Skip to main content

Micro-segmentation

Enterprise Feature

Micro-segmentation is part of Network Security, available exclusively with an Enterprise license. The required feature flag is microsegmentation.

The Micro-segmentation tab of Network Security turns the Proxmox VE firewall into a zone policy you can deploy safely. You group guests into zones, declare the flows allowed between them, watch the real traffic in observation, then block zone by zone once nothing unexpected remains. ProxCenter compiles the policy into native Proxmox firewall objects, so Proxmox keeps enforcing it even when ProxCenter is down.

Micro-segmentation Overview on Production Cluster: the Groups map with the zones monitoring, web, app, db and infra drawn as circles holding their guests, the unzoned guests in a dashed circle, Unknown / External on the left, red dashed unprotected flows, blue blocked flows and green allowed flows labelled with their service, and the sub-tabs Overview, Policy, Observation, Drifts and History with Check now and Emergency stop
The Overview map: zones as circles holding their guests, allowed flows in green, blocked flows in blue, and observed flows that no rule allows in red dashes.

How it works​

ConceptWhat it is in Proxmox
ZoneA group of guests. Membership is a Proxmox tag on the guest: the tag prefix (zone- by default) followed by the zone name, for example zone-web. Tag a new guest and the engine protects it at its next pass, with no other step.
FlowTraffic the policy allows: from a zone or a CIDR, to a zone or a CIDR, on a protocol and ports. Both ends cannot be CIDRs.
ExceptionA narrow opening between chosen VMs, on named services, whatever the flows and the isolation of the zone say.
Inbound stateDisabled (not applied), Observing (traffic outside the policy is logged, nothing is blocked) or Blocking (only the flows of the policy get in).
Outbound stateNot controlled, Observing or Blocking, with a filter to the Local network or to the Local network and Internet.

ProxCenter keeps the policy as a draft. Nothing is written to Proxmox until you review the plan and apply it; each apply records a revision you can restore later.

Everything ProxCenter creates in Proxmox is named pcz-* or carries a comment starting with [pc-microseg]: a cluster security group and an IP set per zone, plus the shared sets pcz-_local and pcz-_known used by the outbound control. On each member, ProxCenter inserts the zone group at the top of the guest rules, turns the firewall on for each network card and sets the guest options. The engine never changes or deletes an object without that marker. Rules you add by hand stay where they are: your DROP rules keep applying, and your ACCEPT rules show up as drifts. A Proxmox object that uses a pcz-* name without the marker blocks the apply until you rename or delete it.

info

Zone names are 1 to 14 characters: lowercase letters, digits and dashes. The limit comes from Proxmox, which caps security group names at 18 characters, pcz- included.

Requirements​

  • The ProxCenter orchestrator runs the engine. On a high-availability install, only the leader orchestrator writes to Proxmox.
  • The cluster firewall must be on. The pre-flight checks say so and open the deployment wizard when it is off.
  • Guests need a known IP address, so that the zone IP sets list them: from the QEMU guest agent, the cloud-init ipconfig, or the container's ip= setting. Members without one appear in Drifts.
  • Micro-segmentation is a provider feature: a vDC tenant does not see it.

Getting started​

On a connection with no policy yet, the tab shows three steps:

  1. Run the pre-flight checks. Check that the cluster and its firewall are ready.
  2. Create your first zone. Group the guests that talk to each other.
  3. Preview the plan. See what would change on Proxmox before anything is written.

Then observe for a while, accept the flows you recognise, and block zone by zone. The recommended order:

  1. Create the zones and assign the guests (Policy > New zone, then Assign guests).
  2. Declare the flows you already know (New flow).
  3. Apply the plan with the zones in Observing.
  4. Review Observation daily: accept what is legitimate, dismiss the rest.
  5. When a zone has seen no pending traffic that would be blocked for the observation window (24 hours by default), it shows as ready. Switch it to Blocking.
  6. Do the same for the outbound traffic, zone by zone, if you need it.

Pre-flight checks​

Policy > Pre-flight checks reads the cluster and reports each check as passed, warning or failed.

