Device SMTP relay through hMailServer and SendGrid

Introduction

This project configures a Windows SMTP relay for infrastructure devices that cannot conveniently authenticate via modern auth to an external mail provider. This SMTP relay replaces what I used to build for many years using the Windows built-in SMTP feature. But in Windows Server 2025 and beyond that is gone, as Microsoft is not going to modernize it and it depended on the legacy IIS 6.0 web server components. But in many environments we still need SMTP relays, as many devices remain that can do only basic auth via STARTTLS or even unauthenticated mail. So I needed another SMTP relay solution, preferably one I can automate. I finally settled on hMailServer as it is open source, free, and has a COM API to automate it. Below you will find the design, setup, and configuration information to help you automate it yourself. It can install, maintain, and repair a deployment, as well as secure your SMTP service as well as possible to ensure it cannot be abused easily.

Design and mail flow

Device SMTP relay through hMailServer and SendGrid

The internal relay has two separate submission paths. Sources without requireSmtpAuth: true use TCP 25 without authentication or TLS. Sources marked true are excluded from TCP 25 and use TCP 587 with SMTP AUTH and STARTTLS required. Windows Firewall and the matching hMailServer IP ranges enforce this separation. Both paths relay onward through SendGrid using required STARTTLS and the apikey account. You can change the port or add one easily for a device that might require it. hMailServer treats domains configured locally as local sender/recipient domains. The script creates the internal domain and the authorized sending domain when missing. Managed device ranges permit local-to-external delivery only. Whether that direction requires SMTP AUTH is configured per device with requireSmtpAuth; the other three local/external delivery combinations remain denied. Configure device senders accordingly.

The global rewrite rule changes the From header to authorizedFrom when that header contains internalDomain. This is not an envelope-sender rewrite, an exact domain match, or a general sender-identity enforcement mechanism. If possible, use the public domain(s) SendGrid trusts.

High availability design

In real life I normally make the SMTP solution highly available behind a load balancer.

Highly available device SMTP relay through Kemp LoadMaster HA, two hMailServer nodes, and SendGrid

The diagram shows a proposed HA topology, not an HA deployment created by this script. A Kemp LoadMaster HA pair presents a shared SMTP virtual IP and hostname (smtp.workinghardinit.work). Devices connect to that endpoint; the load balancer forwards connections to either smtp01.workinghardinit.work or smtp02.workinghardinit.work. Each hMailServer node independently sends mail to SendGrid using STARTTLS and its configured API key. The pictured hostnames are examples and differ from the single-server hostname in the supplied JSON.

The red path represents unauthenticated, unencrypted SMTP on TCP 25. The blue path represents SMTP AUTH with STARTTLS on TCP 587. Keep the two paths separate at the load balancer and relay nodes, and restrict TCP 25 to trusted network segments. HA does not make unauthenticated or unencrypted SMTP secure.

Planning and configuration

  1. Deploy two independent hMailServer nodes and a supported Kemp HA pair. Configure separate TCP 25 and TCP 587 virtual services on the shared virtual IP, with both relay nodes as backends. Configure and test node and virtual service health checks; do not assume that a TCP-open check proves delivery.
  2. Provide DNS for the shared endpoint and both node hostnames. Run the configuration script on each relay with the appropriate node hostname, consistent relay policies, device inventory, sender settings, and upstream configuration. The script configures one local server at a time; it does not configure Kemp, synchronize the nodes, or provision shared DNS.
  3. Decide whether Kemp preserves client source IPs or uses source NAT. The relay’s Windows Firewall and hMailServer ranges use the source address actually seen by the backend. If multiple clients are NATed to the same address, backend per-device restrictions no longer distinguish them: enforce the intended per-device and per-port access at Kemp or upstream, and design backend rules accordingly. In particular, do not assume backend IP ranges can distinguish authenticated from unauthenticated clients sharing one NAT address. Verify the observed source addresses on both nodes and during Kemp failover before enabling access.
  4. Configure TLS and SMTP AUTH end-to-end for TCP 587. If using TCP pass-through, each backend must present a certificate trusted by clients and valid for the shared hostname clients connect to, in addition to satisfying any node-specific requirements. A certificate containing only a node hostname will not validate for the shared name. If terminating TLS at Kemp, separately design and protect the Kemp-to-relay leg; STARTTLS is an SMTP protocol upgrade, so confirm the chosen Kemp configuration supports the intended behavior rather than assuming ordinary HTTPS termination applies.
  5. Provision consistent SMTP AUTH credentials and SendGrid access on both nodes using secure procedures. Reapplying authenticated submission resets the shared account password on the node being configured; coordinate changes across both nodes and devices. Keep certificate private keys and credentials protected on each system. Demo self-signed certificates require client trust and are not a production certificate strategy.
  6. Configure independent monitoring and backups for both relay nodes, Kemp, DNS, and SendGrid connectivity. The relays have independent queues: failover does not automatically transfer messages already accepted by a failed node. Recover that node and its queue to deliver pending messages; avoid replaying messages without considering possible duplicates. SendGrid, DNS, networking, power, and any shared credentials or infrastructure remain common dependencies.

