Tech Blog

everyWAN Blog

Technology, cybersecurity and IT trends that matter

Deep Analysis
Cybersecurity
IT Trends
Filter by:
Macro del interior de un disco duro abierto: el plato y el brazo del cabezal
9 min read

CephX trusted the ticket it had just decrypted, and the fix has no CVE

Ceph v20.2.5 Tentacle, released on 2 November, carries a single security line: "CephX: hardened service ticket verification". It is the second half of the fix you rushed through in August with four CVEs breathing down your neck, and this half has no number, so it never enters a patch queue. Inside: the four checks the verifier no longer skips, why the 300-second tolerance says nothing about your NTP, the date each branch gets it (Squid gets it in its final release) and the ten-minute read-only audit.

Frontal de varios servidores en rack con decenas de bahías de disco, unas cuantas de ellas vacías
9 min read

The ESXi import wizard copies what you provisioned, not what you use

A Proxmox forum thread leaves three figures from the same case: a 5 TB virtual disk, 255 GB actually in use, a good eight hours of import over a single gigabit, and 260 GB on the destination ZFS. They do not contradict each other: the ESXi import wizard sizes your migration by what you provisioned, because the vSphere API does not tell it which blocks are written. Inside: the arithmetic to estimate the window before you promise a date, why there are two bills and almost everyone only looks at one, which destinations absorb the zeros, which machines never go through the wizard at all, and when none of this applies to you.

Sala de servidores real con racks abiertos, latiguillos azules y una escalera de tijera olvidada en el pasillo
8 min read

Proxmox's SDN patch now requires privileges it did not: update and somebody gets a 403

PSA-2026-00092-1, one of the fifteen advisories Proxmox published between 10 and 12 November 2026, does more than plug holes in the SDN API's ACL checks: it changes the permission contract. The advisory says so outright —"it was not, but is now required to have the SDN.Use or SDN.Allocate privilege…"— so when you move libpve-network-api-perl to 1.6.8, an operator with delegated zones who created subnets fine yesterday may start getting refusals. Inside: what to grant and where before you touch anything, why this needs a window rather than an unattended run, what to patch first from the rest of the batch, and the two advisories that fit no comfortable reading.

Sala de servidores con bastidores abiertos medio poblados y paneles de parcheo
13 min read

The API for declaring your VMs is now neutral; the provider that fulfils it is not

On 9 November Broadcom took VM API into the CNCF Sandbox: Kubernetes CRDs for declaring a virtual machine the same way anywhere. The next day, on the same blog, the sentence that qualifies it. Inside: the two readings of the same document, the contract field that still carries no netmask, the Proxmox VE 9.2 provider that already exists and has gone thirty-five days without a reply, the KubeVirt question that is also unanswered, and the five questions to ask before this goes into a contract.

Maleta de transporte con veinte huecos de espuma para discos de 3,5 pulgadas y solo cinco discos dentro
9 min read

Growing Ceph through a single node grows nothing

Three nodes weighted 100, 100 and 1,138 add up to 1,338 of disk. The pool stores 95 and leaves 1,043 — 78% of the raw — where nobody can use it. Inside: the one-line rule that decides your usable capacity in Ceph, the formula to take to the purchasing meeting, why switching the balancer off costs you 7%, and why the same dead disk is paperwork with four nodes and a degraded cluster with three.

Filas de bahías de disco SAS en servidores de rack: el hierro sobre el que corre un cluster Ceph
7 min read

Ceph Umbrella is already in Proxmox, but not for your cluster

Proxmox's Ceph wizard now offers Umbrella (21.2) as a technology preview. The decision is not set by any end-of-life date: it is set by 20.2.4, a number that shows up twice from two directions that never meet. Inside: commit 18c8b132 of 4 November, which gave Squid 32 days and took 149 off Umbrella while announcing neither, what we measured across Proxmox's three repositories (77 packages against 486, and a single version to fall back to), why this is not one window but two, the wiki sentence that decides whether you can touch it today, and what the CPU headline does not mean for you.

Campo de conexión de fibra óptica en un armario de distribución: bandejas con latiguillos azules y amarillos ordenados
6 min read

Everpure supports Proxmox VE; the plugin is still being written

On 5 November 2026 Proxmox announced that Proxmox VE "is supported across the full Everpure FlashArray™ family". The words snapshot, clone, provisioning and plugin never appear in the release: those are on Everpure's blog, in the present continuous ("we're already working on it") and with no date. Inside: the two sentences the release does make, what you actually get today from a shared LUN, why the snapshot-as-volume-chain flag is a technology preview that only applies to newly created disks, the four storage announcements of 2026 and the title that changes, why Pure Storage is now called Everpure, and the three questions we ask before signing off a design.