Pre-flight checks dialog: Cluster firewall passed (the cluster firewall is on), Management access passed (Web 8006 and SSH 22 stay reachable), Log rate limit warning (PVE logs at its default rate of 1 per second, so the observation misses flows) with a Raise while observing button, and ProxCenter address passed (ProxCenter is outside the active zones)
The pre-flight checks, here with the default log rate limit flagged.
CheckWhat it verifies
Cluster firewallThe Proxmox cluster firewall is on. When it is off, Open the deployment wizard reads the nodes, then turns the cluster firewall on and changes nothing else: no rule, no policy. It names the nodes whose own firewall would turn on with the cluster, so you can check their management access first.
Management accessThe web UI (8006) and SSH (22) stay reachable from ProxCenter on every node whose own firewall is on, through a host rule, the cluster rules, or the Proxmox management IP set.
Log rate limitProxmox logs one line per second per chain by default, so the observation misses flows. Raise while observing sets a higher limit while a zone is observed, and sets it back afterwards.
ProxCenter addressProxCenter is not inside a zone it could be cut off by. For a zone that controls its outbound traffic, allow a flow to the Proxmox nodes before blocking.
Outbound subnetsEvery member of a zone with outbound control has a known subnet. A member without one has its local network reduced to its own addresses.
warning

When ProxCenter runs in Docker, the address it detects can be internal to Docker. In Settings > Advanced, enter the address of the host as Proxmox sees it, or the management access check cannot tell whether ProxCenter would be cut off.

Overview​

The Overview sub-tab draws the policy and the observed traffic. Six renderings sit in the toolbar:

RenderingBest for
GroupsZones as circles holding their guests
Compact zonesOne box per zone, for many zones
MatrixSources by destinations, for any number of zones
Guest viewOne zone and its guests, with what talks to them
Sankey diagramThe observed traffic as ribbons, wider with more hits
NodesThe guests on their Proxmox node, with the observed flows and the exceptions from node to node

The Flows checkboxes show or hide the Unprotected (observed, allowed by no flow), Blocked and Allowed edges. Target a VM (name, VMID or IP) keeps the map focused on one guest and lists its observed flows and the allowed flows of its zone. Allowed flows carry no counter: Proxmox does not log what it lets through.

Matrix rendering: source zones app, db, infra, monitoring, web and Unknown / External in rows, destination zones and Outside in columns, each cell holding the services allowed in green and the services observed outside the policy in red dashes with their hit counts, such as 2,260 hits of tcp/3100 from app to monitoring
The Matrix rendering reads at a glance which zone pair carries traffic that no flow allows.
Nodes rendering: one circle per Proxmox node, pve-node-01 to pve-node-12, each with its guest count and its guests coloured by zone, and the observed flows between nodes such as tcp/3100 and MySQL
The Nodes rendering places each guest on its Proxmox node, for questions about where the traffic runs.

Click a zone to open its panel: observed traffic over 1 hour to 7 days, inbound and outbound, the top sources and destinations, the pending suggestions with their actions, the members, the flows and the exceptions. The panel actions edit the zone, change its state, block or reopen its outbound traffic, and Expand opens it in a larger dialog with Traffic, Data and Logs tabs. Clicking an edge between two zones opens the same kind of panel for that pair.

Zone panel of app next to the map focused on its flows: one drift chip, the state icons, edit, outbound, delete and expand actions, the observed traffic chart for the last hour with Both, Inbound and Outbound, the top destinations Outside and monitoring, three pending suggestions (Many destinations HTTPS 5,120 hits, 10.20.90.12 Proxy and 151.101.1.69 HTTPS) and the members app-prod-01 to app-prod-03
The zone panel gathers the traffic, the suggestions and the members of one zone.

Policy​

The Policy sub-tab holds the zone cards and the flow table.

Policy sub-tab: cards for the zones web, app, db, monitoring and infra with their state (Enforced, Observation, Disabled), drift count, members, incoming and outgoing flow counts, member chips, a summary of their flows and a 24-hour sparkline of uncovered traffic split inbound and outbound, then the Flows table grouped by destination zone with Applied states and the Uncovered 24 h chip
Each zone card shows its state, its members, its flows and 24 hours of traffic that no flow covers.

