Advanced Kadeploy

From Grid5000
Revision as of 18:39, 22 January 2022 by Jlawall (talk | contribs)
Jump to navigation Jump to search
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.
Note.png Note

This page is actively maintained by the Grid'5000 team. If you encounter problems, please report them (see the Support page). Additionally, as it is a wiki page, you are free to make minor corrections yourself if needed. If you would like to suggest a more fundamental change, please contact the Grid'5000 team.


Warning.png Warning

Please see the Environment creation guide, which gives automated mechanisms to build kadeploy environnements

What you need to know before starting

The first thing to understand is that by using kadeploy3, you will be running a command that attempts to remotely reboot possibly many nodes at the same time, and boot them over the network using configuration files hosted on a server.

What is an Environment?

Where we describe what exactly is image, kernel, initrd and postinstall

An environment in kadeploy3 is a set of file describing a fully functional Operating System. To be able to setup a Operating System, kadeploy3 needs at least 4 files in the most common cases

  1. An image
    • An image is a file containing all the Operating System files. It is a compressed archive (compressed tarball).
  2. A kernel file
    • Specifies the kernel file to boot. The full file path in the target system is expected.
  3. An initrd file (optional)
    • For the Linux systems, the initrd or initramfs file is the initial ramdisk used by the kernel as a preliminary system before the root filesytem is mounted. More information: Initrd on Wikipedia. The full file path in the target system is expected.
  4. A postinstall file (optional)
    • The postinstall file allows for adapting the deployed environment to specificities of each site and cluster, as well as setting up some platform-specific services.

Once you have this set of files, you can describe your environment to kadeploy3. This description represents an environment in the kadeploy3 sense.

Note: the kadeploy software also supports deploying environments made of binary disk images (dd format) but this is not supported on Grid'5000.

How can I make my own environment?

To create our own environment, there are two main ways:

  • One way is to deploy an existing environment, customize it and save it with tgz-g5k
  • The other one is to build the environment from a recipe, using kameleon, just like the Grid'5000 supported reference environments. See Environment creation.

Disk partitioning

Environments are deployed on the (first) local disk of each node (each Grid'5000 node is equipped with at least one hard drive, HDD or SSD). The following partitioning scheme is used:

Label Role
SWAP Linux swap
PROD Standard environment (default environment installed on nodes)
DEPLOY User's environment (when Kadeploy is used)
EFI For UEFI systems: UEFI system partition (store boot loaders)
TMP Remaining disk space made available in /tmp
Notes
  • Some clusters have more than one local disk. On some clusters, those disks can be reserved apart from the node itself. See Disk reservation to find out how to use them.
  • Kadeploy allows deployments on other partitions than DEPLOY or even creating a custom partitioning scheme. See below for details.

Search and deploy an existing environment

Search an environment

Grid'5000 maintains several reference environments directly available for deployment in all sites. These environments are based on the Debian, Ubuntu, or Centos Linux distribution.

For Debian, different variants of reference environments are offered. For Ubuntu and Centos, only environments with a minimal system are offered.

They are called reference environments because they can be used to generate customized environments and because they are provided and supported by the Grid'5000 technical team.

Kadeploy provides a registry of environments in each site, where reference environments are registered along with environments of users.

Note.png Note

Description of the reference environments: Template:Reference environments

For reference environments, associated filesystem images are stored in the /grid5000 directory of the frontend.

To deploy a registered environment, you must know its name as registered in the Kadeploy registry. This tutorial uses the debian11-x64-base environment.

You can also list all available environment in a site by using the kaenv3 command:

Terminal.png frontend:
kaenv3 -l

This command lists all public as well as your private environments.

We distinguish three levels of visibility for an environment:

  • public: Only administrators can register public environments. They are shown and used by default by the kaenv3 and kadeploy3 commands unless a user is specified (with -u user).
  • shared: Shared user environments. They are shown and used by the kaenv3 and kadeploy3 commands when a user is specified with -u user.
  • private: The environment is only shown and usable by the user the environment belongs to.

For example, a shared environment added by user user is listed this way:

Terminal.png frontend:
kaenv3 -l -u user

You can also look for a specific version with the --env-version version option. Most of the versions of the reference environments images files are available in /grid5000/images. The version number is the last part of the image file.

For instance: debian11-x64-min-2021092316.tar.zst is the image file of the debian11-x64-min reference environment version 2021092316.

Whenever you want to deploy this specific image on some nodes, use:

Terminal.png frontend:
kadeploy3 -f $OAR_NODEFILE -k -e debian11-x64-min --env-version 2021092316

Being able to reproduce a past experiment is a desirable feature. Therefore, you should always try to control as much as possible the environment the experiment is done in. Therefore, we will attempt to check that the environment that was chosen in the environment directory is the one available on a given cluster. On the cluster you would like to deploy, type the following command to print information about an environment:

Terminal.png frontend:
kaenv3 -p debian11-x64-base -u deploy

Adding -u deploy makes sure we get the description of the environment supported by the Grid'5000 staff, as deploy is the user owning them (one might have registred his own environment with the same, that would show up if no user is specified).

In theory, you should also check the post-install script. A post-install script adapts an environment to the site it is deployed on.

If everything seems ok, please proceed to the next step.

Make a job on a deployable node

By default, Grid'5000 nodes are running on the production environment, which already contains most of the important features and can be used to run experiments. This environment however imposes choices and has some limitations due to its general-purpose target. While you can gain the root privileges on it thanks to the sudo-g5k command, you cannot reboot for instance, as this ends your reservation. Deploying on the contrary allows to have a dedicated environment and to have full control over the machine (reboot, serial console, ...).

For this part of the tutorial, jobs made will be interactive (-I), of the deploy type (-t deploy), on only one machine (-l nodes=1) to do environment customization (we will give ourselves 3 hours with -l walltime=3), which gives us the following command, that will open a new shell session on the frontend node:

Terminal.png frontend:
oarsub -I -t deploy -l nodes=1,walltime=3

Indeed, when you submit a job of the deploy type, a new shell is opened on the frontend node and not on the first machine of the job as for standard jobs. When you exit from this shell, the job ends. The shell is populated with OAR_* environment variables. You should look at the list of available variables to get an idea of the information you can use to script deployment later. As usual, if the job is successful, you will get the name of the machine allocated to your job with:

Terminal.png frontend:
cat $OAR_FILE_NODES
Warning.png Warning

At the end of a reservation with the -t deploy option, the reserved nodes will be reboot on the standard environment and thus make them available for another job possibly of another user. Rebooting may take up to 15 minutes, so please mind your commands as they have an implicit cost.

