Cluster and cloud execution¶
Cloud Support¶
Snakemake 4.0 and later supports execution in the cloud via Kubernetes. This is independent of the cloud provider, but we provide the setup steps for GCE below.
Google cloud engine¶
First, install the Google Cloud SDK. Then, run
$ gcloud init
to setup your access. Then, you can create a new kubernetes cluster via
$ gcloud container clusters create $CLUSTER_NAME --num-nodes=$NODES --scopes storage-rw
with $CLUSTER_NAME
being the cluster name and $NODES
being the number of cluster
nodes. If you intent to use google storage, make sure that –scopes storage-rw is set.
This enables Snakemake to write to the google storage from within the cloud nodes.
Next, you configure Kubernetes to use the new cluster via
$ gcloud container clusters get-credentials $CLUSTER_NAME
If you are having issues with authentication, please refer to the help text:
$ gcloud container clusters get-credentials --help
You likely also want to use google storage for reading and writing files. For this, you will additionally need to authenticate with your google cloud account via
$ gcloud auth application-default login
This enables Snakemake to access google storage in order to check existence and modification dates of files. Now, Snakemake is ready to use your cluster.
Important: After finishing your work, do not forget to delete the cluster with
$ gcloud container clusters delete $CLUSTER_NAME
in order to avoid unnecessary charges.
Executing a Snakemake workflow via kubernetes¶
Assuming that kubernetes has been properly configured (see above), you can execute a workflow via:
snakemake --kubernetes --use-conda --default-remote-provider $REMOTE --default-remote-prefix $PREFIX
In this mode, Snakemake will assume all input and output files to be stored in a given
remote location, configured by setting $REMOTE
to your provider of choice
(e.g. GS
for Google cloud storage or S3
for Amazon S3) and $PREFIX
to a bucket name or subfolder within that remote storage.
After successful execution, you find your results in the specified remote storage.
Of course, if any input or output already defines a different remote location, the latter will be used instead.
Importantly, this means that Snakemake does not require a shared network
filesystem to work in the cloud.
Currently, this mode requires that the Snakemake workflow is stored in a git repository. Snakemake uses git to query necessary source files (the Snakefile, scripts, config, …) for workflow execution and encodes them into the kubernetes job.
It is further possible to forward arbitrary environment variables to the kubernetes
jobs via the flag --kubernetes-env
(see snakemake --help
).
When executing, Snakemake will make use of the defined resources and threads to schedule jobs to the correct nodes. In particular, it will forward memory requirements defined as mem_mb to kubernetes. Further, it will propagate the number of threads a job intends to use, such that kubernetes can allocate it to the correct cloud computing node.
Executing a Snakemake workflow via Tibanna on Amazon Web Services¶
First, install Tibanna.
$ pip install -U tibanna
Set up aws configuration either by creating files ~/.aws/credentials
and ~/.aws/config
or by setting up environment variables as below (see Tibanna or AWS documentation for more details):
$ export AWS_ACCESS_KEY_ID=<AWS_ACCESS_KEY>
$ export AWS_SECRET_ACCESS_KEY=<AWS_SECRET_ACCESS_KEY>
$ export AWS_DEFAULT_REGION=<AWS_DEFAULT_REGION>
As an AWS admin, deploy Tibanna Unicorn to Cloud with permissions to a specific S3 bucket.
Name the Unicorn / Unicorn usergroup with the --usergroup
option.
Unicorn is a serverless scheduler, and keeping unicorn on the cloud does not incur extra cost.
One may have many different unicorns with different names and different bucket permissions.
Then, add other (IAM) users to the user group that has permission to use this unicorn / buckets.
$ tibanna deploy_unicorn -g <name> -b <bucket>
$ tibanna add_user -u <username> -g <name>
As a user that has been added to the group (or as an admin), set up the default unicorn.
$ export TIBANNA_DEFAULT_STEP_FUNCTION_NAME=tibanna_unicorn_<name>
Then, you can run as many snakemake runs as you wish as below, inside a directory that contains
Snakefile and other necessary components (e.g. env.yml
, config.json
, …).
$ snakemake --tibanna --default-remote-prefix=<bucketname>/<subdir> [<other options>]
In this mode, Snakemake will assume all input and output files to be stored in the specified remote location
(a subdirectory inside a given S3 bucket.)
