Designing a multi-datacenter FreeIPA topology

A practical guide to placing FreeIPA replicas across data centers, assigning CA, DNS and trust roles, and keeping authentication working when a site is lost.

On this page

When a multi-datacenter FreeIPA topology is needed and what it must answer

A multi-datacenter FreeIPA topology is needed when an organization runs Linux identity across two or more sites and logins must keep working when one site is lost. FreeIPA combines 389 Directory Server for users, groups, hosts and policies, MIT Kerberos for single sign-on, the Dogtag certificate system, SSSD on clients and optionally ISC BIND for DNS. Each of these components is replicated in its own way, so the layout of replicas decides how the whole service behaves during an outage. Before any server is installed, the team should be able to answer which sites hold authentication, which hold certificate and DNS roles, and what happens to clients in a site that loses its link to the rest of the deployment.

The directory runs as a multi-master LDAPv3 service, so any replica can accept writes and pass them on to its peers through replication agreements. That model is what makes a site survivable, but it also makes the number of agreements a design decision. The FreeIPA deployment recommendations advise at least two to three replicas in each datacenter and no more than four replication agreements per replica. Read together, these limits mean a site needs enough replicas to survive a single server failure, while each replica must still reach a small set of peers. The team should draw the agreement graph on paper for every site pair before installing anything, and confirm the current figures on the FreeIPA deployment recommendations page.

The topology also has to answer which servers carry the CA, DNS and trust roles, since the recommendations ask for at least one replica per datacenter running PKI or DNS where those services are used. Those choices determine whether a remote site can still issue certificates or resolve names when the primary site is unreachable. Plan the layout as a whole, record the agreement map and role assignments, and verify both with the team before the first replica joins.

Sizing the replica count per data center and the agreement budget

FreeIPA's published recommendations set practical limits on how many replicas each site runs and how many replication agreements each replica can hold. These limits decide how a multi-site mesh is built, so the cards below cover the replica count, the agreement ceiling, how the two interact, and the server hygiene that keeps the layout supportable.

Replicas per data center

FreeIPA's deployment recommendations advise at least two to three replicas in each datacenter. Two replicas let a site keep serving logins when one server is patched or fails, and three gives room for maintenance while still tolerating a failure. Count replicas per site first, then check that each site can still authenticate clients without reaching another data center. Confirm the count against your user volume and the sizing guidance before you commit to a layout.

The four-agreement ceiling

The recommendations cap replication agreements at no more than four per replica. An agreement is a直接 replication link, so the cap limits how many peers each server must keep synchronized. Plan the agreement graph before installing replicas, because each new server consumes agreement slots on existing servers as well as its own. Document which replica talks to which, and treat any server that approaches the cap as a sign the layout needs rework.

Building a mesh within the budget

With at most four agreements per server and two or three servers per site, a multi-site mesh is built by linking replicas within each site and then choosing a few cross-site links. Each site should have at least two independent paths to the rest of the topology so that losing one link or one site does not split the directory. Sketch the graph for three sites on paper and count the agreements per server. If a site needs more cross-site paths than the budget allows, add a replica to the site rather than stretching existing ones.

Keeping replicas off shared hosts

The recommendations call for dedicated servers without other applications. In a multi-site layout this matters because a FreeIPA replica that shares a host with a web app, a database or a build runner competes for memory, disk and patch windows, and a noisy neighbor can delay replication or certificate work. Ask each site's platform team which hosts will run FreeIPA and confirm that no other service is scheduled there. Include this requirement in the change process so it survives later consolidation projects.

Placing PKI and DNS roles

The recommendations advise at least one replica per datacenter running PKI or DNS, if used. Plan this per site so that a site that loses its nearest server still has a CA and a DNS service nearby. Record which replicas carry the CA and DNS roles, and confirm that the roles are present in every data center that needs them. The role choice affects agreement load and where certificate renewal happens, so review it together with the replica count.