Each zone card shows the inbound state, the outbound state, the drifts of its members, the member chips, its flows, and a sparkline of the last 24 hours of traffic that no flow covers, inbound and outbound. Its actions edit the zone, change its state, block or reopen its outbound traffic, add a flow into it, assign guests, and delete it.

The flow table groups the flows by destination zone. Each row has a state: Applied, Pending add, Pending change, Pending removal, or Not applied for a flow to a CIDR whose source zone does not control its outbound traffic. Learned marks a flow accepted from the observation. The group header shows the traffic into the zone that no flow covers over the last 24 hours.

Zones​

New zone or Edit zone opens the zone dialog.

Edit zone app dialog: Name, Label Applications, Color, Isolate members from each other unchecked, VM exceptions with Manage, Gateway 10.20.0.1, Allow to the gateway DNS, NTP and Ping, Inbound Blocking, Outbound Observing, and Filter to Local network and Internet with the warning that IP filtering cannot follow a service behind a domain name
The zone dialog sets the inbound and outbound states, the outbound filter, the gateway and the exceptions.
FieldEffect
Isolate members from each otherUnchecked, the members of the zone talk to each other freely. Checked, they reach each other only through a declared flow or an exception.
VM exceptionsOpens the exception table of the zone.
GatewayThe router of the zone's network, optional. It is named on the map and in the Observation.
Allow to the gatewayServices allowed to the gateway: DNS, NTP and Ping are offered first. Pick in the list or type a protocol and its ports, such as tcp/8443, udp/5000:5010, icmp or any.
InboundDisabled, Observing or Blocking.
OutboundNot controlled, Observing or Blocking. Requires the inbound to be observing or blocking.
Filter toLocal network filters outbound traffic to the subnets of the zoned VMs, unzoned VMs included, and leaves Internet open. Local network and Internet filters everything: only the flows of the policy go out.
warning

Outbound filtering works on IP addresses only. A service reached through a domain name can change address and be blocked: allow it with a wider CIDR, or keep that zone on Local network.

Moving a zone to Blocking, inbound or outbound, asks you to type the zone name. The dialog says whether the zone is ready: nothing would have been blocked during the observation window. Ready is an indicator, not a lock. A state change is applied to Proxmox right away as a new revision, so it is refused while the draft holds other changes: apply or discard them first.

Assigning guests​

Assign guests lists the guests of the connection with a search by name or VMID. Assigning a guest that already belongs to another zone moves it. The tags are written on the next apply; until then the member shows as Pending. A guest carrying the tags of several zones is left alone until it is assigned to a single one.

Flows​

New flow takes a source (zone or CIDR), a destination (zone or CIDR), a protocol, ports (80, 443, 8000:8100, or empty for all ports) and a comment. A flow from zone A to zone B opens the inbound of B and, when A controls its outbound traffic, the outbound of A: you declare it once.

Exceptions​

An exception lets chosen VMs reach chosen members of the zone on the services you name, whatever the flows and the isolation of the zone say. Use it for one administration host, a backup agent or a single database client, without opening a whole zone pair.

New exception dialog with Source VMs app-prod-01, Destination VMs db-pgsql-prod-01, Allowed services PostgreSQL, and the count of 1 rule
An exception names its source VMs, its destination members and its services.

An exception holds at most 50 rules; past that, allow a flow of the zone instead. A VM cannot be both a source and a destination. The exception table of the zone shows a status for each exception that is not active:

StatusMeaning
Zone not appliedThe zone is neither observing nor blocking: nothing of the exception is written.
No destination in the zoneNone of its destinations is a member of the zone yet.
VM not foundA source or a destination is no longer in the inventory.
Address unknownThe inventory knows no address of a source yet.
Name changedA VM was renamed, or its VMID was reused: check that the exception still names the right VM.

The map draws exceptions as dashed edges between guests, and the Firewalling tab shows their rules as engine rules, edited from their zone.