Deploy a reference environment

To deploy your environment, you must discover the nodes you were allocated by OAR. One simple way of doing this is to look at the content of the file whose name is given by the $OAR_FILE_NODES environment variable (or $OAR_NODE_FILE, or $OAR_NODEFILE). To start the deployment, run the following command:

Terminal.png frontend:
kadeploy3 -e debian11-x64-base -m node.site.grid5000.fr

You can also just provide the nodes file to deploy on all the nodes of your job, with the -f option:

Terminal.png frontend:
kadeploy3 -e debian11-x64-base -f $OAR_FILE_NODES

In order to be able to connect to the node (as root), you must use the -k option and proceed by two ways:

  • You can either specify the public key that will be copied in /root/.ssh/authorized_keys on the deployed nodes:
Terminal.png frontend:
kadeploy3 -e debian11-x64-base -f $OAR_FILE_NODES -k ~/.ssh/my_special_key.pub
  • Or you can supply the -k option without argument. This will automatically copy your ~/.ssh/authorized_keys and replace the /root/.ssh/authorized_keys file on the deployed nodes.
Terminal.png frontend:
kadeploy3 -e debian11-x64-base -f $OAR_FILE_NODES -k

In our case, the node file contains only 1 node.

Once the kadeploy command was executed successfully, the deployed node runs the debian11-x64-base environment as their operating system. It will then be possible to tune this environment according to your needs.

Connect to the deployed environment and customize it

1. Connection

On reference environments managed by the Grid'5000 technical team, you can use the root account for log in with ssh (kadeploy checks that sshd is running before declaring a deployment successful). To connect to the node type:

Terminal.png frontend:
ssh root@node.site.grid5000.fr

In case this doesn't work, please take a look at the kadeploy section of the Sidebar > FAQ

2. Adding software to an environment

You can alter your environment (to add missing libraries that you need, or remove packages that you don't need ; to reduce the size of the image and speed up the deployment process ; etc.) using commands such as:

Terminal.png node:
apt-get update
apt-get upgrade
apt-get install list of desired packages and libraries
apt-get --purge remove list of unwanted packages
apt-get clean

Create a new environment from a node's customized operating system

We now need to save this customized environment, where you have a user account, to be able to use this account again each time you deploy it.
The first step to create an environment is to create an archive of the node's operating system you just customized. You can use tgz-g5k to extract a Grid'5000 environment tarball from a running node. It's usage is describe in the Environment creation tutorial page.

Advanced deployment options

Multisite deployment

In order to achieve a deployment on nodes from different sites, you can use the multiserver option of kadeploy, using the -M option.

Terminal.png frontend:
kadeploy3 -M -f file_with_all_nodes -e debian11-x64-std

Deploy on other partition of disk(s)

Kadeploy default handling of partitions may be too limited for some usages. One may need to use disks differently. Kadeploy offers several options to deploy on another existing partition of the primary disk, or if required to repartition disks entirely and/or use several disks (on nodes with many disks).

Deploy on partition labeled PROD or TMP

As seen earlier, kadeploy manages to label portions to identify them more easily, e.g. PROD, DEPLOY, or TMP. Kadeploy can be instructed to deploy on the PROD or TMP partition instead of the DEPLOY partition.

Because this kind of deployment will break some node standard operations after the end of your job, you must add to your oarsub command for the job creation the -t destructive option. This will cause the node to be completely reinstalled after your job.

Terminal.png frontend:
oarsub -t deploy -t destructive -l nodes=1,walltime=1 -p "cluster='hercule'" -I

Then you can deploy on PROD or TMP with the -p PROD or -p TMP option:

Terminal.png frontend:
kadeploy3 -e debian11-x64-nfs -f $OAR_NODEFILE -p PROD -k

Deploy on secondary disks

Kadeploy identifies disks by their id (disk0, disk1...). You can find them in the Reference API or in the Hardware pages:

Because this kind of deployment will break some node standard operations after the end of your job, you must add to your oarsub command for the job creation the -t destructive option. This will cause the node to be completely reinstalled after your job.

Terminal.png frontend:
oarsub -t deploy -t destructive -l nodes=1,walltime=1 -p "cluster='hercule'" -I

Then you can deploy on an secondary disk such as disk1 with the -b disk1 option:

Terminal.png frontend:
kadeploy3 -e debian11-x64-min -f $OAR_NODEFILE -b disk1 -r ext4 -k
Warning.png Warning

It will not work for reservable disks

Disks can also be handled differently by modifying the deployment automata, see below.

About the kernel and bootloader

As a reminder, a deployed environment may be booted either with kexec (e.g. at the end of the deployment) or by grub (e.g. after the deployment with a kareboot, or a call of the reboot command from the node itself, or if kexec is deactivated for the cluster because not supported).

By default, the bootloader installation step of kadeploy will install and configure grub using the grub commands provided by your environment. The grub configuration file will be generated with grub-mkconfig, which will boot by default the most recent kernel version available in the /boot directory of your environment. In case that version does not match the kernel version provided in the environment description, an error will be reported (Kernel file mismatch between grub and the environment description). This is a sanity check.

If the grub commands are not available in the deployed environment, as a fallback the grub commands provided in the deploy kernel will be used for the installation, and a very simple grub configuration will be generated (unlike when grub-mkconfig is used) that just use the kernel, initrd and kernel params from your environment description.

Note that having grub installed and configured with the commands that are provided in the deployed environment is the preferred way, because it will be isofunctional whenever the deployed system later has upgrades that impact the kernel or grub (or if the bootloader configuration command grub-mkconfig or update-grub is called).