Checking clients for two DNS servers

The recommendations also advise at least two DNS servers in each client's resolv.conf. In a multi-site layout, the two entries should not both sit in a site that might fail. Ask your network team how client resolver configuration is generated per site, and test a client in each location with one resolver stopped. This check ties the replica layout to what users actually experience during an outage.

Which servers should carry CA, DNS and trust roles

The Dogtag certificate authority is optional in a FreeIPA deployment, but where it is installed, the roles need deliberate placement. The FreeIPA deployment recommendations advise that at least one replica in each datacenter runs PKI, so that certificate requests and renewals can be served locally when a link to another site is down. The CA renewal and CRL publishing role is held by the first server by default and is separate from the question of which replicas run the CA service.

  • One CA replica per datacenter: Follow the recommendation so each site can issue and renew certificates without crossing a WAN link.
  • Renewal server location: Confirm which server holds the renewal and CRL role, and plan where it moves if the original server is retired.
  • Certificate replication: Confirm that CA data replicates to every CA replica in your topology before you rely on a site for issuance.
  • Single-CA risk: A topology with one CA replica has no local fallback, so review the number of CA servers before any site loses its link.
  • Sizing for the CA role: Count the CA replicas in each site against your certificate volume and renewal schedule.

Before you finalize the layout, confirm with your team which replicas carry PKI and where the renewal role sits. For the renewal role and the number of CA servers, see the sibling topics on the CA renewal server and on choosing CA servers.

How to build the topology in sequence

This sequence takes a multi-datacenter FreeIPA layout from name resolution to a verified, client-ready set of servers, so each site has local access and the topology stays connected if one site is lost.

  1. Confirm DNS and reverse resolution across sites

    Before any server is installed, the team checks that every planned server name resolves forward and in reverse from every site, using the DNS zones and reverse zones the design calls for. The network team and the platform team run the lookups from each data center, including the path clients will use. Done means every planned FQDN returns the expected address from each site and every address maps back to its planned name. A mismatch found here is far cheaper to fix than one found during replica installation.

  2. Install the initial primary server

    The team installs the first FreeIPA server in the site that holds the most authoritative administrative control, and sets the realm, domain and the DNS and CA roles the design assigns to it. Because this server starts the topology, its name and realm cannot be changed later without a replacement, so the naming decisions from earlier steps must be final. Done means the server answers Kerberos and LDAP requests, the admin account can authenticate, and the CA and DNS services report healthy in the local checks the installer runs.

  3. Add replicas per site with the chosen roles

    Each additional site receives at least two or three replicas, following the FreeIPA deployment recommendations, with the DNS and CA roles placed on the servers the design selected. The platform team installs each replica from an enrollment that is already prepared, and keeps the servers dedicated to FreeIPA. Done means every replica is enrolled, shows the expected roles in the topology view, and every site has at least one replica that runs the PKI or DNS services in use.

  4. Create replication agreements without exceeding four per replica

    The team creates replication agreements so that each replica has no more than four agreements, while every site stays linked to at least one other site through two independent paths. The agreements should be drawn on paper first, then created and reviewed against that plan. Done means the topology listing shows the planned agreements only, no replica exceeds four, and removing any single server or inter-site link still leaves every remaining server reachable from every site.

  5. Verify replication before clients use the servers

    Replication is tested by making a change on one server, such as a test user or group, and confirming it appears on every other server within the expected interval. The team also checks that the CA renewal and CRL roles sit where the design placed them, and that each server's DNS answers match the zone content. Done means test changes replicate in both directions across every site, the replication status reports no errors, and the test objects are removed afterward so the directory stays clean.

  6. Point clients at local and remote DNS servers

    Client configuration is updated so each machine lists at least two DNS servers in its resolver settings, with a nearby server first and a server at another site second. Clients in each site are enrolled against their local replicas, and the team confirms that the nearest server is used for lookups and authentication. Done means test clients in each site resolve the FreeIPA names, obtain Kerberos tickets from a local server, and continue to authenticate after the local DNS server is stopped in a controlled test.

