Home › Topics › VPN & Remote-Access Monitoring › VPN Troubleshooting Playbook

A VPN Troubleshooting Playbook: From "VPN Is Down" to Root Cause


"The VPN is down." Four words, and they could mean anything: one user's cat stepped on their router, or four hundred people are locked out of everything the company runs. VPN outages have a special cruelty to them. The people affected are, by definition, not in the building, so nobody can walk to your desk with useful details. The affected population is scattered across a hundred home networks you cannot see. Every minute of confusion is a minute of the whole remote workforce standing still.


The difference between a miserable VPN incident and a brisk one is almost never cleverness. It is order. A VPN connection succeeds only when a chain of separate systems all work. Those systems are the user's internet, your circuit, the gateway service, the authentication backend, the address pool, and the network behind it all. A troubleshooter who tests that chain in sequence finds the broken link in minutes. One who guesses jumps between links until luck intervenes.


This is the playbook for testing in sequence. It starts with the one question that cuts the problem in half. It walks the shared chain stage by stage with a test for each. It lists the classic culprits worth checking on sight. And it covers the two things playbooks usually forget. One is how to communicate with a workforce whose usual channels may be down. The other is what evidence to collect for the escalation you may need to make. Print it, or at least steal its order.


In this guide

  • Question zero: one user, some users, or all?
  • The shared chain, stage by stage
  • A test for every stage
  • The classic culprits
  • A worked example: quarter to nine on a Tuesday
  • Communicating during a remote-work outage
  • Evidence for escalation
  • The playbook on one page

Question zero: one user, some users, or all?


Before touching anything technical, establish the blast radius, because it determines which half of this playbook applies. Ask the help desk, check the monitoring, or simply count the tickets: is this one user, some users, or everyone?


One user means the shared infrastructure is almost certainly fine — hundreds of other sessions are working through it right now. The problem lives at the edges. It could be their credentials (expired password, locked account, second-factor device out of battery). It could be their client (recent update, mangled configuration, half-installed patch). Or it could be their home internet, the most common answer of all. Verify quickly that the shared chain is clean — your session graph is normal, your probes are green. Then work the user's side with the standard remote-worker toolkit. Supporting the network you cannot see is a discipline of its own. It is covered in the home-office visibility guide.


Some users means hunt the pattern, because a partial outage always has one. Same ISP or region? Their provider is having a day, or a route between them and you soured. Same client version or operating system? An update broke something. Same authentication group? A policy or directory problem. All recent connectors while long-connected sessions purr along? Look hard at the address pool or the session ceiling — the resources consumed at connect time. The pattern is the diagnosis; gather five affected names and the shared attribute usually jumps out.


Everyone means the shared chain is broken somewhere, and the rest of this playbook is your path. Take a breath; you have a sequence for this.


If you monitor the VPN properly, question zero answers itself before the first ticket. The concurrent-session graph is the tell. A normal, gently wobbling line while complaints trickle in means scattered individual problems. A cliff — the count falling off a ledge at a specific minute — means a shared failure. The timestamp of the cliff edge is the single most valuable fact of the whole incident. It is the moment you will match against change logs, certificate lifetimes, and provider status. A graph that plateaus at a flat maximum instead of falling tells a third story: nothing broke, you filled up. Three shapes, three completely different mornings.


The shared chain, stage by stage


Every remote session, regardless of vendor, crosses the same six stages. Learn them once and every VPN incident becomes a search along a straight line instead of a fog.


  • 1. The user's internet. Their packets must leave home successfully. Broken here: individual users fail while your side shows nothing wrong at all.
  • 2. Your public address and circuit. The internet must route to your gateway's public address, over your circuit, which must be up and not drowning. Broken here: nobody reaches you, and usually other inbound things (mail, published services) suffer too.
  • 3. The VPN service itself. The gateway must be running and listening for connections. Broken here: the address answers pings but connection attempts hang or are refused.
  • 4. The authentication path. The gateway checks credentials against a backend — a RADIUS service, a directory, a second-factor provider, often a chain of all three. Broken here: connections reach the login step and die there, for everyone, with "invalid credentials" errors nobody believes. The care and monitoring of these backends is its own pillar: Core Network Services.
  • 5. The address pool. Each authenticated session must be issued an internal address from a finite pool. Broken here: authentication succeeds, then the connection fails at the last moment or connects broken — and early birds are fine while latecomers fail.
  • 6. Internal reachability. The session exists; now traffic must actually flow to the applications — routing, internal firewall rules, and the services themselves. Broken here: "connected but nothing works," the VPN cousin of the lying tunnel from the site-to-site guide.