Whenever relevant, by letting the postinstall create a empty file in the '/var/lib/kadeploy/bootloader/' directory of the deploy kernel (hence before kadeploy runs the bootloader installation step), it allows to trigger a change of the behavior of the bootloader installation step, as such:

  • if file exists: '/var/lib/kadeploy/bootloader/skip', do nothing (grub not installed nor configured)
  • if file exists: '/var/lib/kadeploy/bootloader/no-config', do not configure grub
  • if file exists: '/var/lib/kadeploy/bootloader/no-install', do not install grub
  • if file exists: '/var/lib/kadeploy/bootloader/no-grub-from-deployed-env', do not configure nor install using grub from your deployed env but from the deploy kernel
  • if file exists: '/var/lib/kadeploy/bootloader/no-grub-mkconfig-from-deployed-env', do not configure using grub from your deployed env but from the deploy kernel
  • if file exists: '/var/lib/kadeploy/bootloader/no-grub-install-from-deployed-env', do not install using grub from your deployed env but from the deploy kernel

See below the g5k-postinstall --bootloader option, which handles that, so that you just have to change the postinstall script in the environment description to change the behavior, and let it handle the creation of the before mentioned empty files.

Customizing the postinstalls

In Kadeploy3, postinstalls are scripts that are executed after the copy of the image file in order to customize site-specific or cluster-specific aspects. Since the beginning on 2018, on Grid'5000 the same postinstall script (called g5k-postinstall) is used for all reference environments (and is thus compatible with all supported Debian versions and distributions). That script takes parameters in order to define its behaviour (for example, to choose the style of network configuration to use).

Using g5k-postinstall

The source code for g5k-postinstall is available on gitlab.inria.fr. Its parameters at the time of writing are:

Terminal.png frontend:
g5k-postinstall --help
Usage: g5k-postinstall [options]

Options:
    -d, --debug                      Run in debug mode, with output to terminal
    -v, --version                    Print g5k-postinstall version
    -n, --net n1,n2,n3               Network configuration specification
    -f, --fstab f1,f2,f3             Filesystems configuration specification
    -r, --restrict-user MODE         User restriction mode
        --inittab PATTERN            Configure console in inittab
        --bootloader b1,b2,b3        Set the kadeploy bootloader step setup options
        --no-ref-api                 Do not use the Reference API
        --disable-hacks h1,h2,h3     Hacks to disable
        --no-guix                    Disable guix
        --disk-aliases               Enable disk aliases (e.g. /dev/disk0p1)

Valid tags for network specification:
  debian               write config in /etc/network/interfaces
  debian-bridged       write config in /etc/network/interfaces, with a bridge
                       setup (for the std env)
  netplan              write config in /etc/netplan/01-netcfg.yaml
                       (https://wiki.ubuntu.com/Netplan)
  redhat               write config in /etc/sysconfig/network-scripts/*
  traditional-names    use traditional NIC naming (e.g. eth0) instead of
                       predictable
  force-ref-api-names  force the use of the name provided in the reference API
                       (by default, the predictable name determined by the
                       kernel is used)
  hpc                  add support for HPC (eg InfiniBand) interfaces
Example: --net debian-bridged,traditional-names,hpc

Valid tags for filesystems configuration in fstab:
  nfs            include generic NFS mounts (/home with autofs, /grid5000)
  no-autofs      do not use autofs, just mount the user's NFS directory
  no-uuid        do not use filesystem UUID to identify the partitions, use the
                 block device filenames
  custom         include custom mounts for custom partitionning. Need fstab
                 file included in postinstall archive.
Example: --fstab nfs

Valid modes for user restriction:
  std         if deployed on production partition, restrict to root,oar. else,
              restrict to the current user (see below)
  current     restrict to root and the user having currently reserved the node
  login:jdoe  restrict to a specific login (e.g. jdoe)
  none        no restriction (DEFAULT)

Inittab option:
  Needed for non-systemd systems.
  Example of pattern: s0:12345:respawn:/sbin/agetty -L SPEED TTYSX vt100
  Where SPEED and TTYSX are replaced by g5k-postinstall using information
  retieved from the parameters which are passed in /proc/cmdline.

Valid tags for the kadeploy bootloader step:
  skip        request the bootloader step to do nothing
  no-config   request the bootloader step to skip the bootloader configuration
  no-install  request the bootloader step to skip the bootloader installation
  no-grub-from-deployed-env
              request the bootloader step to not use the grub commands from the
              deployed env to configure and install the bootloader (but write a
              basic grub.cfg and the grub-install from the deploy kernel), same
              as using both next 2 options
  no-grub-mkconfig-from-deployed-env
              request the bootloader step to not use grub-mkconfig from the
              deployed env to configure the bootloader (but write a basic
              grub.cfg)
  no-grub-install-from-deployed-env
              request the bootloader step to not use grub-install from the
              deployed env to install the bootloader (but use grub-install from
              the deploy kernel)
  no-uuid     do not use filesystem UUID to identify the root partition, use the 
              block device filename
  show-menu
              enable the grub menu on the console

No reference API option:
  Do not use the Reference API. This is useful during initial configuration of
  new clusters.

Disable hacks option:
  g5k-postinstall includes hacks that can optionally be disabled.
  Current hacks are:
  - oot-i40e        install the i40e driver on chifflot and chiclet (lille) and
                    grappe (nancy) on debian9 environment.
  - force-net-name  on several clusters, the predictable network interface name
                    can change depending on udev version. This hack forces the
                    predictable network interface name in such cases.
  - beegfs-gr520    configure beegfs shares on grcinq and grvingt (nancy).
Example: --disable-hacks oot-i40e,force-net-name

Disk aliases option:
  g5k-postinstall can optionally install udev rules to provide predictable disk aliases.
  Example of disk aliases automatically created by these udev rules:
    /dev/disk0   -> /dev/sdb
    /dev/disk0p2 -> /dev/sdb2
    /dev/disk1   -> /dev/nvme0n1
    /dev/disk1p4 -> /dev/nvme0n1p4
  This is useful because, starting from Linux 5.3, traditional block device names
  are non-deterministic: /dev/sda might not refer to the same disk on every boot.

