This is a set of scripts for starting kubernetes jobs and interactive shells, given the necessary options on the command line.
WARNING: This is alpha-quality software, and may have bugs, and should not be used for production or mission-critical workloads
usage: kbatch [-h] [--version] [--file FILE] [--image IMAGE]
[--container-name CONTAINER_NAME] [--name NAME] [--cpu CPU]
[--memory MEMORY] [--disk DISK] [--cpu-limit CPU_LIMIT]
[--memory-limit MEMORY_LIMIT] [--disk-limit DISK_LIMIT]
[--time TIME] [--persistent-disk-name PERSISTENT_DISK_NAME]
[--volume-name VOLUME_NAME] [--mount-path MOUNT_PATH]
[--volume-read-write] [--preemptible] [--script SCRIPT]
[--retry-limit RETRY_LIMIT] [--partition PARTITION]
[--labels LABELS]
-- [cmd [args...]]
positional arguments:
-- cmd [args...] Command with arguments to run in the given container
Optional, if your container has a pre-defined command,
otherwise required or your pod/job will 'crash' as it
has no command to run. You can use 'sleep 6000' for
testing an environment, for example.
Also note that this command will be run using /bin/sh
so if your command includes shell-like syntax that you
do not want run in the shell, quote the command, for
example: 'perl -Mbignum=bpi -wle "print bpi(2000)"'
optional arguments:
-h, --help show this help message and exit
--version Show the current version of kbatch
--file FILE, -f FILE Config yaml file to use (optional)
--image IMAGE, -i IMAGE
Docker container image to run (required unless a yaml
is provided)
--container-name CONTAINER_NAME
Container name to use (optional)
--name NAME, -n NAME Name of the job (an autogenerated id will be added to
this name)
--cpu CPU CPU reservation (In CPUs: 100m (== 0.1), 4)
--memory MEMORY Memory reservation (In bytes: 1024, 1e6, 100M, 128Mi)
--disk DISK Disk reservation (In bytes: 1024, 1e6, 100M, 128Mi)
--cpu-limit CPU_LIMIT
CPU limit (In CPUs: 100m (== 0.1), 4)
--memory-limit MEMORY_LIMIT
Memory limit (In bytes: 1024, 1e6, 100M, 128Mi)
--disk-limit DISK_LIMIT
Disk limit (In bytes: 1024, 1e6, 100M, 128Mi)
--time TIME Time limit (minutes)
--persistent-disk-name PERSISTENT_DISK_NAME
Persistent disk name (required to use a
gcePersistentDisk)
--volume-name VOLUME_NAME
Persistent disk volume name (optional)
--mount-path MOUNT_PATH
Mount path for the persistent disk (optional, default
is /static)
--volume-read-write Mount the volume as read-write (default is read-only)
--preemptible, -p Allow scheduling on preemptible nodes
--script SCRIPT Execute a bash script from a file from within the job
before the command args if they are present
--retry-limit RETRY_LIMIT
The number of times a job will retry a failed pod until it
succeeds. If not specified, a job will be retried an
infinite amount of times if not successful
--partition PARTITION
The partition name (value of the partition label) of
the node pool
--labels LABELS A list of node selector key value pairs, e.g. "--
labels key1=value1 key2=value2
This command will run the given docker image or yaml configuration as a batch job through kubernetes, and returns the job id. It will fill in the appropriate fields for the command and args to run, the job name, and memory/cpu/disk reservation and/or limits, if given. Note that whether a job name is specified or not an auto-generated name is used (in Kubernetes) so that a job configuration can be started more than once for different instances.
Example job command line to calculate pi using perl:
kbatch -n kbatch-test-pi -i perl --time 100 'perl -Mbignum=bpi -wle "print bpi(2000)"'
This example job has the name of kbatch-test-pi, uses the image perl, has a maximum execution time of 100 minutes,
and runs the command perl -Mbignum=bpi -wle "print bpi(2000)" inside the container. This also demonstrates the
alternative quoting of the command and args to not parse the parenthesis in the shell.
Note that in order to use preemptible nodes with node taints, you should create a kubernetes node pool with the taint
of gke-preemptible as the key, true as the value, and NoSchedule as the effect. This will prevent jobs that are
not specified as preemptible from being scheduled on the preemptible nodes.
usage: klist
This command lists all currently running jobs. See: kubectl get jobs for advanced options.
usage: kpods
This command lists all currently running pods. See: kubectl get pods for advanced options. Note that kpods -w will
watch pods and show changes dynamically.
usage: kexec pod-id [-- cmd args ... (optional)]
This command will start a bash shell in the given pod, or run the given command if specified.
usage: kstatus job-id
This command lists detailed information for a given job. See: kubectl describe job for advanced options.
usage: kcancel job-id [additional job-ids]
This command will cancel one or more running jobs. See: kubectl delete for advanced options.
Note that this command can also be used to delete old information for completed jobs.
usage: klogs job-id [additional log options such as -f for tailing the logs]
This command will find one of the pods a job is running on and run kubectl logs <POD> <EXTRA_ARGS> on that pod,
appending any arguments passed to klogs
usage: krun [-h] --image IMAGE [--name NAME] [--env ENV] [--script SCRIPT]
-- [cmd [args...]]
positional arguments:
cmd [args...] Command with arguments to run in the given container
(optional)
optional arguments:
-h, --help show this help message and exit
--image IMAGE, -i IMAGE
Docker container image to run
--name NAME, -n NAME Name of the job (an autogenerated id will be added to
this name)
--env ENV Environment variable to set in the format
"VARIABLE=value"
--script SCRIPT Execute a bash script from a file from within the job
before the command args if they are present
This command will start the given container as an interactive job in kubernetes for shell commands or other interactive processes for R&D or debugging purposes. If no command or script is specified, this will default to a bash shell.
To install the local version for development or usage, run:
./setup.py install
Do note that you will need the Google Cloud SDK (for kubectl) installed to use this package.
To lint, run:
flake8
To run tests:
pytest -m pytest . -vvv
After developing, make sure to modify the current version in setup.py before merging. After merging, run the following commands:
git tag v{YOUR_VERSION}
git push --tags
Your version should be a semantic version (i.e. 1.2.1)