Panel frontal iluminado de un servidor de rack con su botón de encendido
9 min read

Proxmox's load balancer only moves HA resources, and declaring HA arms a 60-second watchdog

Proxmox VE 9.2 ships dynamic load balancing through the CRS, and it's been reported as the cluster finally balancing itself. The documentation says those migrations are issued "only for HA resources" and run one at a time: for the balancer to touch a VM you must declare it HA, and a node with active HA resources keeps its watchdog armed and reboots 60 seconds after losing quorum. Inside: the three CRS modes and why the one you have is the wrong one, the factory values (30% threshold, 3 rounds of about 10 s, 10% margin, bruteforce or topsis), why the balancer's reach is smaller than its blast radius, what it won't move even if you declare it, and six questions to answer before flipping the switch.

Una cabina de almacenamiento junto a bastidores abiertos en una sala de servidores real
9 min read

Eight shared storage types in Proxmox, and only two without small print

On 30 October Lightbits announced its native plugin for Proxmox VE and Proxmox's COO summed up the catalogue with a friendly sentence. The official documentation table says the same thing more precisely: eight shared storage types, and only three with no footnote in the snapshots column. Open the footnotes and two are left. We walk through the real routes, the Proxmox 9 switch that only applies to disks you create after turning it on, and what changes when you put someone else's code inside PVE::Storage::Custom.

Pasillo de un centro de datos con bastidores llenos de servidores de 1U y algunas posiciones ya vacías
8 min read

They pulled the VDDK, and your VMware-to-Proxmox exit never ran through it

Around 25 August the public VDDK downloads vanished without notice and it was read as VMware's exit door being shut. Some migration routes read disks through that library and some never touch it; the Proxmox importer is one of the latter. We check it three ways and explain why the most obvious one — opening the package dependencies — proves nothing: the VDDK cannot be redistributed, so no Debian package can declare it, not even the ones that do use it. Also here: the six things the importer does NOT do, and the wall both routes share, vSAN.

Armario de comunicaciones con decenas de latiguillos de red azules y grises cruzados sobre los paneles de parcheo
8 min read

Seven fixed versions in the CVE and none of them boots your Proxmox

CVE-2026-98432 is a 9.8 in the Linux kernel: a use-after-free in the IPv4 TCP handshake, exploitable with no credentials and no interaction. The advisory lists seven fixed versions — 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56 and 7.2.10 — and not one matches the branch a Proxmox node boots. What actually triggers the bug according to the CVE's own field-by-field justification, the clocks of Debian, Ubuntu, Red Hat and Proxmox as of 31 October, what the published exploit proves and what it does not, and six checks with commands.

Filas de discos duros de 3,5 pulgadas montados en un chasis de almacenamiento abierto
8 min read

On Proxmox, the jump to Ceph Tentacle bills you for what you built in 2020

On 31 October the Squid branch of Ceph stops receiving backports, and plenty of hyper-converged Proxmox clusters have the jump to Tentacle on this week's calendar. The procedure works. What nobody writes into the maintenance window is the bill that arrives the day after: OSDs created under Octopus or earlier can grow their META until the Proxmox panel shows less space than you had, two manager modules that had been running since 2020 disappear, the MDS step asks you to give up your standby, and the specific pair of versions between source and target may never have been tested. Our own measurement of the Proxmox repositories, the four possible situations, when NOT to jump this week, and the six checks to run before touching the repository.

Armario de comunicaciones mural de una sede con panel de parcheo, switch y router
9 min read

Proxmox across a thousand sites is not one cluster: it is a thousand clusters

On 26 October 2026 it was reported that Tesco is replacing VMware with Proxmox in its stores. The headline is the hypervisor; the real detail is in the job ad, which does not ask anyone to build a cluster but an "Edge Management Platform" to provision, manage, monitor and maintain the store estate throughout its lifecycle. The reason is a number from Proxmox's own documentation: corosync needs latencies under 5 milliseconds between all nodes, which rules out anything that is not a LAN. With the per-socket arithmetic that flips sign at the edge, the minutes of data loss that ZFS replication brings with it, what the Datacenter Manager still does not do according to its own roadmap, and the three-bucket inventory for deciding what stays in the site and what should never have been there.

Bastidor con varios servidores apilados y sus bahías de disco SAS a la vista, en penumbra
9 min read

VMFS alternatives on Proxmox: no release has been verified on a storage array

This week two clustered filesystems started circulating that do on a shared LUN what VMFS does for vSphere: CetraFS and MeshXFS. It is the missing piece in every migration away from VMware where the array still has life in it. Both are free software, both package for Debian, and both write in their own README why they do not yet belong underneath your machines. The deciding sentence is MeshXFS's: "no release has been verified on a storage array". With the arithmetic of the 120-second stale window against multipath's stock 90, who kills the partitioned host once vSphere HA is gone, why Proxmox's own option has spent 446 days in technology preview, and seven checks before letting any of this near production.

Bahías de disco de una cabina montada en rack, con unidades etiquetadas y varios huecos vacíos
6 min read

The last Ceph Squid patch is not in your Proxmox repository

On 12 October 2026 Ceph shipped 19.2.7, the seventh backport in the Squid series, with the usual line: "we recommend that all deployments update to this release". Today we checked what the Proxmox repository actually serves, and the answer is 19.2.6-pve4, with an index dated 9 September. The branch dies on 31 October. Two different calendars, and the one that governs you is not the one on ceph.io.

