With SharePoint 2010, Microsoft introduced a subset of the full-trust code development model. This was called sandbox solutions and, on paper, the idea was great. Not fabulous, but great.
The thinking at the time was that since organizations were starting to move more and more of their on-premises workloads to hosting environments or even public clouds (pre-Office 365 era), something needed to be done so that SharePoint deployments littered with full-trust code could safely be lifted and shifted or migrated to a multitenant platform. Since full-trust code was, well, fully trusted, it wouldn't take more than a few hours for someone to deploy bad-behaving code that would crash the whole application pool of a given IIS, and possibly even consume all available virtual memory within the SharePoint server.
Sandbox solutions were introduced for both SharePoint 2010 and SharePoint Online. These were initially meant for SharePoint 2010 and were offered as a multitenant platform for organizations so that they could share a common platform, and were later ported to SharePoint Online as a means to govern what kind of custom code would be deployed to a shared platform.
The idea with sandbox solutions was that running code within a package could only access a very limited set of APIs and namespaces within the SharePoint Object Model. One limitation that was set, meant that all calls to methods within the SharePoint development models could only access elements under a single site collection by using the SPSite() object. This way developers couldn't access any structures or data above the site collection that was running the code. Access to direct file I/O was also prohibited, as well as access to directly call HTTP interfaces through SharePoint.
This approach worked well, on paper. But in reality, porting existing code from full-trust packages proved to be troublesome, time-consuming, and sometimes just impossible. Existing code often resorted to using certain methods and features of the SharePoint Object Model, so that a replacement within the SPSite() object was not simply available. Therefore, either big pieces of custom code had to be cut in order to move the platform to the cloud, or fully rewritten without having proper APIs.
As part of this ideology, Microsoft introduced the idea of resources. Each call from the code would consume a small piece of resources from a pool of resources, and administrators could designate pre-allocations for each site collection. When the customization exceeded its allotted resources, it would simply be shut down. Imagine having a graphical component running on your SharePoint site that exposes a tool for your users. Suddenly, the tool simply disappears from the page because it exceeded a hidden resource allocation that end users weren't aware of. Administrators would have to reallocate more resources or start digging through code to understand where and how the resources were spent and how to best optimize this piece of code for a small subset of users complaining loudly.
As such, sandbox solutions never really took off. A couple of years after introducing sandbox solutions in SharePoint Online, Microsoft quietly deprecated the ability to run custom .NET code within a sandbox solution package. This effectively rendered the whole model to a vehicle for provisioning XML-based artifacts for a given SharePoint site or site collection. Not much changed, as developers were now complaining (again) that developing solutions for SharePoint were merely pointless exercises in understanding archaic XML structures in Notepad.
From time to time we still see customer deployment in SharePoint Online happily running sandbox solutions. This is fine but in no way advisable for the future. A good approach is to export the sandbox solution files, rename the packages from .WSP to .ZIP files, and explore what's in the package. This way you can learn and understand what the solution truly implement, and how to best replace the solution.
