Cisco ASA Active/Active failover is a particular high-availability design for ASA devices configured in multiple security context mode. It allows the two units to pass traffic at the same time by assigning different context groups to be active on different units. It is not a generic cluster in which every packet from every context is automatically load-balanced across both appliances.
Cisco's ASA 9.22 documentation describes a maximum of two failover groups, with contexts assigned to those groups. Failover happens at group level, and the surviving unit must be able to handle the combined traffic. These limits shape architecture, maintenance and troubleshooting.
Multiple context mode is a prerequisite
Each security context acts as a logical firewall with relevant interfaces, security policies and routing configuration. In Active/Active mode, a context belongs to one of up to two failover groups; one group may be active on the primary unit and the other on the secondary, subject to configured state. The administrator context and unassigned contexts have documented default behavior that must be understood before designing placement.
A single-context ASA deployment cannot become Active/Active merely by enabling a second unit. If multiple-context support is unsuitable for the organization's features or policy model, Active/Standby may be the appropriate redundancy choice.
Active/Active is not transparent per-session load sharing
The active unit for a context processes traffic for that context. Operators can distribute different contexts across the two units, but a single session should not be understood as having arbitrary simultaneous processing across both members. Packet symmetry, routing and context membership determine which firewall state is relevant.
Record client-to-service flows and active context placement before implementation. An unexpected route through the other unit may fail stateful inspection even though both ASA interfaces show healthy link status.
Failover detection and state replication
Failover links carry control information about peer health and configuration/state exchange, while the dedicated state link can transport eligible connection state. Not all protocol details or transient flows may be synchronized under every software version and feature combination. Losing the state link can lead to dropped sessions during an otherwise successful group failover.
A deterministic test should disable selected interfaces or a peer only within an approved maintenance window and verify actual application recovery. Do not promise every TCP session survives with no retransmission simply because a standby or alternate active context exists.
Capacity after a member failure
When a unit carrying one active group fails, the surviving ASA may need to host both groups and their aggregate traffic. Capacity planning must include concurrent connections, inspection features, encrypted traffic, CPU, memory and link throughput in this degraded state. Designing both units at 100% load in normal operation gives no safe failover headroom.
An application that operates during a small test may still fail under full business load. Run representative traffic and observe session drops, queueing and firewall performance as failover occurs and service is restored.
Security policy and operational safeguards
Contexts retain distinct policies, objects and administrative responsibilities according to the configured model. Replicated configuration can spread a bad policy rapidly; high availability does not replace change review. Protect failover, state and management links with appropriate separation and access controls.
Include context membership, failover group priority, interface monitoring, routing dependencies and expected active placement in change records. Back up configurations and verify the rollback path before upgrading software or changing interface layout.
Verify state before declaring resilience
Use supported ASA operational commands and logs to check peer communication, failover groups, interface health, connection/state statistics and context status. Track exact timestamps of a fault and of restored application service. Identify whether interruption came from failover detection, state recovery, ARP/routing changes or upstream reachability.
Active/Active ASA is useful for distinct logical firewall workloads with controlled failover behavior. Its resilience is an observable property of context design and capacity, not a guarantee that every session and application runs without interruption.
Failover groups determine where a context runs
Cisco ASA Active/Active failover applies to supported multiple-context deployments. Contexts are assigned to failover groups, and the group that is active on one unit can be standby on the other. This is not per-packet balancing of a single firewall context across two appliances. The planned distribution matters because either surviving unit must have enough capacity and interface connectivity to carry the contexts that can move onto it.
Primary and secondary chassis designations are not the same thing as current active or standby status for a failover group. Record unit, context and group ownership separately when checking an incident. A healthy active/active arrangement may have both chassis forwarding different contexts, yet each individual failover group still has its own redundancy state. A failed link, context-specific service or entire chassis must be tested according to the release and monitored failover criteria.
Explain active/active context failover without promising packet-level load balancing
Cisco ASA active/active failover belongs to multiple-context designs supported by the relevant ASA hardware and release. Each failover group has an active unit and a standby unit, and different groups may be active on different physical peers. That is not the same as two ASA units processing each individual connection in parallel for a single context. Clarify which context owns each interface, security policy, NAT and routed service before describing a design as “active/active.”
For a planned resilience test, label the administrative contexts and their failover-group assignments, record the state of the groups and verify that upstream routing and connected switches can reach the active unit for each group. If one unit fails, workloads from its active group move to the survivor, but the survivor must have enough licensed performance capacity and correct interface/routing dependencies to handle the combined load. Redundancy does not manufacture throughput; failure may require accepting degraded capacity.
Observe actual application sessions during the controlled test. Which connection states survive depends on state replication, protocol and configuration, while some sessions must reconnect. Examine the failover history and connection-table behavior together with client errors before declaring success. Separate failover of contexts from failure of shared upstream devices, switch ports or power feeds; those risks require independent controls.
Accept degraded operation under real production policy
Test session preservation and new-flow establishment separately, including inspection, NAT and routes on both units. Some context-mode capabilities or management features differ from single-context ASA behavior; confirm the support matrix for the exact software before designing VPN, multicast, dynamic routing or other dependent services around the active/active architecture.
During a controlled failover, capture failover-state changes, state synchronization alarms, per-context connection counts, interface counters and application errors. Reversal to the original unit should be a deliberate operation with verified routing and policy symmetry rather than an automatic assumption that the preferred device will resume every context without disruption.
Worked example: one ASA unit fails during peak traffic
Two ASA units host separate groups of security contexts, each active on a different box. During a controlled failover, one unit must accept both groups and the combined inspected traffic. Check CPU, session limits, state-link health and application performance in that degraded state. If only one group moves and a route remains asymmetric, inspect the corresponding context path rather than assuming the cluster must be healthy because both units are powered. This demonstrates why two active chassis do not guarantee per-flow load balancing or zero interruption.