Validate failover

From representative allowed and denied source addresses, test TCP 25 and 587 against the shared hostname. Verify authentication, certificate trust and hostname, and actual message delivery through each backend. Then test one relay node unavailable, Kemp node failover, recovery, and queued-message handling. Review load-balancer, hMailServer, firewall, and SendGrid logs; SMTP acceptance alone does not confirm delivery. Capture evidence and repeat tests after changes.

Files

All files you need are at https://github.com/WorkingHardInIT/SMTPRelayThroughhMailServerAndSendgrid

Requirements and preparation

  • Windows with Windows PowerShell 5.1; regression tests also run in PowerShell 7.
  • Administrator rights for installation and Windows Firewall changes. The script requests UAC and relaunches when necessary.
  • Standard hMailServer installation layout and working COM registration.
  • An outbound connection to the configured SendGrid host and port, normally smtp.sendgrid.net:587, plus DNS and HTTPS for installer downloads if needed.
  • A SendGrid API key with Mail Send permission and an appropriately verified sender/domain. Configure provider-required DNS separately.
  • Native Windows New-SelfSignedCertificate support when demo-certificate generation is enabled and either TLS output file is missing. No external certificate utility is downloaded or installed.
  • An internal DNS record for the relay FQDN pointing to the relay server.
  • Back up existing hMailServer configuration/data and record existing firewall settings before applying to an established server.

Review the supplied addresses: they are examples, not discovered allocations. The inventory contains 18 data switches, 5 Omada/UniFi devices, 4 storage switches, 12 OOB switches, 4 proposed iDRAC9 addresses, 4 Kemp LoadMasters, and 2 OPNsense firewalls. Reserve and verify every proposed address, routing, and the source IP actually seen by the relay before applying.

JSON configuration

Edit HMailServerRelay.json, not the script, for routine device and mail-setting changes.

Setting Meaning
smtpHostname Relay’s FQDN, currently smtp01.datawisetech.corp; used for hMailServer’s host identity. Not the SendGrid hostname.
internalDomain Domain created in hMailServer and used by the From-header rewrite condition.
authorizedDomain Sending domain created in hMailServer; authorizedFrom must belong to it.
authorizedFrom Replacement From header address.
sendGrid.host / sendGrid.port Upstream SMTP endpoint; STARTTLS and authentication remain required by code.
tls.preferServerCipherOrder Boolean controlling the server’s cipher preference; the supplied config enables it.
tls.prioritizeChaCha20 Boolean prioritizing ChaCha20 when the client prefers it; the supplied config enables it. Requires tls.preferServerCipherOrder: true.
smtpAuthentication.enabled Enables provisioning of the shared account, certificate/listener, and authenticated TCP 587 path. The checked-in demo config enables it.
smtpAuthentication.generateDemoCertificate When true, generate a 3072-bit, one-year self-signed certificate only when both configured TLS files are missing. It never replaces existing files.
smtpAuthentication.accountAddress Shared account created under authorizedDomain; default [email protected].
smtpAuthentication.port Fixed at 587.
smtpAuthentication.certificateName hMailServer certificate entry name. The certificate must identify smtpHostname.
smtpAuthentication.certificateFile / privateKeyFile PEM certificate and unencrypted PEM private-key paths under C:\hMailServerFiles\Certs; the private-key ACL is limited to SYSTEM, Administrators, and the hMailServer service identity when known.
devices Complete desired list of named IPv4 /32 devices.
devices[].requireSmtpAuth Optional boolean; true requires authenticated submission on TCP 587, while omitted/false selects unauthenticated TCP 25.
devices[].description Optional inventory annotation only; not written to hMailServer IP ranges.