Plan and apply​

Every change of zones, flows, exceptions, members or tag prefix lands in the draft. A strip lists what is not applied yet, with Review and apply and Discard changes. The plan reads Proxmox and lists every write, grouped into cluster objects and one section per guest, before anything is written. Apply runs the steps one after the other and shows each step as waiting, being written, done, failed or not run.

The plan also reports:

  • Proxmox objects that use a ProxCenter name without being ProxCenter's: the apply is blocked until you rename or remove them.
  • Rules added by hand in ProxCenter's security groups: they are kept, not touched.
  • Guests that could not be read: they are handled at the next pass.

Observation​

The Observation sub-tab lists the traffic the firewall logged that the policy does not cover, grouped by zone.

Observation sub-tab with the Type, Status, Zone and Direction filters, the legend, and observed flows grouped by zone, each row with its direction arrow, source, destination, service, hit count, last seen time, a Would be blocked or Blocked status and the accept, accept as CIDR and dismiss actions
Each observed flow can be accepted into the draft, accepted as a CIDR, or dismissed.
  • Would be blocked: seen in observation, outside the policy.
  • Blocked: actually dropped by a zone in blocking.

Filter by Type, Status (Pending, Accepted, Dismissed), Zone and Direction (Inbound, Outbound). Each row offers:

ActionEffect
AcceptCreates the flow between the two zones in the draft.
As CIDRCreates the same flow with the source address as a range, needed when the source is in no zone. For an outbound destination, you choose the prefix length.
Allow this flow onlyFor a flow between two known VMs: opens a pre-filled exception instead of opening the whole zone pair.
DismissHides the suggestion for this pair.

Select several rows to accept, accept as CIDR or dismiss them in one action. An outbound flow accepted towards a zone also allows the inbound of its destination zone.

Details opens the flow: direction, destination, service, hits, first and last seen, a sample pair of addresses, 24 hours of traffic, and the recent occurrences with the guest reached and its node.

Observed flow dialog from 192.168.1.57 to monitoring on Grafana, Would be blocked: summary with direction Inbound, destination monitoring, service Grafana tcp/3000, 870 hits, first and last seen, the sample addresses, a 24-hour bar chart of 514 hits, and the recent occurrences reaching monitoring-01 on pve-node-04
The detail of an observed flow, with its last 24 hours and the guests it reached.
info

The observation is sampled. Proxmox logs at most one line per second per chain by default, so busy flows are under-counted, and the engine keeps one sample pair of addresses per observed flow: the hits count the whole flow, not only that pair. Raise the log rate limit in the settings while you observe.

Drifts​

The Drifts sub-tab lists the differences between the policy and Proxmox found by the last full check, grouped by kind.

Drifts sub-tab as of the last full run, with the All, No zone, Firewall off and Foreign rule filters and three groups: a running guest in no zone with Assign to a zone, a guest whose firewall is off on net0 with Open the plan, and an accept rule at position 3 not created by ProxCenter
Drifts group the guests and rules that differ from the policy, with the fix each one takes.
DriftFix
Running guest without a zoneAssign to a zone, or Ignore.
Network card firewall offOpen the plan: the apply turns it back on.
Rule not managed by ProxCenterAn ACCEPT rule added outside micro-segmentation opens the guest or the group: review it in Proxmox. ProxCenter keeps it as it is.
IP set out of sync, Local network set out of syncRepaired at the next pass.
Member without a known IPInstall the QEMU guest agent, or set an IP in cloud-init or in the container network.
Guest in several zonesKeep a single zone tag on the guest.
Name already used in ProxmoxRename or delete the unmarked pcz-* object in Proxmox.
Outbound rule missing, misplaced or at the wrong verdict, Known destinations rule missing or in excessRepaired at the next pass.
Outbound without a known subnetCheck the guest network configuration in Proxmox: an IP with its prefix length, or the QEMU guest agent.

Ignored drifts are hidden behind Show ignored.

History​

The History sub-tab lists the revisions and the engine runs, newest first, with their actor, summary, result and duration. Restore on a revision replaces the draft with that version and applies it to Proxmox. Each run opens its detail: API calls, writes, steps and errors.

