Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

"Insufficient resources to satisfy configured failover level for vSphere HA": read the admission control policy

Technical article
VMware

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

This message means vSphere HA admission control would be breached by the action you asked for. Admission control keeps spare capacity so that virtual machines can restart after a host failure, and it refuses any action that would eat into that spare capacity. The message appears when a power-on fails: "Insufficient resources to satisfy configured failover level for vSphere HA". Cluster-wide, a similar warning appears without any power-on. The fix is to find which policy is set, do its sum, and change the reservation, the percentage or the host count that makes the sum fail.

What the message means

The TechDocs vSphere 8.0 page "vSphere HA Admission Control" (Last Updated November 19, 2026) says admission control imposes constraints on resource usage, and that any action that might violate them is not permitted. The examples it gives are powering on a virtual machine, migrating a virtual machine and increasing a virtual machine's CPU or memory reservation. Capacity is set by cluster resource percentage, slot policy or dedicated failover hosts.

Broadcom KB 421814 (Updated On: 03-31-2026, vCenter Server 8.x and ESXi 8.x) names the power-on failure and explains it as two conditions that cannot both be met: the reservation of the VM being powered on, and the resource buffer the cluster's failover setting requires. If the reservation, added to current consumption, would leave less than the guaranteed buffer, the power-on fails.

Other strings in this family differ slightly, so match the one on your screen:

  • Power-on task: "Insufficient resources to satisfy configured failover level for vSphere HA" (KB 421814).
  • Cluster summary: "Insufficient configured resources to satisfy the desired vSphere HA failover level on the cluster" (KB 318133 and KB 424373).
  • Alert: "Insufficient resources to satisfy vSphere HA failover level on cluster" followed by the cluster and datacenter names (KB 396016).

What happens after a host has already failed is covered in "vSphere HA did not restart some VMs: reading the event trail and checking admission control and isolation settings". The checks below cover the warning and refusal before any failure.

How each policy does its sum

Cluster resource percentage. TechDocs says vSphere HA totals the reservations of powered-on VMs, totals the host resources available to VMs, and then computes the Current CPU Failover Capacity as (total host CPU minus total CPU requirement) divided by total host CPU. Memory is computed the same way. If either is less than the Configured Failover Capacity, the operation is disallowed. A VM with no reservation counts as 32 MHz of CPU and 0 MB of memory by default (the das.vmcpuminmhz advanced option changes the CPU figure). Only hosts that are connected, not in maintenance mode and without vSphere HA errors are counted.

TechDocs gives an example of three hosts with 24 GHz and 21 GB in total, and VMs requiring 7 GHz and 6 GB. The CPU sum is (24 - 7) / 24 = 17 / 24 = 0.708, which the page gives as 70%. The memory sum is (21 - 6) / 21 = 15 / 21 = 0.714, which the page gives as 71%.

Slot policy. TechDocs says a slot is sized by default to satisfy any powered-on VM, using the largest CPU reservation and the largest memory reservation plus overhead. It divides each host's resources by the slot size, rounds down, and takes the smaller of the CPU and memory counts. The Current Failover Capacity is the number of hosts, starting from the largest, that can fail and still leave enough slots. If it is lower than the configured value, the operation is disallowed. A single VM with a very large reservation inflates the slot size for everyone. The advanced options das.slotcpuinmhz and das.slotmeminmb set an upper bound on the slot size.

Dedicated failover hosts. KB 318133 (Updated On: 04-14-2026) says the warning can arise "if the maximum capacity host is not selected for dedicated host fail over policy".

Likely causes

Cause What you see How to confirm
A VM with a large CPU or memory reservation tries to power on Power-on task fails with the message The VM's reservations plus the cluster total, worked through the policy's sum
Reserved percentage too high for the hosts you have Cluster warning with no power-on; for CPU or memory, the "Desired" reservation is above the "Current" one in vpxd.log KB 424373 compares percentage with host count: 4 failures in 10 hosts is 40%
A few VMs distort the slot size Slot policy; few slots per host Largest CPU and memory reservations in the cluster
Dedicated failover host is not the largest host Warning on a dedicated-host policy KB 318133: pick the host with maximum CPU and memory
Cluster has too few usable hosts Warning on a 2-host cluster, or HA ignoring a host KB 396016; KB 301492 shows vpxd.log lines "Ignoring host" and numHosts=1

Prove the cause before changing anything

All read-only.

1. Read the policy. In the vSphere Client, select the cluster, open Configure, vSphere Availability, Edit, then Admission Control. TechDocs lists the fields: Host failures cluster tolerates, Define host failover capacity by (cluster resource percentage, slot policy, dedicated failover hosts or disabled).

2. List reservations. For the VMs on the cluster, note which have CPU or memory reservations. KB 421814 puts the reservation at the centre of the power-on case.

3. Read vpxd.log. On the vCenter appliance, KB 301492 shows the calculation lines in /var/log/vmware/vpxd/vpxd.log. KB 318133 shows lines that contain "Insufficient Failover resources for admission control." followed by the current and desired CPU and memory reservations, and KB 301492 shows "Failover resources check" lines with numHosts. A numHosts smaller than your host count means HA is ignoring a host.

4. Do the sum yourself with the formula for your policy, using TechDocs' definitions above.

A good result is a sum that passes with margin. A bad result is a sum below the configured value, which tells you which input to change.

Fix

Take a note of the current Admission Control settings (policy, failures tolerated, percentages) before you change anything below. That note is what each back-out uses.

