GPU-accelerated video transcoding with FFmpeg on Cloud Run jobs

This tutorial describes how to transcode low-priority offline videos using Cloud Run jobs.

Prepare your application

To retrieve the code sample for use:

Clone the sample repository to your local machine:

git clone https://github.com/GoogleCloudPlatform/cloud-run-samples

Change to the directory that contains the Cloud Run sample code:
```
cd cloud-run-samples/jobs-video-encoding
```

Create Cloud Storage buckets

To store the videos for processing, and to save the results of encoding, create the following two Cloud Storage buckets:

Create a bucket to store videos before processing:
```
gcloud storage buckets create gs://preprocessing-PROJECT_ID \
  --location LOCATION
```
Replace the following:
- PROJECT_ID: your project ID.
- LOCATION: the Cloud Storage location.

Grant the service account access to read from this bucket:

gcloud storage buckets add-iam-policy-binding gs://preprocessing-PROJECT_ID \
  --member="serviceAccount:video-encoding@PROJECT_ID.iam.gserviceaccount.com" \
  --role="roles/storage.objectViewer"

Replace PROJECT_ID with your project ID.

Create a bucket to store transcoded videos after processing:
```
gcloud storage buckets create gs://transcoded-PROJECT_ID \
  --location LOCATION
```
Replace the following:
- PROJECT_ID: your project ID.
- LOCATION: the Cloud Storage location.

Grant the service account access to read from and write to this bucket:

gcloud storage buckets add-iam-policy-binding gs://transcoded-PROJECT_ID \
  --member="serviceAccount:video-encoding@PROJECT_ID.iam.gserviceaccount.com" \
  --role="roles/storage.objectAdmin"

Replace PROJECT_ID with your project ID.

Deploy a Cloud Run job

Create a Cloud Run job by using the Dockerfile in the sample repository and mounting the buckets that you created:

Navigate to the sample directory:

cd cloud-run-samples/jobs-video-encoding

Create an Artifact Registry if the default Cloud Run registry doesn't already exist:
```
gcloud artifacts repositories create cloud-run-source-deploy \
  --repository-format=docker \
  --location LOCATION
```
Replace LOCATION with the name of the location of the registry.
Build the container image:
```
gcloud builds submit \
  --tag LOCATION-docker.pkg.dev/PROJECT_ID/cloud-run-source-deploy/IMAGE_NAME \
  --machine-type E2-HIGHCPU-32
```
Replace the following:
- PROJECT_ID: your project ID.
- LOCATION:name of the location of the registry.
- IMAGE_NAME: name for the container image, for example: ffmpeg-image.
Cloud Run uses a larger machine type to reduce build time.

Deploy the job:

gcloud beta run jobs create video-encoding-job \
    --image LOCATION-docker.pkg.dev/PROJECT_ID/cloud-run-source-deploy/IMAGE_NAME \
    --region REGION \
    --memory 32Gi \
    --cpu 8 \
    --gpu 1 \
    --gpu-type nvidia-l4 \
    --no-gpu-zonal-redundancy \
    --max-retries 1 \
    --service-account video-encoding@PROJECT_ID.iam.gserviceaccount.com \
    --add-volume=name=input-volume,type=cloud-storage,bucket=preprocessing-PROJECT_ID,readonly=true \
    --add-volume-mount=volume=input-volume,mount-path=/inputs \
    --add-volume=name=output-volume,type=cloud-storage,bucket=transcoded-PROJECT_ID \
    --add-volume-mount=volume=output-volume,mount-path=/outputs

Replace the following:

PROJECT_ID: your project ID.
REGION: the name of the region. Note: This must be the same region that you have GPU quota for.
IMAGE_NAME: name for the container image, for example, ffmpeg-image.

If this is the first time you deployed from source in this project, Cloud Run prompts you to create a default Artifact Registry repository.

Run the job

To run the job, follow these steps:

Upload an example video to encode:

gcloud storage cp gs://cloud-samples-data/video/cat.mp4 gs://preprocessing-PROJECT_ID

Run the job:

gcloud run jobs execute video-encoding-job  \
    --region REGION \
    --wait \
    --args="cat.mp4,encoded_cat.mp4,-vcodec,h264_nvenc,-cq,21,-movflags,+faststart"

The entrypoint.sh file requires an input file, output file, and any arguments to send to FFmpeg.

Review the Cloud Run logs to make sure the video trancoded:

gcloud run jobs logs read video-encoding-job --region REGION

Download the transcoded video:

gcloud storage cp gs://transcoded-PROJECT_ID/encoded_cat.mp4 .