vCenter no healthy upstream is the whole error page: a white screen, three words, and no hint about what broke. I have hit vCenter no healthy upstream after a planned reboot, after a storage alarm nobody read, and once after a certificate quietly expired on a Sunday. The good news is that the message itself is boring. It only means that the reverse proxy inside the vCenter Server Appliance has no running backend service to send your browser to. The real work is finding which service is down and why. Below is the order I check things in, with the commands I actually run, and every fix is backed by a Broadcom knowledge base article so you are not guessing on a production appliance.
What vCenter No Healthy Upstream Really Means
In vCenter 7.x and 8.x the appliance fronts every web request with the Envoy proxy. When the vSphere Client, the Lookup Service, or the vpxd service behind that proxy is not running, Envoy answers with HTTP 503 and the plain text no healthy upstream. Broadcom says this in its own words in several articles: the error is generic and shows up for multiple reasons. That is why the first rule with vCenter no healthy upstream is to stop treating the browser message as the problem. A no healthy upstream vSphere Client page is a symptom, and the cause is almost always one of five things: a stopped service, a full partition, a starved host, an expired certificate, or a broken configuration file.
One more thing I learned the hard way: restarting services without finding the cause rarely works. If a partition is full or the STS certificate expired, the services will stop again, sometimes before you have even finished logging in.
Step 1: Find the Stopped Service Behind vCenter No Healthy Upstream
Open an SSH session to the appliance as root. If you land in the appliance shell, type shell to get a Bash prompt. Then list every service and its state:
service-control --status --allYou get three groups: Running, Stopped and sometimes StartPending. In most vCenter no healthy upstream cases I see vmware-vpxd, vmware-vapi-endpoint or vmware-stsd sitting in the Stopped list. Note the names, because they point at the cause. For a closer look at one service, Broadcom documents vmon-cli, which is the service monitor that starts and health-checks everything on the appliance:
vmon-cli -l
vmon-cli -s vpxdThe second command prints the start type, the run state and a HealthState field. If the service is STOPPED or keeps flapping, the reason is in the vmon log, and the component logs give the details:
tail -n 200 /var/log/vmware/vmon/vmon.log
tail -n 200 /var/log/vmware/vpxd/vpxd.log
tail -n 100 /var/log/vmware/envoy/envoy-access.logCode language: JavaScript (javascript)
The illustration above shows the pattern I see most often: the browser page, a stopped vpxd, and a log partition at 100 percent. Keep that df output in mind, because it is the next check.
Step 2: Check Disk Space and Host Memory
Before touching certificates or config files, run df -h. A full /storage/log, /storage/db or /storage/core partition stops services from initializing, and the UI reports exactly the same vCenter no healthy upstream page.
df -h
du -sh /storage/log/vmware/* | sort -h | tail -n 15Broadcom has separate articles for each partition. On vCenter 8.0 Update 3, /storage/log can fill up because the file /storage/log/vmware/vmware-updatemgr/updatemgr-vmon.log.stderr grows without limit. The documented fix is to back up the file if you need it, truncate it, and then restart the Update Manager service. I take a snapshot first, every time:
> /storage/log/vmware/vmware-updatemgr/updatemgr-vmon.log.stderr
service-control --restart vmware-updatemgrCode language: JavaScript (javascript)If that restart fails, check whether vmware-stsd is down, because the article lists sts as a dependency. You can confirm with service-control --list-dependencies vmware-updatemgr, start it with service-control --start vmware-stsd, then retry the restart. When /storage/db is the full one, the cause is database bloat, and Broadcom’s fix is to extend the disk if needed and then run vacuum full and vacuum analyze in the VCDB database with all other services stopped. That one is slower and riskier, so follow the Broadcom article on the /storage/db partition step by step rather than improvising.
If the appliance itself boots into emergency mode and the VM console shows file system check failures, look at the datastore. Broadcom’s article on datastore space constraints describes a vCenter no healthy upstream case where the appliance sits on a full datastore and drops into emergency mode, and the resolution is to free or extend the datastore, then run fsck from the GRUB menu before powering the appliance back on.
Memory contention on the ESXi host
This one is easy to miss. Broadcom documents a vCenter no healthy upstream case where the ESXi host that runs vCenter has run out of memory. The symptoms are distinctive: a service restart always stalls at vmware-sps, started services fail to write logs, and every command in the shell is painfully slow while disk and log partitions look fine. The fix is not inside the appliance at all. Free host memory by powering off or migrating other virtual machines, make sure the vCenter VM still matches the hardware requirements for its deployment size, and then restart the appliance or its services.
Step 3: Check Whether the STS Certificate Expired
If vCenter no healthy upstream appears right after a reboot and vpxd.log contains Failed to read X509 cert, suspect the Security Token Service signing certificate. Broadcom describes exactly this in its article on the “Signing certificate is not valid” or “No healthy upstream” error. An expired STS certificate keeps vmware-stsd from starting, and the login page can show No Healthy Upstream, Signing certificate is not valid or Cannot connect to vCenter Single Sign-On server depending on how far the browser gets.
The STS certificate typically expires two years after creation, and the normal certificate expiry alarm does not cover it. If the vSphere Client still loads, check it under Administration, Certificates, Certificate Management, where the STS signing certificate card shows a valid-until date. If the client is already down, Broadcom now points to the vCert tool instead of the older checksts.py script, which has been deprecated. Install vCert on the appliance as described in its article, use menu 2 option 8 to view the STS signing certificates, and menu 3 option 8 to replace them. Take a snapshot without memory first, and in Enhanced Linked Mode take powered-off snapshots of every vCenter in the SSO domain before you change anything.

