T Techclick ← All lessons
Check Point · Quantum R81.20 / R82 · Lesson 13 of 17

ClusterXL: when failover is dead

VIP 10.10.10.1 pings. Then it does not. Both members say they are Active — or both say Standby. cphaprob state is the first command, not a reboot of both boxes. This lesson is ClusterXL health and why “both members lie.”

Updated 2026-08-18·20 min read·L2 primary·Quiz at end

After this page you can

The ticket

Users lose HR for 40 seconds every hour. Or VIP is dead and both members claim Active. Night engineer reboots member 1 and member 2 together. Cluster never agrees who owns 10.10.10.1.

Quick interview answer

ClusterXL is Check Point HA (and optional load sharing). Members exchange state on a sync/CCP network. cphaprob state must show one Active (HA) and the other Standby, both “OK.” If both Active or both Down, you have split-brain or a dead CCP path — not “Windows is slow.”

Lab data · dummy only

SMS sms-lab 10.10.10.5 · cluster VIP 10.10.10.1 (cp-gw-01 10.10.10.2 / cp-gw-02 10.10.10.3) · external 203.0.113.25 · internal LAN 10.20.30.0/24 · HR PC 10.20.30.80 TECHCLICK\priya.hr · HR app 10.20.30.41 hr.techclick-lab.in. Not a live customer.

What ClusterXL is

Two Gaia boxes, one cluster object, one VIP per network. In High Availability, one member forwards. The other is ready. Sync copies connections so failover does not drop every TCP session (best effort — some flows still die).

They must see each other on the cluster interfaces. If the sync cable is unplugged, each member may decide it is Active. Two boxes ARP for the same VIP. The LAN melts.

Healthy HA
cp-gw-01 Active owns VIP 10.10.10.1 CCP / sync cp-gw-02 Standby ready

If the middle arrow dies, both sides may paint themselves Active.

HA vs load sharing

ModeWho forwardsWhen
High Availability (Active/Standby)One memberDefault you will support
Load SharingBoth (with restrictions)Special designs — more CCP/pivot care
smartconsole://sms-lab/gateways/cp-cluster
Training mock · not live
Gateways & ServersSecurity PoliciesLogs & MonitorManage & Settings
Gateways & Servers → cp-cluster → ClusterXL

Cluster properties

High Availability
cp-gw-01 10.10.10.2
cp-gw-02 10.10.10.3
10.10.10.1
ClusterXL Administration Guide — HA mode, VIP, sync network. Training mock.

Failover runbook

  1. Side A — do not reboot both

    Pick one member. Console available.

  2. Side B — state

    cphaprob state on both. cphaprob -a if for cluster IPs. cphaprob syncstat if sync is the suspect.

  3. Side C — one change

    Fix the down interface or CCP path. Failover with clusterXL_admin down / up in a window — not power-off both.

dummy healthy
cphaprob state
# ID Unique Address  Assigned Load   State
# 1  10.10.10.2      100%            ACTIVE
# 2  10.10.10.3        0%            STANDBY

cphaprob -a if
# Required interfaces: ...
# eth1  UP    10.10.10.2 (VIP 10.10.10.1)

# Split brain looks like ACTIVE / ACTIVE on both members.

Four cluster failures

1 · Both Active

Sync/CCP down or duplicate cluster IDs. Unplug one member’s data NIC only if you must stop the ARP war — then fix the sync path.

2 · Failover dead (stuck Active)

Standby never becomes Active. Interface criticality, CCP blocked, or clusterXL_admin down left on the good member.

3 · Failover flaps

Bad cable, duplex, or monitored interface bouncing. Find the interface in -a if, do not raise grace timers blindly.

4 · Reboot both

No member left to own the VIP. Double outage you invented.

How to prove it

Close the ticket only when

1) One Active, one Standby, both OK. 2) VIP ARP is one MAC. 3) A planned clusterXL_admin down on Active moves VIP. 4) You did not leave admin-down in place.

Traps

SymptomCluster storyWrong fix
Both ActiveCCP/syncReboot both
VIP deadNobody ActiveNew policy Any-Any
Flap hourlyMonitored NICDisable ClusterXL

Knowledge check

Judgment items. One best answer. Reasons send you back to the matching section.

Q1

Healthy HA cphaprob state shows…

Correct: a. Concept.
Q2

Both members Active usually means…

Correct: a. Failure 1.
Q3

First command?

Correct: a. Runbook.
Q4

Why not reboot both members together?

Correct: a. Failure 4.
Q5

Cluster private conversation rides…

Correct: a. Concept.
Q6

CoreXL is…

Correct: a. Do not mix names.

Check Point class series: Architecture · Gaia first day · SIC reset · Objects + first match · Policy layers · Hide vs Static NAT · Identity Awareness · HTTPS Inspection · Threat Prevention · Find the drop · fw monitor · SecureXL · ClusterXL · VPN Community · Policy install lock · vs PA vs Forti · CCSA / CCSE interview

Sources

Related: Check Point evidence desk · session factory · next lesson in the series above.