Device example:

{
  "name": "DataSwitch-1",
  "cidr": "172.16.10.11/32",
  "description": "First data switch",
  "requireSmtpAuth": false
}

Demo certificate and authenticated submission

The checked-in configuration enables authenticated submission and opts into demo certificate generation. The certificate paths and generation setting are already set in JSON; the installer defaults to C:\hMailServerFiles\Installer. Installer and certificate files are kept under C:\hMailServerFiles.

  1. On apply, if both configured TLS files are missing, native Windows PowerShell creates a one-year RSA-3072 self-signed certificate with a DNS SAN for smtpHostname, exports the certificate and unencrypted RSA private key as PEM, then removes its temporary certificate-store entry. The .crt and .key suffixes are just filenames; both files contain PEM data. hMailServer requires the private key to have no passphrase, and a CA-issued certificate file should include its trust chain. If both files exist, they are reused. If only one exists, the script stops rather than replacing either file.
  2. This self-signed certificate is for demonstrations, not production trust. Clients must be configured to trust this exact certificate (or use a properly trusted internal/public CA certificate instead). Until clients trust it, STARTTLS encrypts the connection but does not reliably authenticate the relay identity. The script checks the generated cert/key match and DNS identity; it does not manage client trust. See the hMailServer SSL certificate documentation for certificate-chain and unencrypted-key requirements.
  3. Apply the script. It prompts securely for the shared account password, provisions the account, configures a STARTTLS-required port 587 listener, updates firewall allowlists, restarts hMailServer, and checks that its process owns a listener on port 587.
  4. Configure each selected device for smtpHostname, port 587, STARTTLS (not implicit SSL), SMTP AUTH, and the provisioned account credentials.

Reapplying with authentication enabled resets the shared account password to the newly prompted value; update the devices to match. Generated files are never overwritten automatically. Do not use -SkipWindowsFirewall while authenticated submission is enabled. On a firewall-managed run with authentication disabled, TCP 587 is explicitly blocked.

Names and addresses must be unique. Do not list 127.0.0.1: the script always adds its managed loopback entry. An explicit "devices": [] removes managed device entries while retaining loopback. Missing/invalid settings fail rather than falling back to defaults. Use exact property spelling; no comments, trailing commas, passwords, or API keys belong in this file.

The script configures the FQDN but does not rename Windows or create DNS records. For this internal relay, the name can resolve internally; SendGrid handles onward Internet delivery. A public receiving MX record or inbound port forwarding is not required for this outbound-only design.

Run and maintain

From the project directory:

# Safe preview: no UAC, credential prompts, or configuration changes.
.\Configure-HMailServerRelay.ps1 -WhatIf

# Apply: review the confirmation, then enter credentials in PowerShell.
.\Configure-HMailServerRelay.ps1

# Select an alternate authoritative configuration.
.\Configure-HMailServerRelay.ps1 -ConfigPath 'C:\SysAdmin\Relay.json' -WhatIf

If elevation is required, approve UAC and finish in the new PowerShell window. The parent returns after launching it; launch success is not configuration success. Parameters, explicit switch values, and the working directory are forwarded. Credentials are requested only after confirmation in the target run.

Other parameters:

  • -HMailAdminUser: administrator username; default Administrator.
  • -InstallerPath: installer file path; default C:\hMailServerFiles\Installer\hMailServer-5.7.1-B3116-x64.exe.
  • -SkipWindowsFirewall: leaves firewall rules untouched, including obsolete addresses. hMailServer device reconciliation still happens.
  • -KeepInternetRange: leaves the existing Internet range unchanged. This is an exception to the normal hardening behavior, not a way to restore previously disabled settings.
  • -Confirm:$false: bypasses the configuration confirmation, not UAC or credential prompts; use intentionally.

To add/remove a device, edit its JSON entry, preview, and reapply. To change its IP, keep the stable name. Renaming creates the new managed range and removes the old. The JSON list is not an incremental list of additions.

Which IP ranges are changed or removed?