Notice the chain is ordered from the user inward. Each stage's failure has a distinct signature at the connection step: cannot reach, reaches but no answer, answer but auth fails, auth passes but no address, address but no traffic. The signature users report — where the connection attempt dies — is itself a pointer to the broken stage, if the help desk asks for it.


A test for every stage


Each stage has a fast, decisive test, and the golden rule is to run them in order and believe the results. Stage one belongs to the user (can they load an ordinary website?). The rest belong to you, and the sequence looks like this from your chair:


1. From OUTSIDE your network (a probe at another site, a phone
   off Wi-Fi, a cloud vantage point):
     ping your gateway's public address        -> circuit reachable?
     attempt a connection to the VPN port      -> service listening?

2. From INSIDE:
     gateway console or SNMP: service running? sessions graph normal
     or cliff-edge? CPU sane?

3. Authentication:
     log in with a monitored TEST ACCOUNT      -> auth chain alive?
     check the auth backend's own health checks

4. Address pool:
     pool usage counter                        -> near 100%?

5. Through the VPN:
     from a connected session, reach one
     internal application                      -> end-to-end proof

Two of these deserve a word. The outside vantage point is the one junior admins most often lack. From inside the network you cannot see what the internet sees. And "it pings fine from my desk" proves nothing about stage two. A monitoring probe at a second site, or any external check service, gives you permanent outside eyes. Set this up on a calm day, not during the incident. The test account is a real, low-privilege account your monitoring logs in with on a schedule. That proves the whole authentication chain continuously. When it works but users fail, the problem is user-specific. When it fails too, you have a timestamp for exactly when the auth path broke.


Add one standing check that is really a pre-answered question: are we at the ceiling? A session graph pinned at a flat maximum, or a pool at full, is a capacity incident wearing an outage costume. That means a different fix and a different conversation. Ideally, it is caught weeks earlier by the headroom thresholds from Remote-Access VPN Capacity. Glance at it before deep diagnosis; it takes five seconds and regularly ends the investigation on the spot.


Tip: the best playbook is the one your monitoring already ran. Every test above can be a scheduled synthetic check — outside reachability, service port, test-account login, pool usage, through-VPN application probe. Then an outage opens not with a scramble but with a dashboard that shows which stage went red first, which is usually the whole diagnosis.


The classic culprits


Experienced admins keep a private list of "it's always one of these." Here is that list, with each culprit's tell. Check them in parallel with the staged tests — several take one glance each.


  • Expired certificate on the gateway. The service is up, the circuit is fine, and every client refuses to connect or throws trust warnings. That often starts at the exact minute the certificate lapsed, and typically first thing in the morning of the day it did. Preventable forever with expiry monitoring: see the Certificate & Expiration Monitoring pillar.
  • Auth backend down or unreachable. Everyone fails at the password step; the test account fails too; the gateway logs show timeouts talking to the backend. The VPN is innocent — it is the messenger.
  • Address pool empty. Auth succeeds, sessions fail at the finish line, early connectors are fine. The pool counter confesses instantly.
  • Circuit down or saturated. Outside probes fail or crawl; everything inbound suffers, not just VPN. If the circuit is merely full, connections limp and drop rather than fail cleanly.
  • License ceiling hit. A flat-topped session graph and refusals that started when the count reached a round number.
  • A firewall or policy change. The service is running but unreachable on its port from outside; the change log (or the colleague looking innocent) has the answer. "What changed last night?" remains the most cost-effective question in IT.
  • Public name pointing wrong. Clients connect by a DNS name; if that record changed or expired, users fail while your address tests pass. Test by name, not just by address.
  • Gateway restarted into a bad state. Up, listening, and subtly broken since the overnight power blip or update — uptime counters and logs date the moment.

Notice how many of these are detectable before they detonate: certificate expiry counts down in plain sight, pools and licenses fill visibly, circuits trend toward saturation. The culprits list doubles as a monitoring shopping list, which is the quiet moral of this entire pillar.


A worked example: quarter to nine on a Tuesday


Here is the playbook running at full speed, in a composite incident assembled from a hundred real ones. At 8:45 the help desk pings you: multiple users cannot connect. Question zero: the session graph shows a cliff at 8:31 — all users, shared chain, and you have your timestamp. Step one, ceiling: sessions fell, not plateaued; pool at 40%. Not capacity. Step two, outside vantage: the external probe history shows the gateway's public address answering pings all morning. But the VPN port check began failing at, of course, 8:31. Circuit up, service dark. Step three, from the console: the gateway is running, CPU bored, service listening on the inside. A service that listens inside but not outside smells like filtering. So you ask the most cost-effective question in IT — what changed? The firewall change log answers: a rule cleanup, applied at 8:31, that tidied away the rule admitting VPN traffic.