History sub-tab with the All, Revisions and Runs filter: light check, log read, full check and IP read runs by the engine with their writes, API calls and durations, revisions v7 and v4 applied by an administrator with Restore, and an apply run that failed because Proxmox answered 500 on one node
Revisions and engine runs side by side, with Restore on each revision.

Checks and load on Proxmox​

The engine runs on the orchestrator, in four loops:

RunDefault intervalWhat it reads
Light check300 sThe cluster IP sets, security groups and firewall options
IP read900 sThe IP addresses of the zone members, through the guest agent or the container interfaces
Full check1800 sThe firewall of every member, and the drifts
Log read60 sThe firewall log of each node, since the previous read, only while a zone observes or blocks

The engine writes to Proxmox only when it finds a difference, one write after the other. Past the write cap per run (50 by default), it stops and holds the remaining writes until you confirm them in the banner. Check now queues a full check at the engine's next pass, within 60 seconds, and shows what it read, compared and repaired.

Emergency stop​

Emergency stop is always visible on the tab. It disables the Proxmox cluster firewall with a single request, effective immediately on every node, and moves every zone of the connection to observation.

Stop micro-segmentation dialog: every zone of this connection moves to observation, it disables the Proxmox cluster firewall with a single request, every firewall rule stops being evaluated including the ones ProxCenter does not manage, the micro-segmentation objects stay in place and Resume re-enables the cluster firewall, with Cancel and Emergency stop
The emergency stop turns the whole cluster firewall off, not only the micro-segmentation.
danger

While the stop is active, every firewall rule of the cluster stops being evaluated, including the rules ProxCenter does not manage: management rules, host rules and guest rules. Use it when the segmentation blocks something it should not, then fix the policy and resume.

The micro-segmentation objects stay in place. Resume re-enables the cluster firewall and re-applies the last revision. It is refused while a node whose own firewall is on does not let the web UI (8006) and SSH (22) in from ProxCenter: allow them in that node's firewall, or turn it off, first.

Settings​

The gear icon opens the micro-segmentation settings of the connection.

Micro-segmentation settings dialog: General with Tag prefix zone-, Load on Proxmox with the light check (300 s), guest IP read (900 s), full check (1800 s) and firewall log read (60 s) intervals, Enforcement with 24 hours of observation before blocking and a write cap of 50 writes, Observation with the log rate limit and burst left unchanged, Notify me about Drifts, Engine paused, Unprotected guests and Blocked flows, Advanced, and Removal with Remove
The settings set the tag prefix, the engine load, the observation window, the log rate, the notifications and the removal.
SectionSettings
GeneralTag prefix: 1 to 16 characters. Changing it makes guests tagged with the old prefix leave their zones at the next apply, until they are tagged again.
Load on ProxmoxThe four loop intervals.
EnforcementObservation before blocking (24 hours by default, inbound and outbound) and Write cap per run.
ObservationLog rate limit and Log burst: empty leaves the Proxmox values untouched; otherwise they are set while a zone is observed.
Notify me aboutDrifts (each new drift, at most once an hour), Engine paused, Unprotected guests, Blocked flows. The notifications go to the channels of Settings > Notifications.
AdvancedAddresses Proxmox sees: the addresses ProxCenter reaches the nodes from. Leave empty to use the detection; the installers record the host addresses.
RemovalRemove micro-segmentation.

Removing micro-segmentation​

Remove… asks you to type the connection name, then releases every guest, deletes the objects ProxCenter created in Proxmox, and restores the log rate limit. The zone tags stay on the guests and the draft is kept, so you can apply it again later. If the removal reports incomplete, run it again.

Access​

PermissionAllows
node.view on the connectionRead the policy, the map, the observation, the drifts and the history
node.manage on the connectionEvery change: draft, apply, state changes, emergency stop, restore, settings, removal

Without node.manage, the controls stay visible but disabled. Every apply, state change, emergency stop, restore and removal is recorded in the audit log under the user who made it.

The public API does not expose micro-segmentation routes.