The script does not remove every hMailServer IP range that is absent from the JSON:

  • Script-managed ranges: Names starting with SendGrid relay are reserved for the script. It creates or updates the ranges for JSON devices and 127.0.0.1, and removes other ranges with that prefix. Do not use the prefix for manual ranges.
  • Recognized older ranges: Unprefixed names matching current device names or the script’s older device-name patterns can be removed or migrated if they are single-IP ranges and their settings match the script’s relay settings. A matching current device gets a prefixed replacement. Recognized entries whose settings differ are left unchanged with a warning.
  • Other manual ranges: Generally left unchanged, even if they are absent from JSON. Previously removed custom unprefixed device names may no longer be recognizable, so review them manually.
  • Default and old broad ranges: The script disables SMTP on My computer and, unless -KeepInternetRange is used, disables SMTP and delivery on Internet. It also disables SMTP on the original DataSwitches, StorageSwitches and OutOfBandSwitches broad ranges when their IP bounds match the original defaults. These ranges are not deleted by those actions. You can use your own naming convention for your devices. These are just examples.

An hMailServer range left intact is not necessarily reachable: the managed Windows Firewall rules independently block sources outside the configured TCP 25 and 587 allowlists. Review other hMailServer ranges and firewall rules before applying changes.

Installation and repair

Readiness requires a standard-location interop DLL, the hMailServer service, and the COM ProgID registration. A DLL alone is not proof of installation.

When setup is needed, the script:

  1. Obtains the pinned official hMailServer 5.7.1 Build 3116 x64 installer.
  2. Uses an existing matching file or downloads it when absent; verifies SHA-256.
  3. Reports download failures with the official download URL and the exact -InstallerPath destination. It never silently accepts a mismatched installer.
  4. Runs unattended setup with the built-in database option, suppressed dialogs, and no automatic restart.
  5. Checks the exit code, installation readiness again, and the service’s running state before proceeding.

The installer is pinned, not dynamically updated to the latest release. When repairing, supply the existing administrator password. Setup currently rejects passwords shorter than six characters or containing whitespace/double quotes because of installer/database-setup limitations. A service present without a standard-location interop DLL stops with an explicit error rather than risking an unintended install over a custom layout.

Security controls and boundaries

Control Purpose and limitations
Individual IPv4 /32 allowlist Avoids trusting every device in a subnet. Trust is source-IP based, not user/device authentication.
Managed hMailServer ranges at priority 1000 Allow SMTP and local-to-external delivery only; deny POP3/IMAP connections for these ranges. Review other equal/higher-priority ranges manually.
Internet range disabled by default Disables SMTP and all delivery combinations on the default IPv4 catch-all. Does not control outbound SendGrid traffic.
My computer SMTP disabled Removes a permissive fallback while retaining the managed 127.0.0.1 relay. Other My computer settings remain unchanged.
Managed firewall rules TCP 25 permits only unauthenticated IPv4 sources plus loopback; TCP 587 permits only authenticated sources when enabled. IPv4 complements and all IPv6 are blocked on both ports.
TLS protocol policy The script enables TLS 1.2/1.3 and disables TLS 1.0/1.1 in hMailServer. If a legacy client truly requires TLS 1.0/1.1, an administrator must re-enable it manually; a later script run restores this policy.
Cipher-order options The JSON controls “Prefer server cipher order” and “Prioritize ChaCha20-Poly1305 when the client prefers it”; both are enabled in the supplied config. These control selection order, not the set of allowed ciphers.
Required upstream STARTTLS Protects SMTP transport to SendGrid; certificate verification remains hMailServer’s responsibility.
Prompted credentials No secrets in JSON/source; SendGrid credential is stored through hMailServer’s settings API. Plaintext is temporarily required by COM/setup APIs.

TLS versions and cipher-order options

hMailServer 5.7.1 exposes settings for each TLS version. The script sets TLS 1.2 and 1.3 on and TLS 1.0 and 1.1 off, restarting the service when that policy changes. This does not force TLS on unauthenticated TCP 25; it controls protocol versions whenever a TLS handshake occurs. SMTP AUTH on TCP 587 still requires STARTTLS.

For authenticated submission, the device and hMailServer negotiate a TLS version and cipher when establishing STARTTLS on TCP 587. The TLS version settings also apply when hMailServer connects to a remote server such as SendGrid. Cipher compatibility matters in both directions; server cipher preference applies to incoming connections where hMailServer is the server.