An example environment description using g5k-postinstall is:

---
name: debian11-x64-min
version: 2021092316
arch: x86_64
description: debian 11 (bullseye) for x64 - min
author: pierre.neyron@imag.fr
visibility: private
destructive: false
os: linux
image:
  file: server:///grid5000/images/debian11-x64-min-2021092316.tar.zst
  kind: tar
  compression: zstd
postinstalls:
- archive: server:///grid5000/postinstalls/g5k-postinstall.tgz
  compression: gzip
  script: g5k-postinstall --net debian --disk-aliases
boot:
  kernel: "/vmlinuz"
  initrd: "/initrd.img"
  kernel_params: ''
filesystem: ext4
partition_type: 131
multipart: false

Things that you can do from there:

  • Use different parameters to change the behaviour of the postinstall. Example parameters for various situations are:
    • Debian min environment with traditional NIC naming: g5k-postinstall --net debian --net traditional-names
    • Debian min environment with predictable NIC naming: g5k-postinstall --net debian
    • Debian min environment with predictable disk aliases: g5k-postinstall --net debian --disk-aliases
    • Debian NFS environment (mount /home, setup LDAP, restrict login to user who reserved the node): g5k-postinstall --net debian --fstab nfs --restrict-user current
    • Debian big environment (NFS + setup HPC networks and mount site-specific directories): g5k-postinstall --net debian --net traditional-names --net hpc --fstab nfs --fstab site-specific
    • Use GPT label (e.g. /dev/disk/by-partlabel/KDPL_DEPLOY_disk0) instead of UUID to identify the root partition in the kernel command line: g5k-postinstall --bootloader no-uuid
    • Use GPT label (e.g. /dev/disk/by-partlabel/KDPL_DEPLOY_disk0) instead of UUID to identify the partitions in the fstab: g5k-postinstall --fstab no-uuid
    • RHEL/Centos style for network configuration: g5k-postinstall --net redhat --net traditional-names
    • Ubuntu 1710 or later: NetPlan for network configuration: g5k-postinstall --net netplan
    • Do not do any network configuration (useful for Gentoo), but force serial console settings: g5k-postinstall --inittab='s0:12345:respawn:/sbin/agetty -L SPEED TTYSX vt100'
  • Use a customized version of g5k-postinstall: after building a modified g5k-postinstall, just point the postinstalls/archive/ field to the new tar archive. See README.md and TechTeam:Postinstalls for details on g5k-postinstall internals.
  • Add an additional postinstall to execute after g5k-postinstall. That way, the additionnal postinstall can be written in any language (e.g. just a shell script) and complete what is already done by g5k-postinstall. See below.

Adding an extra postinstall

Modifying g5k-postinstall may for some purpose not be relevant (or overkill), for instance if actions are independent and can be executed after g5k-postinstall is run. In such a case, it is very easy to provide an additional postinstall that for instance may just be a shell script:

Assuming additional-postinstall.sh is a script located at the root of the /home/jdoe/public/debiantesting-x64-additional-postinstall.tar.gz archive, we just have to declare the additional postinstall in the environment description as follows:

---
author: John Doe
boot:
  initrd: /initrd.img
  kernel: /vmlinuz
description: debian testing with some customizations
destructive: false
filesystem: ext4
image:
  compression: gzip
  file: local:///home/jdoe/public/debiantesting-x64-custom.tar.gz
  kind: tar
multipart: false
name: debiantesting-x64-custom
arch: x86_64
os: linux
partition_type: 131
postinstalls:
- archive: server:///grid5000/postinstalls/g5k-postinstall.tgz
  compression: gzip
  script: g5k-postinstall --net debian --fstab nfs --restrict-user current
- archive: local:///home/jdoe/public//debiantesting-x64-additional-postinstall.tar.gz
  compression: gzip
  script: additional-postinstall.sh
version: 2020071009
visibility: shared

Both postinstalls will execute during the deployment, the one after the other.

Tuning the Kadeploy3 deployment workflow

kadeploy3 allows to fully modify the deployment workflow.

First of all you have to understand the different steps of a deployment. There are 3 macro-steps:

  1. SetDeploymentEnv: this step aims at setting up the deployment environment that contains all the required tools to perform a deployment ;
  2. BroadcastEnv: this step aims at broadcasting the new environment to the nodes and writing it to disk;
  3. BootNewEnv: this step aims at rebooting the nodes on their new environment.

kadeploy3 provides several implementations for each of those 3 macro-steps. You can consult that list in the kadeploy3 page. In Grid'5000, we use the following steps by default in all our clusters:

  • SetDeploymentEnv -> SetDeploymentEnvUntrusted: use an embedded deployment environment
  • BroadcastEnv -> BroadcastEnvKascade: use the Kascade tool to broadcast the environment
  • BootNewEnv -> BootNewEnvKexec: the nodes use kexec to reboot (if it fails, a BootNewEnvClassical, classical reboot, will be performed)

Each one of these implementations is divided in micro steps. You can can see the name of those micro-steps if you use the kadeploy3 option --verbose-level 4. And to see what is actually executed during those micro-steps you can add the debug option of kadeploy3 -d

Terminal.png frontend:
kadeploy3 -f $OAR_FILE_NODES -k -e debian11-x64-base --verbose-level 4 -d > ~/kadeploy3_steps

This command will store the kadeploy3 standard output in the file ~/kadeploy3_steps. Lets analyse its content:

Terminal.png frontend:
grep "Time in" ~/kadeploy3_steps

This command will print on the terminal all the micro-steps executed during the deployment process, and the time spent for each execution. Here are the micro-steps that you should see:

  1. SetDeploymentEnvUntrusted-switch_pxe: Configures the PXE server so that this node will boot on an environment that contains all the required tools to perform the deployment,
  2. SetDeploymentEnvUntrusted-reboot: Sends a reboot signal to the node
  3. SetDeploymentEnvUntrusted-wait_reboot: Waits for the node to restart.
  4. SetDeploymentEnvUntrusted-send_key_in_deploy_env: Sends kadeploy's user's ssh public key into the node's authorized_keys to ease the following ssh connections,
  5. SetDeploymentEnvUntrusted-create_partition_table: Creates the partition table
  6. SetDeploymentEnvUntrusted-format_deploy_part: Format the partition where your environment will be installed. This partition is by default /dev/sda3
  7. SetDeploymentEnvUntrusted-mount_deploy_part: Mounts the deployment partition in a local directory.
  8. SetDeploymentEnvUntrusted-format_tmp_part: Format the partition defined as tmp (by default, /dev/sda5)
  9. SetDeploymentEnvUntrusted-format_swap_part: Format the swap partition
  10. BroadcastEnvKascade-send_environment: Sends your environments into the node and untar it into the deployment partition.
  11. BroadcastEnvKascade-manage_admin_post_install: Execute post installation instructions defined by the site admins, in general to adapt to the specificities of the cluster: console baud rate, Infiniband,...
  12. BroadcastEnvKascade-manage_user_post_install: Execute user defined post installation instructions to automatically configure its node depending on its cluster, site, network capabilities, disk capabilities,...
  13. BroadcastEnvKascade-send_key: Sends the user public ssh key(s) to the node (if the user specified it with the option -k).
  14. BroadcastEnvKascade-install_bootloader: Properly configures the bootloader
  15. BootNewEnvKexec-switch_pxe: Configure the PXE server so that this node will boot on the partition where your environment has been installed
  16. BootNewEnvKexec-umount_deploy_part: Umount the deployment partition from the directory where it has been mounted during the step 7.
  17. BootNewEnvKexec-mount_deploy_part: ReMount the deployment partition
  18. BootNewEnvKexec-kexec: Perform a kexec reboot on the node
  19. BootNewEnvKexec-set_vlan: Properly configure the node's VLAN
  20. BootNewEnvKexec-wait_reboot: Wait for the node to be up.

That is it. You now know all the default micro-steps used to deploy your environments.

Environment boot: grub vs. kexec

On most clusters, kadeploy boots the user environment using kexec from the deploy kernel, after the image deployment and the postinstall steps. With kexec, Linux serves as a bootloader for itself. This means that the node is not "cold" rebooted, thus the classical bootloader (grub) is not used in that "warm" reboot process.

Whenever the node will be rebooted after the deployment, it will however rather use the classical bootloader (grub) than kexec (unless kexec is also configured to do so in the user environment).

You may look at the kadeploy3 command output to see if kexec is indeed used at the end of the deployment.

If you think the use of kexec may introduce a bias and prefer a classical cold reboot at the end of the deployment, you can either change the kadeploy workflow (see below) or just use the --no-kexec option of the kadeploy3 command.

Adjusting timeout for some environments

Since kadeploy3 provides multiple macro-steps and micro-steps, its is important to detect when a step in failing its execution. This error detection is done by using timeout on each step. When a timeout is reached, the nodes that have not completed the given step are discarded from the deployment process.
The value of those timeouts varies from one cluster to another since they depend on the hardware configuration (network speed, hard disk speed, reboot speed, ...). All defaults timeouts are entered in the configurations files on the kadeploy3 server. But you can consult the default timeouts of each macro-steps by using the command kastat3

Terminal.png frontend:
kastat3 -I
 Kadeploy server configuration:
 Custom PXE boot method: PXElinux
 Automata configuration:
   hercule:
     SetDeploymentEnv: SetDeploymentEnvUntrusted,1,600
     BroadcastEnv: BroadcastEnvKascade,0,1000
     BootNewEnv: BootNewEnvKexec,0,180; BootNewEnvHardReboot,0,900
   nova:
     SetDeploymentEnv: SetDeploymentEnvUntrusted,1,600
     BroadcastEnv: BroadcastEnvKascade,0,1000
     BootNewEnv: BootNewEnvKexec,0,150; BootNewEnvHardReboot,0,600
 ...


kadeploy3 allow users to change timeouts in the command line. In some cases, when you try to deploy an environment with a large tarball or with a post-install that lasts too long, you may get discarded nodes. This false positive behavior can be avoided by manually modifying the timeouts for each step at the deployment time.

For instance, in our previous example, the timeout of each steps are:

  • SetDeploymentEnvUntrusted: 143
  • BroadcastEnvKascade: 111
  • BootNewEnvKexec: 33

You can increase the timeout of the second step to 1200 seconds with the following command:

Terminal.png frontend:
kadeploy3 -e my_big_env -f $OAR_FILE_NODES -k --force-steps "SetDeploymentEnv|SetDeploymentEnvUntrusted:1:450&BroadcastEnv|BroadcastEnvKascade:1:1200&BootNewEnv|BootNewEnvClassical:1:400"

Set Break-Point during deployment

As mentioned in the section above, a deployment is a succession of micro steps that can be consulted and modified.
Moreover, kadeploy3 allows user to set a break-point during deployment.

Examples
  • Breakpoint before the postinstall is run:
Terminal.png frontend:
kadeploy3 -f $OAR_FILE_NODES -k -e debian11-x64-base --verbose-level 4 -d --breakpoint BroadcastEnvKascade:manage_user_post_install

This command can be used for debugging purpose. It performs a deployment with the maximum verbose level and it asks to stop the deployment workflow just before executing the manage_user_post_install micro-step of the BroadcastEnvKascade macro-step. Thus you will be able to connect in the deployment environment and to manually run the user post install script to debug it.

  • Breakpoint before rebooting to the deployed environment:
Terminal.png frontend:
kadeploy3 -f $OAR_FILE_NODES -k -e debian11-x64-base --verbose-level 4 -d --breakpoint BootNewEnv

Stops before rebooting to the deployed env. Mind that rebooting the machine will boot again the kadeploy kernel, not the deployed system.

Warning.png Warning

At the current state of kadeploy3, it is not possible to resume the deployment from the break-point step. Thus you will have to redeploy you environment from the first step. This feature may be implemented in future version of kadeploy3.

Modify the deployment workflow with custom operations

In Kadeploy3, we can easily customize the deployment's automata. It's possible to add custom pre, post or substitute operations to each steps. In a custom operation it's possible to: send a file, execute a command or run a script.

This feature in explained in Kadeploy3's documentation (available on Kadeploy3's website) in the section 4.2.2, Use Case 10 and 4.7.

This is illustrated in the following sub-sections.

Note.png Note

When running a custom script, Kadeploy will export different variables, you can get a list of them by running kadeploy3 -I.
A description of each of this variables is available in Kadeploy3's documentation (on Kadeploy3 website) in the section 4.4

Format additional disks

In this example, we will add some custom operations to the deployment workflow: our nodes have two additional hard disks and we want them to be formated during the deployment process.