Step 4: Fix a vpxd That Dumps Core on Start
Sometimes service-control --start --all runs, every other service comes up, only vpxd refuses, and vCenter no healthy upstream persists. In vmon.log you see Service is dumping core, and if you run /sbin/vpxd by hand from the shell it aborts with a file format exception such as Duplicate child: useSSL. Broadcom traces this to a duplicate entry in /etc/vmware-vpx/vpxd.cfg, usually created when SSL for NFC data traffic was enabled or disabled through the UI.
grep -n "useSSL\|maxConcurrentSyncItems" /etc/vmware-vpx/vpxd.cfgCode language: JavaScript (javascript)If the same element appears twice inside the same block, back up the file, remove the duplicate line in vi, and start the services again. The vpxd.cfg article covers vCenter 7.0.x and 8.0.x and insists on a backup or snapshot before the edit, which I would repeat here. A related case in vCenter 8.0.x involves a damaged server.xml for the STS Tomcat instance; the server.xml article has you copy the file from a working vCenter on the same build, so keep that in mind if the catalina log complains about loading its configuration.
Step 5: Restart the Services and Verify the Fix
Only once the cause is handled do I restart everything. The pattern Broadcom documents for a full restart is a stop followed by a start:
service-control --stop --all && service-control --start --all
service-control --status --all
vmon-cli -s vpxd
When the Stopped list is empty and vpxd reports RunState STARTED and HealthState HEALTHY, the vSphere Client loads again and the vCenter no healthy upstream page is gone. I also reload the browser page with a hard refresh, because a cached 503 can make you think the fix did not work. Then I check df -h once more a few minutes later to confirm that whatever filled the partition is not filling it again.
Three vCenter No Healthy Upstream Variants That Look the Same
- Only domain logins fail: on vSphere 8.0 with Entra ID as the identity provider, Broadcom describes a token refresh race condition that returns no healthy upstream for domain accounts while the appliance is otherwise healthy. The known issue is resolved in vCenter 8.0 Update 3g, and the workaround until you upgrade is the stop and start of all services shown above.
- Only the Updates tab of a cluster shows it: that is a Lifecycle Manager backend problem, and Broadcom’s article on the Updates tab reports that restarting all services fixed it.
- Only remote VPN users see it: if LAN users log in fine, check the client side. Broadcom’s VPN article describes split-tunnel clients whose home subnet overlaps the vCenter subnet, with 503 UT and UH flags in
envoy-access.log. The fixes are a lower interface metric on the VPN adapter, a non-overlapping home subnet, or a 1350-byte MTU on the VPN adapter.
How I Stop vCenter No Healthy Upstream From Coming Back
- Put an alarm or a monthly check on
/storage/log,/storage/dband/storage/core. Most of my vCenter no healthy upstream after reboot cases were partitions that had been full for days. - Record the STS signing certificate expiry date in your change calendar. vSphere 7.0 Update 1 and later warn 90 days ahead, but only if someone reads the notification before vCenter no healthy upstream turns up on a Monday morning.
- Keep the vCenter VM on a host with memory headroom, and keep it in a DRS cluster if you can. Broadcom notes the memory contention issue is not observed when DRS is enabled.
- Stay current on patches. The Entra ID login variant is fixed in 8.0 Update 3g, and the updatemgr log growth is specific to 8.0 Update 3.
- Snapshot before every fix. Every Broadcom article above starts with that sentence for a reason.
Read more: ESXi Host Not Responding in vCenter: 7 Proven Steps to Fix It
Read more: Greyed Out “Migrate” Button Issue in VMware vCenter and Its Solution
Read more: Reset a Lost ESXi Root Password
FAQ
What does “no healthy upstream” mean in vCenter?
It is an HTTP 503 from the Envoy proxy inside the vCenter Server Appliance. A backend service such as vpxd, the STS or the vAPI endpoint is not running, so the proxy has nowhere to send the vSphere Client request.
How do I find which service is stopped behind vCenter no healthy upstream?
SSH to the appliance, type shell, then run service-control –status –all. Use vmon-cli -s followed by the service name for the health state, and read /var/log/vmware/vmon/vmon.log for the reason it stopped.
Can a full disk cause vCenter no healthy upstream?
Yes. A full /storage/log, /storage/db or /storage/core partition prevents services from starting. Run df -h first, free or extend the partition using the matching Broadcom article, then restart the services.
How do I check whether the STS certificate expired?
In the vSphere Client go to Administration, Certificates, Certificate Management and read the STS signing certificate card. If the client is down, install the vCert tool on the appliance and use its View Certificate Info menu.
Is it safe to run service-control –stop –all and –start –all?
Yes, it is the documented way to restart all services, but it causes a short outage of the vSphere Client and it does not fix a full disk or an expired certificate. Fix the cause first, then restart.
References
Broadcom KB: vCenter Server UI displays “no healthy upstream” due to vpxd service failing to start
Broadcom KB: “Signing certificate is not valid” or “No healthy upstream” error in vCenter Server Appliance
Broadcom KB: Checking the STS certificate expiration and replacing expired STS certificates
Broadcom KB: vCenter UI shows error “No healthy upstream”, due to memory contention
Broadcom KB: updatemgr-vmon.log.stderr is very large causing vCenter services to not start
Broadcom KB: No healthy upstream due to /storage/db partition full
Broadcom KB: “No Healthy Upstream” error due to datastore space constraints
Broadcom KB: ERROR: no healthy upstream while login to vCenter using domain accounts
Broadcom KB: vCenter UI returns “no healthy upstream” for remote VPN users over conflicting subnets
Broadcom KB: “No Healthy Upstream” seen under Updates tab in vCenter for cluster
Broadcom KB: Checking a Service Health State in vCenter using vmon-cli