The two cipher options are controlled by the required tls JSON settings:

  • Prefer server cipher order makes the server choose according to its configured cipher order instead of the client’s order. It is a useful control when the server’s cipher list is deliberately strong, but does not itself remove weak ciphers.
  • Prioritize ChaCha20-Poly1305 when the client prefers it favors ChaCha20 for clients that advertise it first, often improving performance on mobile devices without AES hardware acceleration. This is a performance preference, not an additional security layer, and OpenSSL requires server cipher preference to be enabled for it to take effect.

An alternate configuration file must include both boolean tls settings; the ChaCha option requires server cipher preference to be enabled. OpenSSL documents the server-preference and ChaCha options.

You can manually remove weaker ciphers from Settings > Advanced > Security > SSL/TLS ciphers after confirming that no required device or remote mail server needs them. Save the original list first, then test STARTTLS with all relevant devices and delivery to SendGrid; incompatible peers may no longer connect. The script does not change this cipher list.

Remote server certificate verification

Keep Verify remote server SSL/TLS certificates enabled. It checks the identity of servers hMailServer connects to, such as SendGrid, including certificate trust, hostname and validity. This setting does not verify connecting devices and does not make devices trust the relay’s self-signed certificate. Each device that validates the relay’s certificate needs to trust it separately and connect using its matching hostname. The script leaves remote-server certificate verification unchanged.

The firewall rules are named:

  • HMailServer-SMTP25-Trusted-Allow
  • HMailServer-SMTP25-Untrusted-IPv4-Block
  • HMailServer-SMTP25-Untrusted-IPv6-Block
  • HMailServer-SMTP587-Trusted-Allow
  • HMailServer-SMTP587-Untrusted-IPv4-Block
  • HMailServer-SMTP587-Untrusted-IPv6-Block

The TCP 25 and 587 IPv4 blocks are calculated independently from the same authoritative device list; they are not manually maintained or SendGrid destination-IP lists. IPv6 uses ::/1 and 8000::/1 because ::/0 was rejected by the tested Windows Firewall. Rules apply to all profiles and are updated in place, block rules first. Windows Firewall must be enabled and effective policy must permit these local rules; Group Policy and unrelated explicit block rules can affect results.

Important boundaries:

  • Any application on the relay host can use its unauthenticated loopback relay.
  • A compromised allowlisted device can abuse its relay permission. Recipient restrictions and rate limiting are not implemented here. SMTP AUTH protects the credentials/transport but is shared across authenticated devices.
  • If NAT combines clients behind an allowed source IP, they share its trust.
  • The script does not globally disable POP3/IMAP services, close every installer listener, or harden every pre-existing hMailServer range or firewall rule.
  • Unattended setup requires the administrator password on a process command line. Local process inspection/monitoring may expose it; SecureString does not eliminate that installer requirement.
  • Changes are not transactional. A failure can leave partial changes, including firewall rules installed before setup. Correct the cause and reconcile again.

These controls are a scoped relay design, not proof of comprehensive hardening.

Testing and troubleshooting

Regression tests (mocked management APIs; no mail is sent):

powershell.exe -NoProfile -File .\Configure-HMailServerRelay.Tests.ps1
pwsh.exe -NoProfile -File .\Configure-HMailServerRelay.Tests.ps1

Network-only check from the actual relay server:

Test-NetConnection smtp.sendgrid.net -Port 587

A successful result proves TCP reachability only, not STARTTLS, authentication, SendGrid acceptance, or inbox delivery. UniFi policies, IDS/IPS, routing, or other network controls can block traffic; inspect their logs rather than assuming. Do not add inbound port forwarding to solve an outbound SMTP problem.

To send a real message, substitute an address you control and explicitly set the sender and recipient instead of relying on script defaults:

.\Test-SmtpRelay.ps1 -SmtpServer '127.0.0.1' `
  -From '[email protected]' -To '[email protected]'

Run from the relay host for loopback, or from an allowed device/IP against the relay’s FQDN to test the network path. Local SMTP acceptance is not final delivery: inspect hMailServer’s queue/SMTP/error logs, SendGrid activity, and the recipient mailbox (including spam). Handle logs as sensitive; redact credentials and message data before sharing.

For 80040154 / Class not registered, inspect service, binaries, COM registration, and process architecture. The updated detection schedules setup for missing COM registration, but registry presence alone does not guarantee COM activation. Do not delete the database or blindly reinstall over a custom installation. But we do detect leftovers from a previous installation so as not to mistake them for a properly installed hMailServer.

