Compute Options

Compute is the part of a cloud account that runs your code. Storage holds bytes, a database answers queries, and compute is where the employees API actually executes when a request arrives.

There are three shapes to choose from, and every provider offers all three under different names: a whole machine you look after, a container the platform runs for you, and a function the platform starts when something happens. They differ in how much you operate, how you pay, and what kind of work they suit. The question that separates them is not which is most modern. It is how your work arrives — steadily, in bursts, or as long jobs.

// The platform calls this method. Your code does not start a web server,
// does not own the schedule, and cannot assume the previous call ran here.
public class PhotoUploadedHandler
{
    private readonly IThumbnailStore _thumbnails;
    private readonly ILogger<PhotoUploadedHandler> _logger;

    public PhotoUploadedHandler(IThumbnailStore thumbnails, ILogger<PhotoUploadedHandler> logger)
    {
        _thumbnails = thumbnails;
        _logger = logger;
    }

    public async Task HandleAsync(string photoKey, Stream photo, CancellationToken ct)
    {
        var thumbnailKey = photoKey.Replace("photos/", "thumbnails/");

        // Writing the same key twice produces the same result, so a repeated
        // delivery is harmless. That property is not optional here.
        await _thumbnails.WriteAsync(thumbnailKey, photo, ct);

        _logger.LogInformation("Thumbnail written. Key={ThumbnailKey}", thumbnailKey);
    }
}

• There is no host, no port and no route. The platform owns the trigger, which is the real difference between a function and a small web service.
• The handler is written so that running it twice for one upload changes nothing. Event platforms generally promise to deliver an event at least once, not exactly once, so a retry after a timeout can call you again with the same photo. Code that appends a row or increments a counter would double it.
• Nothing is cached in a field between calls. The instance holding this object may be discarded at any point, so anything you want to keep has to go to storage or a database.
• The log line uses named placeholders rather than an assembled sentence. That keeps the employee-facing values as separate fields you can filter on, which the monitoring lesson later in this module builds on.

A workable default for a new service: package it as a container image and run it on a managed container platform. You get reproducible environments, the provider patches the host, scaling is a setting, and you have not painted yourself into an unusual runtime.

Reach for functions when work arrives as discrete events and finishes quickly — a thumbnail to generate, a message to process, a report to kick off. Reach for a virtual machine when something about the host is a stated requirement rather than a preference.

Most real systems use two of the three. The employees API on a container platform, with a function handling photo uploads, is a reasonable shape and it is the one this course assumes from here on.