We want to a new partition scheme such as:

  • classical grid5000 partitioning on sda
  • data1 ext4 on sdb1
  • data2 ext2 on sdc1

The three following sections describe how to perform such an operation.

1. Make the reservation in destructive mode

First of all, when you do your reservation, you must tell to OAR that it should redeploy the node entirely after the reservation with the -t destructive parameter:

Terminal.png frontend:
oarsub -t deploy -t destructive -l nodes=1,walltime=2 -p "cluster='hercule'" -I
2. Describe the custom operations

After that you have to create a file that describe the custom operations you want to be performed during the deployment. In our example we will first repartition the additional disks (using parted) and then format them (using the script format.sh).

  • The operation description file (let's say custom-partitioning.yml) should look like something like this:
---
# Our custom steps should be performed during the SetDeploymentEnv macro-step
SetDeploymentEnvUntrusted:
  # Custom partitioning step, done after the create_partition_table micro-step
  # In the sample this step is exploded in 4 steps but it can be done in 1 using a single parted command
  create_partition_table:
      post-ops:
        # We send a file on the node
        - action: send
          file: sdb.parted
          # The variable $KADEPLOY_TMP_DIR will be substitued by kadeploy
          destination: $KADEPLOY_TMP_DIR 
          name: send_partition_map_sdb
        # Then we execute the parted command using the previously sent file
        - action: exec
          name: partitioning_sdb
          # The variable $KADEPLOY_TMP_DIR will be substitued by kadeploy
          command: parted -a optimal /dev/sdb --script $(cat $KADEPLOY_TMP_DIR/sdb.parted)
        # Same operation for the second disk
        - action: send
          file: sdc.parted
          destination: $KADEPLOY_TMP_DIR 
          name: send_partition_map_sdc
        - action: exec
          name: partitioning_sdc
          command: parted -a optimal /dev/sdc --script $(cat $KADEPLOY_TMP_DIR/sdc.parted)
  # Custom format step, done after the format_deploy_part micro-step
  format_deploy_part:
      post-ops:
        # We run the script contained in the file 'format.sh'
        - action: run 
          name: format_disks
          file: format.sh
  • The file sdb.parted will look like something like this:
mklabel msdos
u GB mkpart primary ext4 0% 100%
align-check optimal 1
  • The file sdc.parted will look like something like this:
mklabel msdos
u GB mkpart primary ext2 0% 100%
align-check optimal 1
  • The file format.sh will look like something like this:
#!/bin/sh
set -e
# formating /dev/sdb
mkfs -t ext4 -b 4096 -O sparse_super,filetype,resize_inode,dir_index -q /dev/sdb1
# formating /dev/sdc
mkfs -t ext2 -b 4096 -O sparse_super,filetype,resize_inode,dir_index -q /dev/sdc1
3. Run the deployment

Now you can deploy you environment with this custom operation:

Terminal.png frontend:
kadeploy3 -e debian11-x64-min -f $OAR_NODE_FILE -k --custom-steps ./custom-partitioning.yml
Warning.png Warning

In some cases you should increase the step timeout (for some long formatting for example) see Advanced_Kadeploy#Adjusting timeout for some environments for details.

Note: Both partitions are not mounted on boot. To mount those partitions you should do:

Terminal.png NODE:
mkdir -p /media/data1
Terminal.png NODE:
mkdir /media/data2
Terminal.png NODE:
mount /dev/sdb1 /media/data1
Terminal.png NODE:
mount /dev/sdc1 /media/data2

Use a custom partitioning scheme

Example 1: Deploy on the whole disk

In this example, we will modify the deployment workflow to deploy the system on a unique disk partition ( '/' on sda1 )

1. Make the reservation in destructive mode

As you will change partitioning of the disk, you must tell to OAR that it should redeploy the node entirely after the reservation with the -t destructive parameter:

Terminal.png frontend:
oarsub -t deploy -t destructive -l nodes=1,walltime=2 -I
2. Describe the custom operations

After that you have to create a file that describe the custom operations you want to be performed during the deployment. In this example we will create our custom partitioning scheme and bypass some steps that are not necessary to deploy the system on a unique partition.

  • The operation description file (let's say custom-partitioning.yml) should look like something like this:
---
# Our custom steps should be performed during the SetDeploymentEnv macro-step
SetDeploymentEnv:
  # Custom partitioning step that is substituted to the create_partition_table micro-step
  create_partition_table:
    substitute:
      # We send a file on the node
      - action: send
        file: map.parted
        # The variable $KADEPLOY_TMP_DIR will be substituted by Kadeploy
        destination: $KADEPLOY_TMP_DIR
        name: send_partition_map
      # Then we execute the parted command using the previously sent file
      - action: exec
        name: partitioning
        # The variable $KADEPLOY_TMP_DIR will be substituted by Kadeploy
        command: parted -a optimal /dev/sda --script $(cat $KADEPLOY_TMP_DIR/map.parted)
# Hack to disable useless steps
  format_tmp_part:
    substitute:
      - action: exec
        name: remove_format_tmp_part_step
        command: /bin/true
  format_swap_part:
    substitute:
      - action: exec
        name: remove_format_swap_part_step
        command: /bin/true


  • The file map.parted, which will be passed to parted, will look like this:
mklabel gpt
mkpart KDPL_SYSTEM_disk0 ext4 0% 100%
toggle 1 boot
align-check optimal 1
Note.png Note

Kadeploy identify the partitions using GPT label. The form of the label should be KDPL_name_disk. So the label KDPL_SYSTEM_disk1 will be use on a deployment

Terminal.png frontend:
kadeploy3 -a custom-env.yml -f $OAR_NODE_FILE -b disk1 -p SYSTEM -k --custom-steps custom-partitioning.yml
The default disk on Grid'5000 is disk1 and the partition is DEPLOY, so by default, kadeploy look for the label KDPL_DEPLOY_disk0

3. Customize the environment's postinstall

In order for our new partitions to be mounted at boot time we will modify the Grid'5000 postinstall files.