After successful execution, you find your results in the specified remote storage.
Of course, if any input or output already defines a different remote location, the latter will be used instead.
In that case, Tibanna Unicorn must be deployed with all the relevant buckets (-b bucket1,bucket2,bucket3,...
)
to allow access to the Unicorn serverless components.
Snakemake will assign 3x of the total input size as the allocated space for each execution. The execution may fail
if the total input + output + temp file sizes exceed this estimate.
In addition to regular snakemake options, --precommand=<command>
option allows sending a command to execute before
executing each step on an isolated environment. This kind of command could involve downloading or installing
necessary files that cannot be handled using conda (e.g. the command may begin with wget
, git clone
, etc.)
To check Tibanna execution logs, first use tibanna stat
to see the list of all the individual runs.
$ tibanna stat -n <number_of_executions_to_view> -l
Then, check the detailed log for each job using the Tibanna job id that can be obtained from the first column
of the output of tibanna stat
.
$ tibanna log -j <jobid>
When executing, Snakemake will make use of the defined resources and threads to schedule jobs to the correct nodes. In particular, it will forward memory requirements defined as mem_mb to Tibanna. Further, it will propagate the number of threads a job intends to use, such that Tibanna can allocate it to the most cost-effective cloud compute instance available.
Cluster Execution¶
Snakemake can make use of cluster engines that support shell scripts and have access to a common filesystem, (e.g. the Sun Grid Engine). In this case, Snakemake simply needs to be given a submit command that accepts a shell script as first positional argument:
$ snakemake --cluster qsub -j 32
Here, -j
denotes the number of jobs submitted being submitted to the cluster at the same time (here 32).
The cluster command can be decorated with job specific information, e.g.
$ snakemake --cluster "qsub {threads}"
Thereby, all keywords of a rule are allowed (e.g. rulename, params, input, output, threads, priority, …). For example, you could encode the expected running time into params:
rule:
input: ...
output: ...
params: runtime="4h"
shell: ...
and forward it to the cluster scheduler:
$ snakemake --cluster "qsub --runtime {params.runtime}"
If your cluster system supports DRMAA, Snakemake can make use of that to increase the control over jobs.
E.g. jobs can be cancelled upon pressing Ctrl+C
, which is not possible with the generic --cluster
support.
With DRMAA, no qsub
command needs to be provided, but system specific arguments can still be given as a string, e.g.
$ snakemake --drmaa " -q username" -j 32
Note that the string has to contain a leading whitespace. Else, the arguments will be interpreted as part of the normal Snakemake arguments, and execution will fail.
Job Properties¶
When executing a workflow on a cluster using the --cluster
parameter (see below), Snakemake creates a job script for each job to execute. This script is then invoked using the provided cluster submission command (e.g. qsub
). Sometimes you want to provide a custom wrapper for the cluster submission command that decides about additional parameters. As this might be based on properties of the job, Snakemake stores the job properties (e.g. name, rulename, threads, input, output, params etc.) as JSON inside the job script (for group jobs, the rulename will be “GROUP”, otherwise it will be the same as the job name). For convenience, there exists a parser function snakemake.utils.read_job_properties that can be used to access the properties. The following shows an example job submission wrapper:
#!python
#!/usr/bin/env python3
import os
import sys
from snakemake.utils import read_job_properties
jobscript = sys.argv[1]
job_properties = read_job_properties(jobscript)
# do something useful with the threads
threads = job_properties[threads]
# access property defined in the cluster configuration file (Snakemake >=3.6.0)
job_properties["cluster"]["time"]
os.system("qsub -t {threads} {script}".format(threads=threads, script=jobscript))
Visualization¶
To visualize the workflow, one can use the option --dag
.
This creates a representation of the DAG in the graphviz dot language which has to be postprocessed by the graphviz tool dot
.
E.g. to visualize the DAG that would be executed, you can issue:
$ snakemake --dag | dot | display
For saving this to a file, you can specify the desired format:
$ snakemake --dag | dot -Tpdf > dag.pdf
To visualize the whole DAG regardless of the eventual presence of files, the forceall
option can be used:
$ snakemake --forceall --dag | dot -Tpdf > dag.pdf
Of course the visual appearance can be modified by providing further command line arguments to dot
.