Conclusion

This script provides me with an automated, reproducible way of deploying and maintaining an internal SMTP relay on Windows Server for various devices that still require one. It can be made highly available when required. I am happy with the results, and it helps me deliver value in niche environments that the big companies no longer serve.

SME VMware-to-Hyper-V migration project madness

Inroduction

We will discuss the ridiculous decisions some bean-counting managers make when migrating their aging, small VMware deployments to the excellent, affordable Hyper-V solution, especially in a Microsoft ecosystem. I’d call it “FinOps gone wrong,” but it’s really a mix of stupidity, extreme bean-counting, and a lack of knowledge about a process they clearly shouldn’t influence. In other words, incompetence.

Everyone knows the reasons why these migrations happen. Customers do not appreciate being squeezed out like a lemon by a company that doesn’t respect their business and would rather you leave.

Let’s talk about a classic IT horror story. Management reads an article, looks at licensing costs, gets spooked by vendor changes, or, in Broadcom’s case, insulted and exploited. So they decide: “We’re migrating off VMware to Hyper-V. It’ll be easy, right? And hey, we don’t need to buy new hardware. We can repurpose the existing gear or squeeze it onto whatever’s lying around the server room.”

Stop right there. Hyper-V is a rock-solid hypervisor, and if you’re willing to leave the vCenter reservation behind, you will find it a flexible, capable solution. That is not the problem.

If you are planning a migration from VMware ESXi to Microsoft Hyper-V and management’s strategy can be summed up as “do it with what we already have; we are too cheap to buy spare or new hardware,” you are essentially being asked to defuse a bomb while blindfolded in a dark room because the exponentially safer approach will cost us some extra money. They are willing to risk their reputation, yours, and the livelihoods of their customers or business.

Why is this a bad idea

Let’s break down why trying to pull off a cluster migration on a shoestring hardware budget is a disaster waiting to happen, focusing on the twin horsemen of the infrastructure apocalypse: Single Node Ops and Untested Failover Clustering.

A high availability cluster must achieve its goal by providing redundancy and failover. You size them, configure them, and test them for that. Sizing is usually N-X (N-1, N-2, N-3) setups, depending on your cluster size and how many clusters you have. If you have multiple clusters and N-2, you might be able to carve out enough hardware without adding much risk to set up a new Hyper-V cluster. The bare minimum is two nodes if you want to validate and test your cluster and backups before taking it into production. If your idea is to leverage backups to save your proverbial behind, that is cool if you know they are good and you can restore, but it will hit hard because of downtime. This is not your best shot, A-game proposal. To add insult to injury, teams often make these decisions in 2-node clusters, where you can’t pull spare hardware from a larger pool. That option does not exist.

Penny Wise, Pound Foolish

In an ideal world, a migration project looks like this:

  • Build a shiny new target cluster (Hyper-V).
  • Thoroughly validate storage, networking, and cluster health.
  • Set up backups, test them, and recover test workloads.
  • Perform various failover scenario testing to validate your tests and assumptions. Shut down a controller, unplug a switch, a server node, etc.
  • Migrate workloads gracefully over time.
  • Decommission the old environment.

In the real world, way too often the following idea comes up: “Can’t you just wipe Node A of the old cluster, install Windows Server / Hyper-V on it, move a few VMs over, and repeat?”

No! Unless this is the only option left, outside of anything you can control, and your business is toast if you don’t, there is no sane reason to take these risks.

When you have zero spare hardware, you lose the single most important safety net in enterprise IT: headroom. You have no staging ground, no sandbox to test quirks, and no buffer when things inevitably go sideways during conversion.

The Danger of Single Node Operations

When management refuses to buy adequate hardware for a transition, you are forced into Single Node Operations right out of the gate. That is supported for last-man-standing scenarios, but it is NOT healthy, and you should not use it where it is not intended.

  • Where is your High Availability? A cluster requires a quorum. If you are stripping down nodes or running on a minimal configuration because you don’t have enough physical hosts to maintain N+1 redundancy during the migration phase, you aren’t running a cluster—you’re running a series of single points of failure duct-taped together.
  • The Resource Ceiling: Hyper-V and Failover Clustering are fantastic, but they aren’t magic. If you dump a massive chunk of your migrated workloads onto a single node because the rest of the target cluster hardware hasn’t arrived (or never will), you saturate memory, CPU queues, and IOPS. Maybe you are “lucky” and have enough headroom, but you still shouldn’t bet on it.
  • The Domino Effect: If that single overloaded node hiccups, crashes, or hits a resource exhaustion wall, your entire migrated workload goes down with it. There’s no automatic failover because the secondary node is either empty, unconfigured, or nonexistent.