Rollback at 8:58. Sessions climb the cliff. Total time: thirteen minutes. The diagnosis was carried entirely by tests you could have run half-asleep. Those tests were a graph, a counter, an outside probe, a console check, a change log. No inspiration was required — which is precisely the point of a playbook. The 8:31 timestamp did the detective work; the sequence just made sure you looked at it in the right order. And note what closed the case. It was not a packet capture or vendor support. It was the boring, pre-built outside vantage point that could see what your desk could not.


Communicating during a remote-work outage


A VPN outage has a property no other outage has: everyone affected is somewhere else. They cannot see the ticket system if it sits behind the VPN. They may not see email if that requires the VPN too. What they will do, in the absence of information, is retry. That means hundreds of clients hammering the gateway in a loop. That has a way of turning a recovering service back into a broken one. Meanwhile, the help desk drowns in duplicates of a ticket you already have.


So the communication channel is an engineering decision. It must be chosen on a calm day: something reachable without the VPN. It could be a status page hosted outside your network, a text-message list, a phone-system greeting, or a chat platform that does not route through you. Anything works, as long as it is written down and users know it is the place to look. Then pre-write the first message, because during the incident you will not have spare eloquence. It needs only four parts. Say we know ("VPN connections are failing for all users"), and since when. Say what to do ("do not repeatedly retry; work offline where you can"), and when the next update comes. Post it early. A message saying "we know, next update in 30 minutes" converts hundreds of would-be tickets into silence. Keep the update promise even when the update is "still working on it."


One more calm-day task: make sure the incident responders themselves have a path in that does not depend on the broken thing. That could be an out-of-band management route, a console, or someone physically near the hardware. Discovering that your only way to fix the VPN is through the VPN is a rite of passage nobody needs.


Evidence for escalation


Some VPN incidents end with a fix; others end with an escalation. That could be to the ISP whose circuit is sick, the vendor whose gateway is misbehaving, or the management asking what happened. All three audiences run on evidence. And the moment to start collecting it is the moment the incident starts, because scrollback vanishes and memories improve with retelling.


The escalation kit starts with a timeline with timestamps (first failure, blast radius, each test and its result, every change made). It includes graphs covering the window — sessions, pool usage, gateway CPU, circuit utilization, and your outside-vantage probe history. That history is the piece an ISP cannot argue with. Include log excerpts from the gateway and auth backend for the same window. And include the change record — what was modified on your side recently, honestly included. A vendor's first questions are always what changed, when did it start, and what do the logs say. An ISP's first move is always "it looks fine from here." That dissolves against a probe graph showing exactly when their path stopped answering. Handing over the kit in the first message routinely saves a full day of back-and-forth. The same habit of keeping before-and-after proof is what The Exoneration File turns into a general practice.


The playbook on one page


StepQuestionFast testIf it fails
0One, some, or all?Ticket count, session graphOne: user's side. Some: find the pattern. All: continue.
1Ceiling?Sessions and pool vs maximumCapacity incident: free stale sessions, widen, communicate
2Reachable from outside?External ping and port check, by nameCircuit, DNS record, or upstream: escalate with probe history
3Service alive?Gateway console, service status, logsRestart or fail over per vendor guidance; check certificate
4Auth working?Test-account login; backend healthFix or escalate the backend; the VPN is the messenger
5Addresses left?Pool usage counterShorten holds, widen pool, split across pools
6Traffic flowing?Probe an internal app through a sessionRouting and internal firewall rules; check recent changes
—Everyone informed?Status message posted, out-of-bandPost it now; promise and keep an update cadence

Laminate mentally, or literally. The order is the value. Each step either finds the problem or certifies a stage as innocent. So by the time you reach the bottom you have either a root cause or a tight, evidence-backed escalation. Either one beats an hour of inspired guessing.


Where to go next


A playbook is what you reach for when prevention lost; the rest of this pillar is the prevention. Tunnel-truth checks, capacity headroom, per-tunnel graphs, and split-tunnel clarity make this playbook mostly unnecessary. Those topics live in the other guides at VPN & Remote-Access Monitoring. Nearly every step above gets faster when the tests already run on a schedule and the graphs already exist. Build those on a quiet afternoon. And the next "VPN is down" morning will be a short walk down a numbered list instead of a bad day.


Put this into practice with SysPeep Monitor

Most of this playbook can run before you even pick up the ticket. SysPeep Monitor schedules the synthetic checks — reachability, service ports, test logins, pool and session thresholds. It keeps the graphs that become your escalation evidence. Remote probes provide the outside vantage point. Alert rules with dependencies tell you which stage broke first instead of paging you eight times. The free trial installs in minutes.

Download Free TrialLearn About SysPeep