You saw in the Admin Tutorial how to both launch and monitor workflows that have been scheduled by a user. This guide provides a bit more details on this process.
In the SeqWare Pipeline system there are two ways to launch workflows. A user can use WorkflowLauncher to directly launch a workflow provided the workflow engine is setup on this host (e.g. it is an Oozie or Pegasus submission host). Alternatively, a user can “schedule” a workflow using WorkflowLauncher through a web service. Another process can monitor the web service and launch the workflows that have been scheduled. On the SeqWare VM/AMI this all happens on the same box. But it is possible to have the user scheduling a workflow, the SeqWare Web Service, and the workflow launcher for SeqWare Pipeline all can (and likely should) be on separate servers. This latter method for launching workflows is the preferred mechanism, especially in a production environment, since it allows the scheduling of workflows and the running to be decoupled (different machines, different users, different user roles, etc).
The setup and configuration of the Web Service and user command line tools are covered elsewhere. Here we detail the needed cron jobs running on the SeqWare Pipeline host that will query the Web Service, launch scheduled workflows, and monitor their progress.
One core limitation of SeqWare Pipeline is the lack of a single daemon for controlling workflow launching and monitoring. Instead the coordination of workflows happens via the SeqWare Web Service. Each workflow that gets scheduled by a user is associated with that users Web Service credentials. In order to launch that scheduled workflow the WorkflowLauncher needs to connect to the Web Service with the same credentials, find the workflow, and then launch it on the SeqWare Pipeline box. If workflows are scheduled by multiple accounts each account needs its own launcher and monitor cron job to periodically launch and monitor workflows.
Typically this is not a huge limitation since the number of SeqWare accounts responsible in a production environment is relatively limited. It does make it more difficult, though, when the number of SeqWare users is high and each must have their own distinct account to monitor and launch workflows.
Show table showing user launching and host interactions. In this example the user “Bob” logs into server1 and connects to the SeqWare Web Service running on server2 using the login “email@example.com”. He then schedules a workflow with WorkflowLauncher and specifies the host to run this workflow on with –host server2. On server2 another user account, “seqware-bob” runs the daemons documented below. They connect to the Web Service locally using the same “firstname.lastname@example.org” username and find workflows that have been scheduled to run on server2. These daemons launch the workflows on server2 and the resulting jobs run on the cluster connected to server2 as user “seqware-bob”, resulting in files associated with this workflow being owned by the user “seqware-bob”.
|Host - server1||Network||Host - server2|
|User Scheduling||Web Service Account||Host param||->||User Running||Web Service Account|
In our reference SeqWare environment, we typically schedule jobs and then launch them asynchronously via a cronjob.
A user will schedule workflow launches using a command similar to that below:
$ seqware workflow schedule --accession 1 --parent-accession 99 --ini workflow.ini --host `hostname --long`
Then in a cronjob we use the following command to launch scheduled jobs:
$ seqware workflow-run launch-scheduled
Note that in the first command, we allow jobs to be scheduled on a specific host. When we launch scheduled workflows, we check this value in order to determine whether a particular scheduled workflow should be launched on this host. Note that while we normally use a fully qualified hostname, any unique string can be used to designate a host for launching (for example on Amazon S3).
Since the engine that executes the workflow is separate from the SeqWare MetaDB, a separate process is used to propagate statuses between the workflow engine and MetaDB:
$ seqware workflow-run propagate-statuses
Once this is executed, workflow-run reports will reflect the updated status.
The SeqWare VM performs both of the above functions via a cronjob:
$ crontab -l * * * * * /home/seqware/crons/status.cron >> /home/seqware/logs/status.log $ cat /home/seqware/crons/status.cron #!/bin/bash source /home/seqware/.bash_profile seqware workflow-run launch-scheduled seqware workflow-run propagate-statuses --threads 10
This script runs every minute and uses the first command to launch workflows that have been previously scheduled while the second command is used to check the status of launched workflows.