Skipping Failover Clustering Validation and Other Checks Before Cutover

Hyper-V relies heavily on Failover Clustering for high availability, live migration, and resilient storage architectures (like Cluster Shared Volumes – CSVs). Setting up a rock-solid cluster requires rigorous pre-flight validation.

Normally, you run the built-in Cluster Validation Wizard to check storage, networking, drivers, and quorum configuration. Then test real-world failure scenarios before signing off on the production deployment. But when you are rushing a zero-budget migration under immense pressure, corners get cut:

  • Untested Storage Paths: Are your CSVs performing under load? Do you have proper multipathing (MPIO) configured, or is every I/O operation bottlenecked down a single sluggish path because nobody wanted to buy proper redundant switches or HBAs? You won’t find out until production hits it.
  • Network Partition Risks Live Migration: Live migration requires robust, dedicated cluster networks (ideally SMB Direct/RDMA or high-speed dedicated VLANs). On repurposed, constrained hardware, you’re often fighting for every megabit of bandwidth. Without proper validation testing, a live migration can easily stall, drop packets, or split-brain your cluster.
  • The Backup Fall-Safe Illusion: If you are skipping staging hardware and banking entirely on backups to bail you out if the migration craters, you better have verified those restores beforehand—otherwise, relying on cold backups mid-crisis will hammer the business with massive downtime.
  • The “Cutover and Pray” Methodology: Because there’s no spare hardware to run a parallel pilot, the cutover becomes a high-stakes, all-or-nothing event. You flip the switch, cross your fingers, and hope that the cluster validation errors you ignored don’t come back to bite you at 2:00 AM. Spoiler alert: they will.
  • To complete your migration, you need to flatten the remaining nodes, convert them to a cluster node, and add them to the new Hyper-V clusters. You have introduced risk into your existing solution, introduced risk into your new solution, and then created even more risk by wiping the remaining node(s) of the old solution.

The Bottom Line

Infrastructure migrations are complex enough when you do have the right tools, proper hardware headroom, and adequate time. Stripping away the hardware budget doesn’t save the company money. It defers the cost, trading upfront capital expenditure for a high risk of catastrophic downtime, burned-out sysadmins, emergency consulting fees, potential disaster recovery, reputational damage, and economic loss.

If management tells you to migrate a mission-critical VMware environment to Hyper-V with zero budget for new or spare hardware, show them the math on what a few hours of total downtime costs the business. Because sooner or later, running a cluster on a shoestring budget is going to snap. YOLO managers have no business making decisions about such technical matters. But economic risk management should be their business, and one would hope they don’t choose such a bad approach. Alas, I have heard and seen it way more than I would like to.

Veeam 100 Summit 2026 – Day One

Veeam 100 Summit 2026 – Day One

The Veeam 100 Summit 2026 is taking place in Lisbon, Portugal, as I write. On day one, I attended the Object First track. Object First was acquired by Veeam in Q1 2026. While I cannot disclose anything discussed here, I can share that these people are (very) ambitious. It was a technical, no-nonsense experience where I voiced my enthusiasm, wishes, and concerns in my usual feedback style. Direct, unapologetic but polite, with the occasional strong “force multiplier” wording.

That led to some interesting discussions with lead architect Alexey Vasilev and Anthony Cusimano at the evening networking event. The level of interaction didn’t disappoint, as we have come to expect at the Veeam 100 Summit 2026. Great first day.

Veeam Application Backup Repository (ABR) in v13.1 for network devices

Introduction

The release of Veeam Backup & Replication v13.1 introduces a significant new capability: the Application Backup Repository (ABR). Positioned as a secure landing zone for application‑generated backups, ABR brings hardened storage, snapshot lifecycle management, and immutability to workloads that generate backups of data or configurations to an NFS share. Databases and network devices come to mind and are mentioned in the documentation here: Application Backup Repositories – Veeam Backup & Replication User Guide.

“Configuration data of network devices: firewall or switch configurations.”

