Skip to main content Link Search Menu Expand Document (external link)

Staff Advice - Implementation

This page does not contain any additional design requirements or specifications.

Tips on Database Architecture

Whenever you’re thinking about storing lists or maps of things in Datastore, it may help to think about how you can “flatten” your data structure - this might help with security, efficiency, and/or code complexity! As an example: let’s say we have a list of restaurants, where each restaurant has a list of menu items. Let’s say I want to figure out whether a particular restaurant has a specific menu item, and we’re using a datastore-like backend (a key-value store).

Consider these two approaches to representing this information in some key-value store:

// Approach A
    "movies": {
        "mcu": ["iron man", "the incredible hulk", "iron man 2", "thor", "captain america", "the avengers"],
        "dceu": ["batman"],
        "pixar": ["turning red"]

// Approach B
   "movies/mcu": ["iron man", "the incredible hulk", "iron man 2", "thor", "captain america", "the avengers"],
   "movies/dceu": ["batman"],
   "movies/pixar": ["turning red"]

Both of these represent the same data, except instead of storing all of this information in one place, we’re flattening this data out across multiple entries in our key-value store. In approach B, all I need to do is to index into the datastore at the string value I construct (e.g. in this case, "movies/pixar"), and I can retrieve all of the movies that are affiliated with Pixar without having to retrieve a bunch of other data that was previously stored at the same location!

If efficiency is measured in read/write bandwidth (like it is in this project), then it’s much more efficient to represent our data in this flattened structure.

Minimizing Complexity

If you don’t need to store a certain piece of data, then don’t store it. Just recompute it (or re-fetch it, if it’s coming from datastore) “on-the-fly” when you need it. It’ll make your life easier!

Authenticated Encryption

In order to build in support for authenticated encryption, you may find it helpful to create a struct with two properties: (a) some arbitrary ciphertext, and (b) some tag (e.g. MAC, signature) over the ciphertext. Then, you could marshal that struct into a sequence of bytes, and store that in datastore. When loading it from datastore, you could (a) unmarshal it and check the tag for authenticity/integrity, and then decrypt the ciphertext and pass the plaintext downstream.

Checking for Errors

Throughout each API method, you’ll probably have several dozen error checks. Take a look at the following code for an example of good and bad practice when it comes to handling errors.

// This is bad practice: the error is discarded!
value, _ = SomeAPICall(...)

// This is good practice: the error is checked!
value, err = SomeAPICall(...)
if (err != nil) {
    return err;

When an error is detected (e.g. malicious tampering occurred), your client just needs to immediately return. We don’t care about what happens afterwards (e.g. program state could be messed up, could have file inconsistency, etc. - none of this matters). In other words, we don’t care about recovery – we solely care about detection.

You should almost never discard the output of an error: always, always check to see if error occurred!