0%
Forward Deployed Engineering
The Role
Discovery and Scoping
Building in the Field
Running Deployments
Enterprise Integration, SSO and APIs
Enterprise customers do not want another password. They want their employees to sign in with the identity provider they already run, they want access to stop the moment someone leaves and they want an auditor to see who could do what. Integration with identity, with the customer systems of record and with their security controls is often the longest part of a deployment and the part most likely to block launch. This lesson covers single sign on, user provisioning, role mapping, APIs and webhooks, and the network and security work that sits around them.
Why SSO matters to the customer
Single sign on lets users authenticate once with the corporate identity provider and reach every approved application. For the customer it means one place to enforce password rules and multi factor authentication, one place to remove access when someone leaves and a log of every sign in. Security teams at large companies often refuse any application that does not support it. For the FDE, SSO is usually the first integration the customer asks about and the first one they test.
SAML and OpenID Connect
Two protocols cover nearly every enterprise case. SAML is older and based on signed XML assertions passed through the browser. OpenID Connect is built on OAuth 2 and uses signed JSON tokens. Both follow the same idea: the application redirects the user to the identity provider, the provider authenticates them and sends back a signed statement of who they are, and the application checks the signature and starts a session. Support both if the product serves large companies, since many still run SAML only.
The parts of a SAML setup
A SAML integration needs a few values exchanged between the two sides. The application gives the customer its entity identifier and the URL that receives assertions. The customer gives back the identity provider metadata: its own entity identifier, its sign in URL and the certificate used to sign assertions. Many setups fail on small mismatches in these values, such as a trailing slash in a URL, so copy them exactly and prefer exchanging metadata files over typing values by hand.
Attributes and claims
The assertion or token carries attributes about the user: email, name, a stable user identifier and often group memberships. Agree with the customer which attributes will be sent and in what format. Use the stable identifier as the key for the user account, not the email, because emails change when people marry or the company rebrands. Groups are usually the basis for role mapping, covered later, so make sure the identity provider sends the groups that matter and not every group in the directory.
Testing with the customer
Testing SSO needs a real account in the customer identity provider, so schedule a working session with their identity administrator early. Start with a test user in a test group, confirm sign in, check the attributes received and test sign out. Then try the failure cases: a user not assigned to the application, an expired certificate, a clock that is a few minutes off. Each failure should produce a clear message, not a blank page or a generic error.
Sessions and sign out
Once the user is signed in, the application runs its own session. Decide how long sessions last and whether they end when the identity provider session ends. Many security teams ask for idle timeouts and a maximum session length. Single logout, where signing out of the identity provider signs the user out of every application, is supported unevenly and fragile, so short sessions with re authentication through the provider are often the more reliable way to meet the requirement.
Certificates and rotation
Identity provider signing certificates expire, typically every one to three years. When the customer rotates the certificate without telling the application, every sign in fails at once. Support more than one trusted certificate per connection so the new one can be added before the old one is removed, monitor expiry dates and alert weeks ahead. Write the rotation steps into the runbook the customer receives at handover.
Just in time accounts
Many applications create the user account the first time someone signs in through SSO, using the attributes in the assertion. This is called just in time provisioning. It is simple and works for small groups, but it has limits: accounts exist only after first sign in, so nobody can assign work to a new colleague in advance, and accounts are never removed when people leave. For that, the customer needs provisioning, covered in the next section.
Multiple identity providers
Large customers sometimes run more than one identity provider, after an acquisition or for contractors. The application must route each user to the right provider, usually by email domain on the sign in page, and must not let one provider assert users that belong to another. Keep each connection separate in configuration, and test that a user from one domain cannot sign in through the other connection.
Break glass access
Plan what happens when the identity provider is down or misconfigured. If every administrator account is SSO only, a broken connection locks everyone out, including the people who could fix it. Agree a small number of break glass accounts with strong passwords and multi factor authentication, stored securely and used only in emergencies, with every use logged and reviewed. Security teams usually accept this when it is documented and audited.
Multi factor authentication
Multi factor authentication belongs to the identity provider, not the application. When SSO is on, the application should not ask for a second factor of its own, because users will see two prompts and the customer loses a single place to set policy. Some customers want proof that a second factor was used, which OpenID Connect can carry as an authentication context claim. Check for it when the customer requires it, and document the behaviour when the claim is missing.
Enforcing SSO
Once SSO works, most customers want it enforced: no local passwords for their domain, so nobody can bypass the identity provider. Add a setting per customer that blocks password sign in for users in their email domains, keeping only the break glass accounts as exceptions. Turn it on after a period where both work, and tell users beforehand, since anyone who saved a password will be surprised.
Domain verification
Before a customer can claim an email domain for SSO or enforcement, prove they own it, usually by asking them to publish a DNS record with a value the application generates. Without verification, any customer could claim a domain they do not own and capture sign ins for users of another company. This check is small to build and a common question in security questionnaires.
Clear errors for users
Most SSO support requests come from users who see a confusing error. Write messages for the common cases in plain language: you are not assigned to this application, ask your IT team; your company requires sign in through its identity provider; the sign in link has expired, try again. Include a short reference code that support can look up in the logs. Good messages turn a ticket that takes an hour into one that takes a minute.
Staging connections
Ask the customer whether they have a test identity provider tenant. If they do, connect the staging environment to it, so certificate rotations, new groups and attribute changes can be tried there first. If they do not, use a separate application entry in their production provider for staging, with its own test users and groups. Testing identity changes directly against production is how a small change locks out a whole department.
Support SAML and OpenID Connect, exchange metadata files, key accounts on a stable identifier and test the failure cases with the customer identity administrator. Trust more than one certificate so rotations do not break sign in, keep sessions short and agree audited break glass accounts for when the provider fails.