
Note: pricing and features may have changed since publication. Check each provider’s documentation for current details.
The main alternatives to Google Cloud Run Jobs are AWS Batch, Azure Container Apps jobs, Render cron jobs, Modal, and Upsun Cloud task containers. Each one runs a container or function to completion and bills only while it runs.
They differ on four things: how long a run can last, whether failed runs retry automatically, how many runs execute in parallel, and how close the job sits to the application and data it works on.
A Cloud Run job runs a container to completion instead of serving requests. Its strengths:
Teams look for an alternative for a few common reasons:
| Option | Longest run | Automatic retries | Parallel runs | Scheduling | Billing |
| Cloud Run Jobs | 7 days (1 hour with GPUs) | Yes, 3 by default, up to 10 | Up to 10,000 tasks per job | Cloud Scheduler | Per vCPU-second and GiB-second |
| Upsun Cloud task containers | 1 day (1 hour by default) | No; the code that starts the task retries | 3 at once by default; more triggers queue | A cron in the app starts the task through the Upsun API | Per second while the task runs |
| AWS Batch | No limit by default; about 14 days on Fargate | Yes, 1 to 10 attempts | Array jobs of up to 10,000 child jobs | EventBridge | No Batch fee; you pay for EC2 or Fargate |
| Azure Container Apps jobs | Set per job (replica timeout) | Yes, retry limit per job (0 by default) | Parallel replicas per execution | Built-in cron and event triggers | Per second while the job runs (Consumption plan) |
| Render cron jobs | 12 hours | No; the next run is a new run | One active run per cron job | Built-in cron | Per second, $1 minimum per cron job per month |
| Modal | 24 hours per attempt | Yes, configurable | Fan-out across containers with .map() | Built-in Cron and Period schedules | Per second for CPU, memory, and GPU |
AWS Batch runs containerized jobs from a queue on EC2, EC2 Spot, or Fargate. It handles large workloads: an array job can have up to 10,000 child jobs, and each job can retry 1 to 10 times. Jobs have no timeout unless you set one, though jobs on Fargate shouldn't expect to run longer than 14 days. AWS Batch itself has no charge; you pay for the EC2 or Fargate capacity your jobs use.
The trade-off is setup. You define compute environments, job queues, and job definitions before the first job runs, and recurring jobs are scheduled through Amazon EventBridge.
Azure Container Apps jobs run containers that do their work and exit. A job can start manually, on a cron schedule, or from an event such as a message arriving in a queue. Jobs run in the same Container Apps environment as your apps, so they share networking and logging. On the Consumption plan, a job is billed at the active rate only while an execution runs.
The trade-off is configuration you own. Every job needs a replica timeout, and failed replicas are only retried if you set a retry limit.
Render cron jobs run a command on a schedule you write as a cron expression. Each run gets a fresh container that is removed when the command exits. Billing is per second of run time, with a minimum of $1 per cron job per month. Render guarantees that only one run of a cron job is active at a time, and you can trigger a run by hand from the dashboard.
The trade-off is scope. Render stops any run after 12 hours, cron jobs can't use a persistent disk, and a failed run isn't retried; the next scheduled run starts fresh. For longer or multi-step work, Render points to background workers or its Workflows product, whose tasks run for 2 hours by default and up to 24 hours.
Modal runs Python functions in the cloud, billed per second for CPU, memory, and GPU. A function can run on a Cron or Period schedule, retry on failure, and fan out across many containers with .map(). Timeouts default to 5 minutes and go up to 24 hours per attempt, which suits model training and other GPU-heavy batch work.
The trade-off is distance from the application. Jobs are written as Python functions with Modal's SDK, and they connect to your application's database over the network with their own credentials.
Upsun Cloud task containers run one command to completion inside the same environment as your application, then stop. A task is declared in .upsun/config.yaml next to your apps and services. Its relationships give it a direct connection to the environment's database and other services, so a reindex or an import works on the application's own data. Each run is billed per second while it runs, and it is recorded as an activity with a status and logs.
Because tasks are part of the project configuration, every active branch environment runs the task definitions from its own branch. A changed job can run in a branch environment before the branch is merged and deployed to production.
The trade-offs:
Job | Good fit | Why |
| Process thousands of files in parallel on AWS | AWS Batch | Array jobs of up to 10,000 child jobs, with retries |
| Start a job when a message lands in a queue on Azure | Azure Container Apps jobs | Built-in event triggers |
| Nightly cleanup for an app hosted on Render | Render cron jobs | Built-in cron, billed per second of run time |
| Batch inference or model training on GPUs | Modal | GPUs, up to 24 hours per attempt |
| Reindex, import, or migrate the application's own database | Upsun Cloud task containers | Runs next to the app's services and can be tested in a branch environment first |
| Work that runs longer than a day in one piece | Cloud Run Jobs or AWS Batch | 7 days on Cloud Run Jobs; no default limit on AWS Batch |
If your application already runs on Google Cloud and your jobs need multi-day runs or thousands of parallel tasks, Cloud Run Jobs may still be the right choice.
Cloud Run Jobs | Upsun Cloud task containers |
| Container image | A runtime type and a build hook; Upsun builds the task from your repository |
| Task timeout | run.timeout, up to 86,400 seconds |
| Retries (--max-retries) | Retry logic in the code that starts the task |
| Multiple tasks per execution | Separate runs, up to 3 at once by default |
| Cloud Scheduler | A cron in your application that starts the task |
| Environment variables and secrets | Project and environment variables, which can be marked sensitive |
| Connection to the database | A relationship to the service in the same environment |
| gcloud run jobs execute | upsun task:run <name>, or a POST to the task's run endpoint |
The example below moves a nightly reindex job. It defines the task, gives the application permission to start it, and adds a cron that starts it on production only.
tasks:
reindex:
type: "python:3.14"
hooks:
build: "pip install -r requirements.txt"
relationships:
database: {} # your existing PostgreSQL service
run:
command: "python jobs/reindex.py"
timeout: 3600To run the task on a schedule, add a cron to your application that starts it through the Upsun API, and grant the app permission with an authorizations entry. See Offload heavy jobs to task containers with cron for the full configuration.
To complete the move:
Can Upsun Cloud tasks run on Google Cloud infrastructure?
Yes. Upsun Cloud has regions on Google Cloud, as well as on AWS, Azure, and OVHcloud. You choose the region when you create a project, and tasks run in the same region as the rest of the project.
Can an Upsun Cloud task run my existing container image?
A task uses an Upsun runtime, set with type in the same way as an application, and Upsun builds it from your repository code. Moving a Cloud Run job usually means moving its build steps into the task's build hook.
How do you retry a failed Upsun Cloud task?
The code that started the task retries it. Each run is recorded as an activity with a status, so a trigger script or the application can check the result and start the task again. Write the job so that running it twice is safe.
What happens to a running task when you deploy?
A deploy stops any task running in that environment. The task receives SIGTERM, then SIGKILL a few seconds later. Save progress to a database or storage mount when SIGTERM arrives, and start the task again after the deploy.
How is an Upsun Cloud task billed?
Per second for the length of each run, for the CPU and memory you allocate to the task, at the same rates as application containers. No CPU or memory is billed between runs. You set a task’s resources with upsun resources:set or in the Console.