The vm universe facilitates an HTCondor job that matches and then lands a disk image on an execute machine within an HTCondor pool. This disk image is intended to be a virtual machine. In this manner, the virtual machine is the job to be executed.
This section describes this type of HTCondor job. See section 3.5.25 for details of configuration variables.
Different than all other universe jobs, the vm universe job specifies a disk image, not an executable. Therefore, the submit commands input, output, and error do not apply. If specified, condor_submit rejects the job with an error. The executable command changes definition within a vm universe job. It no longer specifies an executable file, but instead provides a string that identifies the job for tools such as condor_q. Other commands specific to the type of virtual machine software identify the disk image.
VMware, Xen, and KVM virtual machine software are supported. As these differ from each other, the submit description file specifies one of
or
or
The job is required to specify its memory needs for the disk image with vm_memory, which is given in Mbytes. HTCondor uses this number to assure a match with a machine that can provide the needed memory space.
Virtual machine networking is enabled with the command
And, when networking is enabled, a definition of vm_networking_type as bridge matches the job only with a machine that is configured to use bridge networking. A definition of vm_networking_type as nat matches the job only with a machine that is configured to use NAT networking. When no definition of vm_networking_type is given, HTCondor may match the job with a machine that enables networking, and further, the choice of bridge or NAT networking is determined by the machine’s configuration.
Modified disk images are transferred back to the machine from which the job was submitted as the vm universe job completes. Job completion for a vm universe job occurs when the virtual machine is shut down, and HTCondor notices (as the result of a periodic check on the state of the virtual machine). Should the job not want any files transferred back (modified or not), for example because the job explicitly transferred its own files, the submit command to prevent the transfer is
The required disk image must be identified for a virtual machine. This vm_disk command specifies a list of comma-separated files. Each disk file is specified by colon-separated fields. The first field is the path and file name of the disk file. The second field specifies the device. The third field specifies permissions, and the optional fourth specifies the format. Here is an example that identifies a single file:
If HTCondor will be transferring the disk file, then the file name given in vm_disk should not contain any path information. Otherwise, the full path to the file should be given.
Setting values in the submit description file for some commands have consequences for the virtual machine description file. These commands are
For VMware virtual machines, setting values for these commands causes HTCondor to modify the .vmx file, overwriting existing values. For KVM and Xen virtual machines, HTCondor uses these values when it produces the description file.
For Xen and KVM jobs, if any files need to be transferred from the submit machine to the machine where the vm universe job will execute, HTCondor must be explicitly told to do so with the standard file transfer attributes:
Any and all needed files that will not accessible directly from the machines where the job may execute must be listed.
Further commands specify information that is specific to the virtual machine type targeted.
Specific to VMware, the submit description file command vmware_dir gives the path and directory (on the machine from which the job is submitted) to where VMware-specific files and applications reside. One example of a VMware-specific application is the VMDK files, which form a virtual hard drive (disk image) for the virtual machine. VMX files containing the primary configuration for the virtual machine would also be in this directory.
HTCondor must be told whether or not the contents of the vmware_dir directory must be transferred to the machine where the job is to be executed. This required information is given with the submit command vmware_should_transfer_files. With a value of True, HTCondor does transfer the contents of the directory. With a value of False, HTCondor does not transfer the contents of the directory, and instead presumes that access to this directory is available through a shared file system.
By default, HTCondor uses a snapshot disk for new and modified files. They may also be utilized for checkpoints. The snapshot disk is initially quite small, growing only as new files are created or files are modified. When vmware_should_transfer_files is True, a job may specify that a snapshot disk is not to be used with the command
In this case, HTCondor will utilize original disk files in producing checkpoints. Note that condor_submit issues an error message and does not submit the job if both vmware_should_transfer_files and vmware_snapshot_disk are False.
Because VMware Player does not support snapshots, machines using VMware Player may only run vm jobs that set vmware_snapshot_disk to False. These jobs will also set vmware_should_transfer_files to True. A job using VMware Player will go on hold if it attempts to use a snapshot. The pool administrator should have configured the pool such that machines will not start jobs they can not run.
Note that if snapshot disks are requested and file transfer is not being used, the vmware_dir setting given in the submit description file should not contain any symbolic link path components, as described on the https://htcondor-wiki.cs.wisc.edu/index.cgi/wiki?p=HowToAdminRecipes page under the answer to why VMware jobs with symbolic links fail.
Here is a sample submit description file for a VMware virtual machine:
This sample uses the vmware_dir command to identify the location of the disk image to be executed as an HTCondor job. The contents of this directory are transferred to the machine assigned to execute the HTCondor job.
A Xen vm universe job requires specification of the guest kernel. The xen_kernel command accomplishes this, utilizing one of the following definitions.
This form of the xen_kernel command also requires further definition of the xen_root command. xen_root defines the device containing files needed by root.
Creating a checkpoint is straightforward for a virtual machine, as a checkpoint is a set of files that represent a snapshot of both disk image and memory. The checkpoint is created and all files are transferred back to the $(SPOOL) directory on the machine from which the job was submitted. The submit command to create checkpoints is
Without this command, no checkpoints are created (by default). With the command, a checkpoint is created any time the vm universe jobs is evicted from the machine upon which it is executing. This occurs as a result of the machine configuration indicating that it will no longer execute this job.
vm universe jobs can not use a checkpoint server.
Periodic creation of checkpoints is not supported at this time.
Enabling both networking and checkpointing for a vm universe job can cause networking problems when the job restarts, particularly if the job migrates to a different machine. condor_submit will normally reject such jobs. To enable both, then add the command
Take care with respect to the use of network connections within the virtual machine and their interaction with checkpoints. Open network connections at the time of the checkpoint will likely be lost when the checkpoint is subsequently used to resume execution of the virtual machine. This occurs whether or not the execution resumes on the same machine or a different one within the HTCondor pool.
Following the platform-specific guest OS installation instructions found at http://partnerweb.vmware.com/GOSIG/home.html, creates a VMware disk image.
While the following web page contains instructions specific to Fedora on how to create a virtual guest image, it should provide a good starting point for other platforms as well.
http://fedoraproject.org/wiki/Virtualization_Quick_Start
Job completion for a vm universe job occurs when the virtual machine is shut down, and HTCondor notices (as the result of a periodic check on the state of the virtual machine). This is different from jobs executed under the environment of other universes.
Shut down of a virtual machine occurs from within the virtual machine environment. A script, executed with the proper authorization level, is the likely source of the shut down commands.
Under a Windows 2000, Windows XP, or Vista virtual machine, an administrator issues the command
Under a Linux virtual machine, the root user executes
The command /sbin/halt will not completely shut down some Linux distributions, and instead causes the job to hang.
Since the successful completion of the vm universe job requires the successful shut down of the virtual machine, it is good advice to try the shut down procedure outside of HTCondor, before a vm universe job is submitted.
It is not uncommon for a vm universe job to fail to launch because of a problem with the execute machine. In these cases, HTCondor will reschedule the job and note, in its user event log (if requested), the reason for the failure and that the job will be rescheduled. The reason is unlikely to be directly useful to you as an HTCondor user, but may help your HTCondor administrator understand the problem.
If the VM fails to launch for other reasons, the job will be placed on hold and the reason placed in the job ClassAd’s HoldReason attribute. The following table may help in understanding such reasons.