Reduce or remove non-critical reservations

KB 421814 says to find non-critical VMs with high CPU or memory reservations and reduce or remove them, because reservations should only be used for critical workloads that need guaranteed resources.

Risk: a lower reservation removes a guarantee, so the VM may be slower under contention. Save first: record each VM's current reservation values. Back out: put the recorded values back in the VM's resource settings, if the admission control sum still allows it.

Align the percentage with the host failures you need

KB 424373 (Updated On: 07-27-2026) gives the formula (host failures to tolerate / total hosts) x 100, and the example 4 / 10 x 100 = 40% for 4 failures in a 10-host cluster. In the vSphere Client it uses Admission Control, Define host failover capacity by, Cluster resource Percentage, Override calculated failover capacity. The KB advises checking the new redundancy level against your service level agreements.

Risk: a lower percentage tolerates fewer host failures. Save first: note the current CPU and memory percentages. Back out: type the noted percentages back in.

Disruptive step: restarting vpxd interrupts vCenter management operations, including vSphere Client sessions and running vCenter tasks, until it is back. Save first: the note of the current Admission Control settings from the start of this section, and a list of any vCenter tasks still running. Back out: there is no undo for a restart; wait for the service to return and check the cluster, and reselect the noted settings if they were not kept.

KB 424373 adds that if the policy was changed in the interface yet the old calculation seems to apply, restarting the vpxd service forces vCenter to apply it, with service-control --restart vpxd. The KB shows only the command; on the vCenter Server Appliance run it in an SSH session as root.

Slot policy distortion

Save first: record the existing advanced options table. Risk: a fixed slot may make a large VM take several slots. Then set das.slotcpuinmhz or das.slotmeminmb in the vSphere HA advanced options to cap the slot size. Back out: delete the option you added.

Dedicated failover host, small clusters

For a dedicated-host policy, KB 318133 suggests using another policy or choosing the host with the most CPU and memory.

Risk: changing the policy or the dedicated host changes which capacity is held back, so VMs that powered on before may be refused, or a failover may land on a smaller host. Save first: note the current policy, failures tolerated and dedicated host. Back out: reselect the noted policy and host in Admission Control.

For a cluster with too few hosts, KB 396016 says to plan enough hosts to meet the demand when one or more hosts fail. In the 2-node cluster case of KB 301492, where HA ignores a host and numHosts shows 1 instead of 2, the KB's resolution is "Remove and re-add the problematic host to vCenter", then verify that numHosts is 2. The KB gives no removal procedure. Broadcom's "Removing a Host from a Cluster" TechDocs page covers removal from a cluster, not from the vCenter inventory. It says a host must be in maintenance mode first, with its powered-on VMs migrated off; the same precaution is sensible here.

Disruptive step: as advice, not from the KB: removing a host from the inventory ends vCenter's management of that host and its VMs until it is re-added. Save first: note the host's root credentials, cluster membership, network and storage settings and distributed switch membership, and list its VMs with their resource pool, tags, permissions and DRS rules. Then put the host in maintenance mode so its powered-on VMs migrate away; if the cluster cannot take them, stop. Back out: re-add the host to vCenter and the cluster using the noted settings, which is also the second half of the KB's resolution; then take the host out of maintenance mode and restore the noted VM settings.

Switching admission control off

KB 421814 offers temporarily disabling admission control as a workaround: Configure, vSphere Availability, Edit, Admission Control, Define host failover capacity by, Disabled. TechDocs says: "Do not permanently deactivate admission control", because without it you have no assurance that the expected number of VMs can be restarted after a failure. KB 396016 notes that disabling it suppresses the alerts.

Risk: the cluster may be unable to restart VMs after a host failure. Save first: record the current setting. Back out: set Define host failover capacity by back to the recorded policy once the VM is running.

Example (illustrative): Cluster cl01 has four hosts with 40 GHz of CPU each, so 4 x 40 = 160 GHz. Admission control uses the percentage policy at 25%, which is 1 failure / 4 hosts x 100. Powered-on reservations total 118 GHz, so the CPU failover capacity is (160 - 118) / 160 = 26.25%. db01 needs a 4 GHz reservation to power on. With it, the total would be 122 GHz and (160 - 122) / 160 = 23.75%, below 25%, so the power-on fails with the message. The administrator finds three test VMs reserving 6 GHz in total that need no guarantee and reduces that by 4 GHz. The total with db01 is now 118 GHz, the capacity is 26.25%, and the power-on succeeds. Memory passes in both cases.

Confirm the fix

  1. The failing power-on works, or the cluster warning clears after the settings are saved.
  2. Recalculate the policy's sum with the new numbers, and check there is margin above the configured value.
  3. Admission control is set to the intended policy and not "Disabled".

What to watch afterwards

  • Check the sum whenever reservations grow, or a host enters maintenance mode, because TechDocs counts only connected hosts that are not in maintenance mode.
  • Keep the percentage tied to the number of failures you intend to survive.
  • Treat any use of "Disabled" as temporary, with a reminder to restore it.

Risks and limits

  • These commands have not been run or lab-tested.
  • TechDocs gives two host counts on different pages: "To use vSphere HA admission control you must have at least 3 hosts in the cluster", and, for the percentage policy, a check that at least two vSphere HA-enabled hosts exist. KB 396016 says a cluster of 2 hosts needs 3 hosts available for successful admission control. Neither page says which count governs for your cluster, so plan for 3 hosts.
  • The TechDocs pages are vSphere 8.0 pages, and each KB lists its own versions. HA behaviour in other versions may differ.

Sources