The catalog image generation process of einstein web recommendation is one of the worst possible designed processes i have ever seen. Therefore it would be a pleasure if you can share your experiences like:
- How have you configured your working catalog image generation process?
- What kind of hardware and how many web servers have you used to conquer this process?
- How do you update your images that contain custom fields (automated)?
- Have you customized the requests / s parameter on the backend of salesforce, if yes to what kind of request / s parameter?
Default Publish Process (How i have experienced it):
- More than 2000 Requests per Second to a Server [this is customizable, but low values like 50 req/s can still crash servers with an maximum of 1000 workers]
- Requests, that are adding up , for us more than 500k Requests in less than 5min :D:D:D This is like a DDOS-Attack
- You cannot set a maximum of concurrent requests
- No handling of responses from the server (he does not react with a decreasing request / s when the server replies to slow or with a closed connection).
- If you have a Product Catalog with multiple Languages the image of one product gets loaded as often as there are languages (even if the imagelink is not different and the field is not locale).
- The result is that the productcatalog items times languages requests will be made.
- If you insert your own catalog fields into a display it does not get updated when this value get changed.
- Displays only get updated when ProductName, ImageLink, Regprice or Salesprice get updated (WTF).
- When you want to update one particular item where you changed the "customfield" inside the image you have to publish ALL products again.
- I can only imagine how great this works with streaming updates and automated journeys that use einstein predictive content.
Thanks for reading, I am happy about any insights or advices.