  7. Record the topology and the failover test results

    The team documents the final server list, roles, agreements and resolver order, then keeps the record with the change process for later additions and removals. The site loss test from the verification step is repeated and its results are written down with the date and the people who ran it. Done means the topology record matches the live system, and the test results show which services survived a simulated site outage and which needed manual action.

Layout options and when each fits

This table compares three ways to lay out FreeIPA replicas across sites: hub-and-spoke, a bounded full mesh, and a primary site with a warm secondary. Read each row against your own site count and the sites you can least afford to lose. The main takeaway is that agreement count, CA and DNS placement, and the behavior after a site loss all move together, so choose the pattern before placing roles.

LayoutReplication agreementsCA and DNS placementBehavior when a site is lost
Hub-and-spokeEach spoke connects to one or more hub replicas, so spoke agreements stay low and hub replicas carry most linksPlace CA and DNS roles on hub replicas, with at least one per datacenter if used, so spokes do not depend on a single remote CALosing a hub weakens the topology most, so confirm each spoke still reaches a second hub before relying on it
Full mesh within a bounded replica countEvery replica links to others, which raises agreement count quickly, so keep each replica within the recommended limit of four agreementsSpread CA and DNS across sites so each datacenter has a local copy, with the CA renewal role on a server you can recoverA lost site leaves other replicas already connected, so authentication continues, but agreements to the lost site must be cleaned up
Primary site with warm secondaryPrimary replicas link to each other and to the secondary, and the secondary keeps agreements back to the primaryPrimary holds the CA renewal role and most DNS duties, while the secondary runs its own DNS and CA replicas ready to take overLosing the primary halts changes until a secondary is promoted, so confirm the secondary holds current data and roles before a failure
Recommended starting point (illustrative)Two or three sites with two to three replicas each usually keep agreement counts within the documented limitEach datacenter runs at least one replica with CA or DNS if those services are usedTest a simulated loss of one site in a lab before production rollout to confirm clients fail over

Keeping the topology connected when a site is lost

Surviving replicas keep replicating with each other only if the replication agreements that remain still form a connected graph once the lost site's servers are removed. Before you finalize a layout, map every agreement and confirm that each surviving site has at least one path to every other surviving site that does not run through the lost site. The deployment recommendations cap each replica at four replication agreements, so cross-site links are a scarce resource and should be placed on purpose, not left to chance. Use ipa topologysegment-find and ipa-replica-manage list to review the segments and the servers each one connects, and rehearse the loss of each site on paper before any outage occurs.

The CA renewal and CRL publication role is the point most likely to surprise a team during a site loss. The backup documentation notes that the first server handles certificate renewal and CRL publication by default. If that server sits in the site you might lose, renewals and CRL publishing stop with it, and the surviving servers will not take over automatically. Confirm which server holds the role before any planned outage, and if it sits in the site under maintenance, move the role to a server in a surviving site first. Record the result in your runbook so the on-call engineer does not have to discover it during an incident.

Backups are the last line of recovery, and they can cause damage if used carelessly. Restoring an outdated ipa-backup into a live replica topology can break replication, because the restored data no longer matches the changes the other servers already hold. Treat restore as a rebuild procedure: use the backup taken closest to the failure, restore it only in the manner the backup documentation describes, and avoid pushing it into servers that are still replicating with healthy peers. Test the full procedure in a lab that mirrors your topology, and write down which backup set goes with which site so the correct one is chosen under pressure.

Security and operational risks of a split topology

Spreading FreeIPA servers across sites widens the set of machines that hold keys, directory data and certificate roles. The cards below cover the risks that follow from that spread and the control to put in place for each one.

Trust boundaries between sites

Every FreeIPA replica holds a full copy of the directory, Kerberos keys and CA material, so a compromised server in a weaker site can affect the whole realm. Treat each data center as a trust zone and decide which sites may run CA, DNS and KRA roles. Check that replication traffic is restricted to known replica addresses and that administrative access to each server is limited to named staff.

