Let's look at the security measures available in Cosmos DB. There are two methods for controlling access to your Cosmos database. Firstly, you can use IP address filtering. By default, this is disabled, so any IP address could potentially access your database. You can configure network and IP access through Firewall and virtual networks under settings. This allows you to specify either particular virtual networks access, or specified IP addresses.

Once connected, a request needs to be authenticated and authorized. This is done using keys: a primary and secondary key that are generated as part of the account creation. These are the read-write keys. There are also two read-only keys. Having two sets of keys allows you to regenerate one key in each pair, that is, rotating the keys while maintaining uninterrupted access to the database for users and applications.

It's definitely not ideal to use the master keys for all access to your Cosmos database. You can use resource tokens to grant access to specified resources within the database. A token is created when a user is given permission to access a resource. Once a user has been authenticated, say, via Azure Active Directory, a resource token is requested for that user from the Cosmos DB account and relayed back to the client. The client app can then use that token to access the specified Cosmos resources until the token expires. Tokens, by default, are only valid for an hour, but this can be extended to up to five hours. Once expired, a new token will be created for the client if access is still permitted.

Apart from keys and resource tokens, you can assign roles and permissions to users through the Cosmos DB account Access control function.

As with other Azure services, Cosmos DB records audit logs which can be viewed through Activity log. In addition to activity logs, you can enable diagnostic logging by going into Diagnostic settings under the monitoring menu. Diagnostic logs can be saved to a storage account, streamed to an event hub, or sent to Log analytics. From a security point of view, Query Runtime Statistics, which records details of queries, could hold valuable information when tracking down suspicious activity. You can also record Mongo requests and Data plane requests, which include the resource unit charge.