These “backups” or exports traditionally operate outside Veeam’s agent‑based or hypervisor‑based protection models. But while ABR is a powerful addition, it’s important to understand where it truly fits and where it doesn’t. When I first read the blog posts from Michael Cade and Anthony Spiteri, my first use case were network and IoT devices as they both mention it.

In particular, the idea that “network devices and IoT systems” can directly use ABR is more theoretical than realistic for most devices. The majority of those devices do not have an NFS client to write to an NFS share directly. That said, we can leverage the Application Backup Repository to improve backup and recovery services to our (internal) customers.

This post gives you a clear, realistic overview.

What ABR Actually Is

ABR is a feature of the Veeam Hardened Repository appliance. It exposes managed NFS volumes that applications can write their own backup files to. Each application receives its own isolated NFS export, and Veeam handles:

  • Snapshot creation and retention
  • Immutability enforcement
  • Instant rollback to any snapshot
  • Backup copy to secondary storage
  • Secure access control (Kerberos optional)

This allows application teams to continue using their native backup tools, such as RMAN, Backint, pg_dump, MongoDB dump, MySQL dump, custom scripts, while gaining Veeam’s hardened storage and recovery workflows.

Where ABR Shines

ABR is ideal for workloads that already support writing backup files to NFS:

  • Oracle RMAN Incremental Merge
  • SAP HANA Backint or file-based backups
  • PostgreSQL (pg_dump, WAL archiving)
  • MongoDB / MySQL / MariaDB dumps
  • Linux-based application exports
  • DevOps pipelines writing artifacts to NFS

These workloads benefit from ABR’s snapshot rollback, immutability, and backup copy integration without changing their native backup processes.

Network appliances and IoT device limitation

When it comes to most network appliances and IoT devices we are confronted with the limitation that they cannot write to NFS as they lack:

  • an NFS client
  • POSIX-compliant file handling
  • kernel-level NFS support
  • secure session handling (Kerberos, AUTH_SYS)

Typical network gear supports:

  • TFTP
  • FTP / FTPS
  • SFTP / SCP
  • HTTPS API upload
  • Proprietary backup formats

Typical IoT devices support:

  • local flash storage
  • MQTT / HTTP upload
  • cloud sync
  • proprietary export formats

NFS is almost never supported. So while ABR can technically accept NFS writes from any capable device, the practical reality is that network and IoT devices will not use ABR directly. This is something to keep in mind.

A Practical Workaround: How to Still Protect Network & IoT Backups with ABR

Even though network/IoT devices cannot write directly to NFS, you can still protect their backups using ABR by introducing a simple intermediary.

Workflow

  • Network/IoT device exports its backup (TFTP, SFTP, SCP, API, etc.)
  • A Linux server or automation node collects these backups
  • That server or servers write(s) the collected files into the ABR NFS share
  • Veeam handles snapshots, immutability, and backup copy from ABR

Benefits

  • Hardened storage for device backups
  • Snapshot rollback
  • Immutability
  • Backup copy to secondary storage
  • No unprotected local storage
  • This is the realistic way to use ABR for network gear and IoT systems.

Pricing

Well, it could work out to be cheaper than a NAS license for Veeam, but maybe not if you only store config file backups on there. I mean 1TB goes a long way for config backups only.

That said, in both cases you pay some money to have additional Veeam capabilities and immutability with granular and backup admin-managed recovery options. That might be the reason you want it over just backing up one or more virtual machines with the network appliance and IoT device backups on it. Think about it, a backup admin can provide a network engineer with a secured download of an encrypted zip at zero-dark-thirty in the morning without needing to restore files from a virtual machine, which might require a recovery key etc. Does that fit your security rules? Maybe, maybe not. But that is up to management to decide when they are down an extra 6 hours waiting for the file(s) from inside the Virtual Machine to be restored.

Conclusion

The Application Backup Repository is one of the most interestingl additions to Veeam v13.1, especially for enterprise databases and Linux-based applications that already support NFS. It brings hardened storage, immutability, and snapshot-driven recovery to workloads that previously lived outside Veeam’s ecosystem.

However, ABR is not a direct backup target for most network devices or IoT as far as I can see. These devices simply do not support NFS. The good news: with a lightweight collection server acting as an intermediary, you can still bring their backups into ABR and benefit from its security and lifecycle management.

ABR is powerful — but only when used where it actually fits. All this is lab and theory, it is early days yet and I am sharing my thinking process out loud with you. Nothing more, nothing less.