  • Create and go in your public directory:
Terminal.png frontend:
mkdir public/custom-postinstall && cd public/custom-postinstall
  • Then decompress the postinstall archive:
Terminal.png frontend:
tar xzf /grid5000/postinstalls/g5k-postinstall.tgz
  • Add your custom /etc/fstab file in this directory, named fstab:
PARTLABEL=KDPL_SYSTEM_disk0     /          ext4    defaults 1      2

When you will pass "--fstab custom" option to the postinstall, it will copy this file in /etc/fstab

  • Regenerate the postinstall archive:
Terminal.png frontend:
tar -czvf ~/public/g5k-postinstall-custom.tgz *
  • Make some cleanup:
  • Create the environment's description file (let's say custom-env.dsc) based on the reference one:
    • use kaenv3 -p debian10-x64-base to have an example of environment description.

Your custom-env.dsc should look like this:

--- 
name: custom-env
version: 1
arch: x86_64
description: Custom env based on Debian 10
author: me@domain.tld
visibility: shared
destructive: true
os: linux
image:
  file: server:///grid5000/images/debian10-x64-base-2020012812.tgz
  kind: tar
  compression: gzip
postinstalls:
- archive: http://public/~<login>/g5k-postinstall-custom.tgz
  compression: gzip
  script: g5k-postinstall --net debian --fstab custom
boot:
  kernel: "/vmlinuz"
  initrd: "/initrd.img"
filesystem: ext4
partition_type: 131
multipart: false
4. Run the deployment

Finally, we deploy our custom environment with your custom operations:

Terminal.png frontend:
kadeploy3 -a custom-env.dsc -f $OAR_NODE_FILE -p SYSTEM -k --custom-steps custom-partitioning.yml
Note.png Note

In some case you should increase the step timeout (for some long formatting for example) see Advanced_Kadeploy#Adjusting timeout for some environments for details.

Example 2: Deploy on multiple partitions

In this example, we will modify the deployment workflow: a different partition will be used for each of the /, /home, /opt and /tmp directories. Imagine that you want to make your own partitioning scheme like that:

Mount point Partition Disk space File System
swap SWAP 2G linux-swap
/ SYSTEM 18G ext4
/var VAR 30G ext4
/opt OPT 20G ext4
/tmp TMP everything else ext4

The four following sections describe how to perform such an operation.

1. Make the reservation in destructive mode

First of all, when you do your reservation, you must tell to OAR that it should redeploy the node entirely after the reservation with the -t destructive parameter:

Terminal.png frontend:
oarsub -t deploy -t destructive -l nodes=1,walltime=2 -I
2. Describe the custom operations

After that you have to create a file that describe the custom operations you want to be performed during the deployment. In our example we will first create apply our custom partitioning scheme, format the partition and the mount them.

  • The operation description file (let's say custom-partitioning.yml) should look like something like this:
---
# Our custom steps should be performed during the SetDeploymentEnv macro-step
SetDeploymentEnvUntrusted:
  # Custom partitioning step that is substituted to the create_partition_table micro-step
  create_partition_table:
    substitute:
      # We send a file on the node
      - action: send
        file: map.parted
        # The variable $KADEPLOY_TMP_DIR will be substituted by Kadeploy
        destination: $KADEPLOY_TMP_DIR 
        name: send_partition_map
      # Then we execute the parted command using the previously sent file
      - action: exec
        name: partitioning
        # The variable $KADEPLOY_TMP_DIR will be substituted by Kadeploy
        command: parted -a optimal /dev/sda --script $(cat $KADEPLOY_TMP_DIR/map.parted)
  # Custom format step, done after the format_deploy_part micro-step
  format_deploy_part:
    post-ops:
      # We run the script contained in the file 'format.sh'
      - action: run 
        name: format_partitions
        file: format.sh
  # Custom mount step, done after the mount_deploy_part micro-step
  mount_deploy_part:
    post-ops:
      # We run the script contained in the file 'format.sh'
      - action: run 
        name: mount_partitions
        file: mount.sh
Note.png Note

In order for Kadeploy to be able to perform the installation correctly, every partitions have to be mounted before the installation process which is done in the macro-step BroadcastEnv

  • The file map.parted will look like something like this:
mklabel gpt
u GB mkpart KDPL_SWAP_disk0 linux-swap 0% 2
u GB mkpart KDPL_SYSTEM_disk0 ext4 2 20
u GB mkpart KDPL_VAR_disk0 ext4 20 50
u GB mkpart KDPL_OPT_disk0 ext4 50 70
u GB mkpart KDPL_TMP_disk0 ext4 70 100%
toggle 2 boot
align-check optimal 1
align-check optimal 2
align-check optimal 3
align-check optimal 4
align-check optimal 5


Note.png Note

Kadeploy identify the partitions using GPT label. The form of the label should be KDPL_name_disk. So the label KDPL_SYSTEM_disk1 will be use on a deployment

Terminal.png frontend:
kadeploy3 -a custom-env.yml -f $OAR_NODE_FILE -b disk1 -p SYSTEM -k --custom-steps custom-partitioning.yml
The default disk on Grid'5000 is disk1 and the partition is DEPLOY, so by default, kadeploy look for the label KDPL_DEPLOY_disk0. The KDPL_SWAP_disk is also detected as the swap partition and will be formated by kadeploy when detected.

  • The file format.sh will look like something like this:
#!/bin/sh
set -e

mkfs_opts="sparse_super,filetype,resize_inode,dir_index"
ext4_blocksize="4096"

# / will be formated by Kadeploy since we will precise the -p SYSTEM option
# formating /var
mkfs -t ext4 -b ${ext4_blocksize} -O ${mkfs_opts} -q /dev/disk/by-partlabel/KDPL_VAR_disk0
# formating /opt
mkfs -t ext4 -b ${ext4_blocksize} -O ${mkfs_opts} -q /dev/disk/by-partlabel/KDPL_OPT_disk0
# formating /tmp
mkfs -t ext4 -b ${ext4_blocksize} -O ${mkfs_opts} -q /dev/disk/by-partlabel/KDPL_TMP_disk0
Note.png Note

When running a custom script, Kadeploy will export different variables, you can get a list of them by running "kadeploy -I".

