Context
I have containerised a machine learning inference batch job that I am currently running on an on-prem ubuntu instance as a cron job. Basically, the cron job runs (there are a few other lines, but this is the core)
docker run $IMAGE
and this performs some inference using python.
Challenge
I want to migrate this job to the cloud using GCP infrastructure, but a bit confused about which service to use and how. Someone from Google customer team initially suggested Google Batch, but after some research I realised batch cannot support cron job at all. Then some forums/tutorials seem to support Google cloud scheduler. But after further reading it seems they trigger jobs if I hit an http endpoint. (That means hitting the http endpoint has to be established as a cron job with my on-prem infra? But why would anyone use that?)
So want to ask the SOF community on what’s the appropriate Google service to run a cron job, without myself managing an instance (and keeping it on 24*7).
For information, I have already pushed the $IMAGE to Google artefact registry. So all I need for google is to run the image with the following cron expression.
30 0 * * *
Further Details (Based on Comment)
I have a docker image in google artefact registry. The last line of the image is
# Run the script at container boot time.
CMD ["./run_manager.sh"]
This run_manager.sh performs the feature engineering, model training+inference using tensorflow and persists the result in an appropriate database. The whole thing takes about 20-25 minutes to run and then the container terminates.
I need to run this job daily, let’s say at 12.30 am every day. Currently I have this set up as a cron job on an on premise Ubuntu instance.
Requirement
Leverage Google cloud to perform this, without having to keep a VM on 24*7 just for one job. The behaviour I want is a Google cloud service to which I supply
- The artefact URL in the artefact registry
- The cron expression (or something equivalent)
and then GCP takes care of running the artefact/container according to the cron expression?
There is no web-service, no http-endpoint, no pub-sub here. Just a cron job.
3
Leverage Google cloud to perform this, without having to keep a VM on 24*7 just for one job.
In that scenario, you can think of using Instance Schedules through which you can start and stop your VM’s at a particular time everyday.
You can follow these steps to create instance schedule:
-
In the Google Cloud console, go to the VM instances page.
-
Go to VM instances
-
Click the Instance schedules tab at the top of the page.
Note: If you don’t see this option, create a VM instance first.
-
Click date_range Create schedule. The Create a schedule pane opens.
-
Enter a Name.
-
In the Region drop-down menu, select the location for this instance schedule.
-
Define when the instance schedule starts and stops any attached VM instances. If you need VM instances to start or stop at a specific time, schedule the operation 15 minutes earlier than needed. Make sure each start and stop operation are at least 15 minutes apart.
-
Use either the default Start time, Stop time, and Frequency fields or, if you want to configure a more complex schedule, use cron expressions.
Default fields: Enter a Start time, a Stop time, or both.
In the Start time field, type or click access_time to select the time to start VM instances.
In the Stop time field, type or click access_time to select the time to stop VM instances. In the Frequency drop-down menu at the bottom of the pane, select how often the Start time and Stop time repeat.
-
If you want to use Cron expressions:
To enable cron expressions, click the Use CRON expression toggle at the top of the pane.
Enter a Start CRON expression, Stop CRON expression, or both.
In the Start CRON expression field, enter a cron expression that describes when to start VM instances.
In the Stop CRON expression field, enter a cron expression that describes when to stop VM instances.
-
In the Time zone drop-down menu, select the time zone for the Start time and Stop time.
-
Click Submit.
According to the time specified while creating the Instance Schedule, your VM would start, the cron job will be run, and then finally the VM would be stopped.
For more details check this public document, this StackOverflow thread and this article.