Dedicated replica hosts

The deployment recommendations call for dedicated servers without other applications. Co-locating FreeIPA with general workloads in a remote site adds patch cycles, local administrators and attack surface that the identity service does not need. Confirm that each replica host in every site runs only FreeIPA, its DNS and its required services, and that anyone with root on the host is covered by the same access review as the directory.

Protecting backups in each site

The ipa-backup command produces full or online data-only backups that contain sensitive directory and key material. Backups copied to a site without equivalent physical and logical controls become a second copy of the realm's secrets. Encrypt backup archives, restrict who can read the backup store in each data center, and keep at least one copy outside the site where the primary replicas run.

Restoring stale backups

The backup documentation warns that restoring an outdated backup into an active replica topology can break replication. In a split layout, an operator may restore a site's backup while other sites have moved on. Record the backup timestamp for every server, compare it with the latest changes elsewhere, and rebuild a damaged replica from a current replica instead of restoring an old archive on top of live peers.

Knowing where renewal lives

The first server holds certificate renewal and CRL publication by default. If that server sits in one site and that site is lost, shared certificates may stop renewing and revocation status may stop updating across the realm. Before a site outage is possible, record which server holds these roles and confirm that another CA-capable replica in a different site can take them over.

Keeping server patches aligned

Replicas in different sites often receive updates on different schedules, which leaves the topology running mixed package levels. FreeIPA 4.13.4 is a stabilization release that fixed two CVEs, including LDAP access control flaws that could let unauthenticated clients obtain credentials, and hardened authentication checks across the API. Apply it to every replica in the same change window, starting with the servers that are reachable from untrusted networks, and record the installed version per server.

Questions teams ask before designing a multi-site FreeIPA layout

How much effort does a multi-site FreeIPA design take?

The effort depends mostly on how many sites you run, how many replicas each one needs, and what already sits beside FreeIPA. A design for two sites with no existing directory is a small exercise in agreement layout and role placement. A design that must also absorb an existing LDAP directory, an Active Directory forest or a DNS estate owned by another team takes considerably longer, because each of those adds decisions that cannot be made in isolation. The FreeIPA deployment recommendations advise at least two to three replicas in each datacenter and no more than four replication agreements per replica, so the number of sites multiplies the number of servers and agreements to plan. Ask for the design to list every server, role, agreement and DNS location before any build starts.

Can our internal team design a multi-datacenter FreeIPA topology?

Yes, if the team has someone who can read the FreeIPA deployment recommendations closely, test name resolution from every site, and reason about what happens when a site drops off the network. The harder part is usually not the installation commands but the decisions: which servers carry the CA, DNS and trust roles, where the first server's renewal and CRL duties should sit, and how the topology stays connected when a site is lost. Teams that have run FreeIPA in one location often underestimate how much of the design depends on DNS ownership and on the network paths between sites. An outside review of a draft topology is a common middle path, and it can be scoped separately from a full build.

What do we need to provide for a multi-site design?

Expect to supply a current network diagram showing every data center, the links between them and any firewall or routing limits, plus the DNS zones that FreeIPA will serve or delegate to. You will also need the list of existing systems that authenticate against the directory, any LDAP or Active Directory servers that must stay in service during migration, and the names of the people who approve DNS and firewall changes. Change windows matter because adding replicas, moving roles and changing agreements all touch live authentication. Confirm with your own change board how many windows you can book, how long each one lasts and who must be on call. Missing DNS ownership is the most common reason a design stalls.

How long does a multi-site FreeIPA project take?

The timeline follows the scope, and no duration is fixed until the written scope and your environment are reviewed. The main variables are the number of sites, whether the work is a new deployment or a change to a live topology, how many external systems must be integrated, and how quickly your DNS, network and change approvals move. A new deployment in a single lab can be measured in days of build work, while a live topology change across several sites that must keep authentication running usually needs staged windows and a rollback plan. Ask for a phase plan with dependencies, and check which phases depend on your teams rather than on the implementer.