Parte trasera de un armario de servidores con los latiguillos de red saliendo de cada nodo hacia los de al lado
10 min read

The switchless Ceph mesh has four variants. Two of them break on one cable

Three nodes wired directly to each other at 25 GbE, with no switch to buy, is one of the best decisions you can make on a small budget. Proxmox documents four ways to wire it and only two survive a single link going down: in the other two, its own wiki says, "the nodes now have a different view" and Ceph reports services as down that are perfectly alive. Since Proxmox VE 9 that mesh is configured from the web interface, with four fabric protocols. And the official documentation still labels that layer tech preview.

Pasillo de un centro de datos con un rack abierto, un carro de consola con monitor y teclado, y una silla vacía
9 min read

A VM escape with no CVE and no patch: what root on a Proxmox node reaches

On Tuesday 6 October 2026 The Register reported that a researcher claims a guest-to-host escape in KVM, that Vercel's CEO confirmed it inside their bounty programme, and named Proxmox among those relying on KVM. There is no CVE, no patch and no technical detail. Since there is nothing to patch, what is measurable today is the radius: Proxmox VE's documentation says QEMU runs as a root process, so it is one hop and not two; and root on one node reads /etc/pve/priv, which pmxcfs replicates in real time to every node. Inside: the table of ten files with their official descriptions, why the backup password sits there "in plain text" next to the encryption key, and the rotation list, line by line, with what each one costs.

Parte trasera de un rack con servidores de 1U, sus puertos de consola a la vista y los mazos de cables recogidos
10 min read

Moving to Proxmox: permissions do not travel, and the role that sounds like "user" opens the console

The migration brings the disks, the configuration and the networks. It brings not one line of the permissions table, and because the two tables look alike in shape, almost nobody rebuilds theirs: they copy it from memory, going by the names of the roles. Inside: the mapping between a vCenter permission and a Proxmox ACL, with its three gaps; the literal definition of the PVEVMUser role —"view, backup, configure CD-ROM, VM console, VM power management"— and why console, CD-ROM and power together do not describe a reader; the question only your installation can answer; the inheritance rule that subtracts instead of adding and bites everyone arriving from Active Directory; the family of privileges that reads and writes files inside the guest, has no mental equivalent in vCenter and is a virtue of the model that nobody uses; root@pam, the unconfined account that cannot be deleted; and the four commands that tell you, the following Monday, who can do what on your cluster.

Filas de bahías de disco SAS en servidores de almacenamiento montados en rack
13 min read

You set noout to upgrade Ceph and switched off re-replication

The Ceph upgrade procedure on Proxmox recommends, as its fourth step, that you run "ceph osd set noout": "optional, but recommended", says the wiki. Ceph's own OSD troubleshooting page says this is "more a thought exercise" than a suggestion that anyone "in the post-Luminous world" should run it. Both sentences are published today and both are defensible. What that flag switches off is not rebalancing: it is a "down" OSD ever reaching "out", which is precisely the step that makes CRUSH recompute and recreate the missing copy. And the flag does not tell the disk you rebooted apart from the disk that genuinely died. Inside: why mon_osd_down_out_subtree_limit already brakes the scenario that actually frightens you, the chain almost nobody spells out (degraded PGs are not flagged as clean, and what is not clean is not scrubbed, so checksum verification stops on exactly the data that just lost a copy), the arithmetic of the two warnings arriving at 10.5 and 12.25 days to a dashboard that has been amber since minute one, the table of what each flag really switches off, and the four commands that do the same job on one disk or one host without touching the rest of the cluster.

Sala de servidores con racks cableados y un módulo de gestión de potencia
9 min read

Moving to Proxmox 9.2 wasn't your call: the repository made it

Ask whoever runs your cluster who decided to go from 9.1 to 9.2, and in which meeting it was approved. In most places nobody decided: Proxmox VE ships no repository per minor version —there is nothing like a pve-9.1-security— so a routine apt dist-upgrade, the same one applying the patches you do want, leaves you on the latest published minor. The software that takes your backups does have a supported-version matrix, and it runs behind: Proxmox VE 9.2 shipped on 21 May 2026 and the Veeam plug-in covering it arrived on 29 July, 69 days later. And one detail misleads: plug-in 3.3 for the 13.0 branch shipped on 25 August, twenty-seven days AFTER 4.0, so "I am up to date" can mean up to date on the branch that does not carry the new support. Nothing turns red; what you lose is the right to open a case. The three facts that must live on the same line before you open the window, why the right order is matrix first and apt second, the ARM architecture boundary, and where this crack does not exist.

Subscribe to our newsletter

To receive IT stories, everyWAN news and exclusive subscriber offers, sign up to our mailing list

Minorisa de Sistemas Informaticos y Gestión S.L. © 2026
everyWAN
everyWAN