  • The file mount.sh will look like something like this:
#!/bin/sh
set -e

# / will be mounted in ${KADEPLOY_ENV_EXTRACTION_DIR} by Kadeploy
# mount /var
mkdir ${KADEPLOY_ENV_EXTRACTION_DIR}/var
mount /dev/disk/by-partlabel/KDPL_VAR_disk0 ${KADEPLOY_ENV_EXTRACTION_DIR}/var/
# mount /opt
mkdir ${KADEPLOY_ENV_EXTRACTION_DIR}/opt
mount /dev/disk/by-partlabel/KDPL_OPT_disk0 ${KADEPLOY_ENV_EXTRACTION_DIR}/opt/
# mount /tmp
mkdir ${KADEPLOY_ENV_EXTRACTION_DIR}/tmp
mount /dev/disk/by-partlabel/KDPL_TMP_disk0 ${KADEPLOY_ENV_EXTRACTION_DIR}/tmp/


3. Customize the environment's postinstall

In order for our new partitions to be mounted at boot time we can modify the Grid'5000 postinstall files (this customization can also be done by adding another custom operation).

  • Create and go in a temporary directory:
Terminal.png frontend:
tmpdir=$(mktemp -d) && export tmpdir && pushd $tmpdir
  • Then decompress the postinstall archive:
Terminal.png frontend:
tar xzf /grid5000/postinstalls/g5k-postinstall.tgz
Note.png Note

We assume that the current shell is BASH, if not please replace the "export" instruction

  • Add your custom /etc/fstab file in this temporary directory, named fstab:
PARTLABEL=KDPL_SWAP_disk0       none          swap    sw       0      0
PARTLABEL=KDPL_VAR_disk0      /var          ext4    defaults 1      2
PARTLABEL=KDPL_OPT_disk0       /opt          ext4    defaults 1      2
PARTLABEL=KDPL_TMP_disk0       /tmp          ext4    defaults 1      2

/ will be added by Kadeploy since we will precise the -p SYSTEM option

  • Regenerate the postinstall archive:
Terminal.png frontend:
tar -czvf ~/g5k-postinstall-custom.tgz *
  • Make some cleanup:
Terminal.png frontend:
popd && rm -R $tmpdir
  • Create the environment's description file (let's say custom-env.yml) based on the reference one:
Terminal.png frontend:
kaenv3 -p debian11-x64-base -u deploy | sed -e "s/archive:.*$/archive: \/home\/${USER}\/g5k-postinstall-custom.tgz/" -e 's/public/shared/' > custom-env.yml

and customize the custom-env.yml file to suit your needs (especially your archive path):

--- 
name: custom-env
version: 1
description: Custom env based on Debian 10
author: me@domain.tld
visibility: shared
destructive: true
os: linux
image:
  file: server:///grid5000/images/debian11-x64-base-2021092316.tar.zst
  kind: tar
  compression: zstd
postinstalls:
- archive: /home/me/g5k-postinstall-custom.tgz
  compression: gzip
  script: g5k-postinstall --net debian --fstab custom
boot:
  kernel: "/vmlinuz"
  initrd: "/initrd.img"
filesystem: ext4
partition_type: 131
multipart: false
Warning.png Warning

Do not forget the --fstab custom option to g5k-postinstall.

4. Run the deployment

Finally, we deploy our custom environment with your custom operations:

Terminal.png frontend:
kadeploy3 -a custom-env.yml -f $OAR_NODE_FILE -p SYSTEM -k --custom-steps custom-partitioning.yml
Note.png Note

In some case you should increase the step timeout (for some long formatting for example) see Advanced_Kadeploy#Adjusting timeout for some environments for details.

Boot a custom kernel with kexec

In this example, we change some custom operations of the reboot workflow: our nodes will use kexec to start a custom kernel that we upload beforehand, instead of rebooting. This allows us to boot a custom kernel without deploying a whole system (just use the kareboot3 command, not the kadeploy3 one).

1. Make the reservation in destructive mode

First of all, when we do our reservation, we have to tell OAR that it must redeploy the node entirely after our reservation. For this, we use the -t destructive parameter:

Terminal.png frontend:
oarsub -t deploy -t destructive -l nodes=1,walltime=2 -p "cluster='hercule'" -I
2. Describe the custom operations

Next, we must create a file that describes custom operations to perform during the reboot step.

In our example, we first send kernel and initrd files to the nodes, then run kexec using them. A operation description file (here custom-kexec-reboot.yml) to perform those actions looks like this:

---
Simple:
    reboot:
        pre-ops:
            - action: send
              file: /home/me/custom_kernel.vmlinuz
              destination: $KADEPLOY_TMP_DIR
              name: send_custom_kernel
            - action: send
              file: /home/me/custom_kernel.initrd.img
              destination: $KADEPLOY_TMP_DIR
              name: send_custom_initrd
        substitute:
            - action: exec
              name: kexec
              # nohup to be sure the script exit correctly before the kexec start
              command: kexec -l $KADEPLOY_TMP_DIR/custom_krenel.vmlinuz --initrd=$KADEPLOY_TMP_DIR/custom_kernel.initrd.img && nohup /bin/sh -c 'sleep 1; systemctl kexec' 1>/dev/null 2>/dev/null </dev/null &
3. Run the reboot

Finally, we call kareboot3 with our custom operations:

Terminal.png frontend:
kareboot3 simple --custom-steps custom-kexec-reboot.yml

Skip the first reboot/kexec

If you are already on the deployment kernel (by using as breakpoint or rebooting with kareboot3 -r deploy_env) you may want to skip the first reboot/kexec. This can be done by masking the reboot/kexec via custom operations. You need to be in a destructive job to be able to use custom operations (oarsub -t deploy -t destructive)).Write a file with your custom operation, eg custom-kexec-reboot.yml:

---
SetDeploymentEnv:
    reboot:
        substitute:
            - action: exec
              name: nothing
              command: 'true'
    kexec:
        substitute:
            - action: exec
              name: nothing
              command: 'true'

Start the deployment with our custom operations:

Terminal.png frontend:
kadeploy3 -e debian11-x64-min -f $OAR_NODE_FILE --custom-steps custom-kexec-reboot.yml