What does a fixed-scope FreeIPA quote cover?

A fixed-scope quote covers the work described in a written scope, including the agreed deliverables, the timeline and the acceptance criteria, and it does not cover work outside that document. For a multi-site design, the scope should name the sites, the number of replicas per site, the role placement for CA, DNS and trust, the agreements to be created, the verification tests and the documentation handed over. It should also say what is excluded, such as ongoing operations, hardware procurement, or changes to other directories. DeployOpen does not publish a price list, so the quote follows from that written scope. Ask for the assumptions and the client responsibilities listed alongside each deliverable.

What drives the cost of a multi-site FreeIPA design?

Cost is driven by the same factors as effort: site count, replica count, the number of systems that must integrate, and the state of your DNS and network. A design project has one-time costs for discovery, design, build and verification, and some teams add recurring costs for ongoing support or upgrades. Hardware or virtual machines, storage and any cross-site bandwidth are paid to your own vendors or cloud providers and are separate from implementation work. The FreeIPA project is community licensed under GPL-3.0-or-later with no paid edition, so software licensing is not a line item. Ask for each cost component to be shown separately in a quote, with the recurring items marked.

What breaks if a site is lost during a multi-site design?

The design must answer this before any server is added. If the site that holds the only CA, the renewal and CRL role, or the only DNS servers goes offline, clients in other sites can lose authentication or certificate services even though their own replicas are healthy. The FreeIPA deployment recommendations call for at least one replica per datacenter running PKI or DNS when those services are used, and for at least two DNS servers in each client's resolv.conf. Confirm with your team how replication paths reroute during an outage, which roles must be moved before a planned shutdown, and how the topology is tested with a site intentionally isolated.

Does a multi-site FreeIPA design need an Active Directory decision first?

It can, and the answer changes the layout. If you plan a cross-forest trust with Active Directory, the IPA domain must use a DNS domain that does not overlap with any domain in the AD forest, and that choice affects naming and DNS delegation across every site. Existing LDAP directories also shape the plan because users and groups may need migrating into FreeIPA, and the migration path differs for sudo and HBAC rules. Settle naming, trust intent and migration scope before the replica layout is fixed, so the topology does not need to be rebuilt later. Confirm the trust and migration requirements with your directory and DNS owners first.

How DeployOpen scopes and delivers a FreeIPA topology design

DeployOpen approaches multi-datacenter FreeIPA topology as a written design followed by a controlled build in infrastructure you control. We start from your sites, failure scenarios and compliance obligations, then agree the replica layout, role placement and verification steps before any server is changed.

How we scope

  • Inventory every data center, network path and planned site, with the expected client and user populations at each one.
  • Agree which replicas carry CA, DNS and trust roles, and document why each placement was chosen.
  • Define the written scope for topology design, replication agreements, DNS locations and loss-of-site runbooks before work starts.

How we build

  • Draw the replication agreement map and check that no server exceeds the recommended agreement count.
  • Test name resolution, reverse DNS and replication health in a lab or staging copy before production changes.
  • Confirm that the first server's CA renewal and CRL publication role has a documented fallback at each site.

How we hand over

  • Deliver the topology diagram, role register, agreement inventory and site-loss runbook as versioned documents.
  • Walk your platform and security teams through failover tests and the steps to add, remove or rename a server.
  • Record the backup schedule and the restore constraints that apply when an outdated backup meets a live replica topology.

How we support

  • Offer ongoing support, upgrades and continuous testing as a retainer after handover, scoped in writing.
  • Review the topology when sites, FreeIPA releases or security advisories change the design.
  • Confirm with your team the support hours, escalation path and change approval process before the engagement is signed.

Talk to our team.

Tell us what you're working on, whether it's a deployment, an audit, a security test or a cyber range. You'll speak with an engineer who can help you scope it.

  • 30-minute call: free, with no obligation.
  • NDA on request: we can sign before you share details.
  • Clear next steps: a scope and plan after the call.