of the image, to copy the raw image file into the EBS device. I use "ddpt if=<imagefile> of=/dev/xvdf bs=512 conv=sparse oflag=sparse,fsync". The sparseness aspect is important, because otherwise you're copying empty blocks onto the EBS device, which makes the snapshot take much longer, and is really just a waste of time. EBS volumes already read-as-zero, so there's no sense copying zero blocks.
- Detach the EBS volume and terminate the instances.
- Snapshot the EBS volume.
- Delete the volume (not needed now).
- Use ec2-register to create an AMI from the snapshotted volume.
I've started creating AMIs which have CUDA preinstalled. These are for the cg1.4xlarge instance type.
Nothing too exciting lately. Today's release has Linux 3.7.6.
New AMI releases, now with cloud-init. Thanks to Jeremy D for contributing his time and effort to making cloud-init work well on Arch Linux.
Released new AMIs, primarily for the new AWS region in Sydney, Australia (ap-southeast-2).
Today's AMIs are released. Nothing too fancy in this build: just updated packages, including linux-ec2 3.6.6-1.
I've added a new linux-ec2 package which contains a patched v3.6.2 kernel. There are a few major differences between this kernel and the Arch Linux stock kernel:
- Hangs on Xen fixed (patches from 3.6.3 stable-queue).
- CONFIG_PREEMPT_VOLUNTARY instead of CONFIG_PREEMPT, this will allow for better scheduling as a domU.
- CONFIG_HZ=100 instead of CONFIG_HZ=300, this allows for better performance on many-CPU instances, as there are fewer timer interrupts to preempt other tasks.
- Many drivers removed, particularly those that didn't make sense for running in an EC2 instance. I've left drivers for my own hardware so I can experiment with it as a dom0 kernel as well. The kernel size is roughly half the stock Arch Linux kernel due to the stripped drivers.
I am also building new AMIs right now, and am beating the i386 AMIs into working order. Once done I'll publish the next release (which should be 2012.10.21). Once it's available, it will show in the tables above.
Do not upgrade HVM instance kernels to anything between 3.6.0 and 3.6.2 inclusive. You must wait for 3.6.3 or else your instance will not boot. We're currently waiting on this patch to be integrated into the mainline stable tree. This is also why I am probably not doing an AMI release this week, as the HVM AMIs would be totally broken.
I've also taken a look at building i386 (well, i686) AMIs. I'm not really sure that it's worth the effort. Nobody really uses 32-bit AMIs anymore, and we'd need to fork the kernel just to make it happen. For now, i686 is on ice.
These AMIs are as close to a "vanilla" install as I can make them without making them functionally impaired on EC2. But here's the complete list of differences between the EC2 builds and a stock install:
High performance kernel specifically for EC2, including paravirtualization support on i386 and x86_64 AMIs, and more Xen-friendly process scheduling.
Kernel modules included in initrd, some of which are relevant outside of EC2 contexts (e.g. if you want to run the image in a non-EC2 environment such as KVM or Hyper-V):
- KVM: virtio virtio-blk virtio-net virtio-pci virtio-ring
- Xen: xen-blkfront xen-netfront xen-pcifront xen-privcmd
- Hyper-V: hv_storvsc hv_balloon hv_vmbus hv_utils hv_netvsc
- IPMI (e.g. EC2 reboot request): button ipmi-msghandler ipmi-poweroff
- EC2 enhanced networking SR-IOV driver: ixgbevf
Extra packages installed: audit, cloud-init, ec2-keyring, ec2-pacman-mirrors, irqbalance, lrzip, openssh, rng-tools, rsync, systemd-sysvcompat
Added an extra package source for ec2-specific packages. The repository currently contains numerous packages useful on EC2. You can view the list of packages by doing 'pacman -Sy; pacman -Ss | grep ^ec2'
Additional services enabled at boot: rngd, sshd, cloud-init, irqbalance, auditd, systemd-networkd, systemd-timesyncd, systemd-resolved
User's public key is pulled from the EC2 instance metadata service at startup, and added to /root/.ssh/authorized_keys
SSH configured with 'PasswordAuthentication no', enforcing public key authentication
pacman loads (and automatically lsigns) the 'archlinux' and 'ec2' keyrings on the first boot (the latter keyring contains my public key used for package signing in the ec2 repo).
pacman mirror list is automatically selected at boot based on a list I created (based on rankmirrors run on instances in each region). These lists are provided by the package ec2-pacman-mirrors, which is in the ec2 repo.
dhclient is used instead of dhcpcd for robustness reasons. I found that dhcpcd gave up too quickly if it tried to do a DHCPREQUEST when the vif wasn't completely up, making the EC2 instance inaccessible.
dhclient is configured to retry forever, and request the following dhcp options: subnet-mask, broadcast-address, time-offset, routers, domain-name, domain-name-servers, host-name, interface-mtu, fqdn
/usr/bin/pinentry is symlinked to /usr/bin/pinentry-curses instead of the default pinentry-gtk, since gtk isn't available in this install and the primary access method is SSH.