Fix it now
Event 41 from the OWA source records that the web application threw before it could render anything, so the browser gets a generic apology page. The event itself rarely names the fault. Start with the two failures Microsoft actually documents for this symptom: a security update installed without elevation, and an expired OAuth certificate.
Get-AuthConfig | Format-List CurrentCertificateThumbprint
Get-ExchangeCertificate | Format-List Thumbprint,Subject,NotAfter,Services
Get-WebAppPoolState MSExchangeOWAAppPool
Get-WebAppPoolState MSExchangeECPAppPool
Test-ServiceHealth
- Check free space on every Exchange volume first, including the one holding the logging tree. A full disk takes the web applications down before it takes anything else down.
- If the failure started after a security update, reinstall that .msp from an elevated Command Prompt by typing its full path, and restart the server. Microsoft documents installing it without elevation as the cause of exactly this symptom.
- If the admin centre still fails afterwards, run
.\UpdateCas.ps1and.\UpdateConfigFiles.ps1from the Exchange Bin folder in an elevated shell, theniisreset. - If the auth certificate returned by
Get-AuthConfighas expired, the documented resolution for the OWA and admin centre symptom is to install Exchange Server 2019 CU12 or Exchange Server 2016 CU23 or later. - Restart any application pool showing as Stopped with
Restart-WebAppPool MSExchangeOWAAppPool, then test a second mailbox on a different database.
Test one more mailbox before you touch the server. If only one mailbox fails and others on the same database are fine, stop investigating the server and look at that mailbox.
If OWA and the admin centre both render, you can stop. If not, the next section maps the generic page onto the four layers that produce it.
Why it happens
OWA and the Exchange admin centre are ASP.NET applications published on the Default Web Site, which proxies authenticated requests to the Exchange Back End site over HTTPS on port 444. When anything in that chain throws, the front end deliberately returns a generic error, because a mail server should not print exception detail to the internet. Event 41 is the record that something threw and nothing more.
Microsoft publishes no description for event 41 from the MSExchange OWA source, nor for event 140, nor for the ASP.NET event 1310 that appears alongside it. Treat the numbers as labels. What is useful is the text inside those events – the application path, the exception type, and often the component – and the timestamp, which lets you line them up with the request you made.
Two causes for this exact symptom are documented, and they are worth trying before anything speculative. The first is a security update installed on a server with User Account Control enabled but without elevated permissions: Microsoft’s article describes OWA returning “Something went wrong … HTTP Status code: 500” and the admin centre reporting that it could not load Microsoft.Exchange.Common. The repair is to reinstall the update elevated and, if the admin centre still fails, correct BinSearchFolders on the back end ECP virtual directory and run UpdateCas.ps1 and UpdateConfigFiles.ps1 followed by iisreset.
The second is the OAuth certificate. Where it is expired, missing or misconfigured, Microsoft documents the browser freezing or reporting that the redirect limit was reached, with Event ID 1003 from MSExchange Front End HTTPS Proxy naming a NullReferenceException in FbaModule.ParseCadataCookies. The documented resolution is not a manual certificate dance: it is to install Exchange Server 2019 CU12 or Exchange Server 2016 CU23 or later.
A security update went on without elevation
You have this one if OWA returns a generic error with HTTP 500, and the admin centre reports that it cannot load an Exchange assembly. It began after a manual .msp install.
- Open an elevated Command Prompt, type the full path to the .msp file, run it, and restart the server.
- If the admin centre still fails, open IIS Manager, go to Exchange Back End, the ECP virtual directory, Application Settings, and replace the environment-variable paths in BinSearchFolders with absolute ones.
- From an elevated Exchange Management Shell in the Exchange Bin folder, run
.\UpdateCas.ps1then.\UpdateConfigFiles.ps1. - Run
iisresetfrom an elevated Command Prompt, restart the server, and retest both applications.
This is the whole reason Microsoft’s cumulative update guidance insists on an elevated command prompt. It applies to security updates as well.
The OAuth certificate has expired
You have this one if The browser freezes or reports a redirect limit, and Event ID 1003 from MSExchange Front End HTTPS Proxy names a NullReferenceException in FbaModule.
- Read the thumbprint in use:
Get-AuthConfig | Format-List CurrentCertificateThumbprint, then check its NotAfter withGet-ExchangeCertificate. - Install Exchange Server 2019 CU12 or Exchange Server 2016 CU23 or later. That is Microsoft’s documented resolution for this symptom.
- Where you need to roll the certificate by hand, the documented sequence is
New-ExchangeCertificate, thenSet-AuthConfig -NewCertificateThumbprint <thumbprint> -NewCertificateEffectiveDate <date>, thenSet-AuthConfig -PublishCertificate, thenSet-AuthConfig -ClearPreviousCertificate. - Allow for Active Directory replication before testing on other Exchange servers.
This certificate is internal to the organisation. It is not the one browsers see, and replacing it changes nothing about what a client trusts.
A volume holding Exchange logs has filled
You have this one if Everyone loses OWA at roughly the same moment, often overnight, and a drive holding logging or transport data is at or near zero free.
- Check free space on every Exchange volume, not just the database drives.
- Clear old IIS logs under
C:\inetpub\logs\LogFilesand old Exchange logs under the Logging tree in the Exchange installation path, keeping whatever retention you are required to hold. - Restart the affected application pools, or run
iisreset /noforceif several are stuck. - Put scheduled housekeeping in place. Exchange logging grows without limit by default, so this recurs otherwise.
The virtual directory configuration is inconsistent after an update
You have this one if The ASP.NET entry names a configuration or module problem, and the failure began immediately after a cumulative update or a framework change.
- Confirm the installed .NET Framework matches the requirement for your build. Exchange Server SE requires .NET Framework 4.8 or 4.8.1.
- Record any customisations – logon format, URLs, authentication choices – before you touch anything.
- Recreate the affected directory with the matching Remove and New cmdlets, using
-Role ClientAccessor-Role Mailboxto target the front end or back end copy deliberately. - Run
iisreset /noforce, reapply your customisations and retest.
Each of these directories exists twice, once on the Default Web Site and once on Exchange Back End. Recreating the front end copy achieves nothing if the back end copy is the broken one.
One mailbox cannot be rendered
You have this one if Other mailboxes on the same database and the same server work normally.
- Test the mailbox in Outlook. If Outlook is fine and the browser is not, the fault is in mailbox content or settings rather than logon.
- Run
New-MailboxRepairRequestagainst that mailbox and let it complete. - If it still fails, move the mailbox to another database with
New-MoveRequest, which rebuilds its structures in the process. - Compare its regional and language settings against a working mailbox, since those are set on first logon and are per-mailbox.
Microsoft’s own warning about mailbox repair matters here: the mailbox being repaired is inaccessible while the request runs, and once started the request cannot be stopped without dismounting the database.
Full reference
Matching the pattern to the layer
| Who it affects | Where to look |
|---|---|
| Everyone, both OWA and the admin centre | Server-wide: disk space, application pools, the auth certificate, a badly installed update |
| The admin centre only | Its own virtual directory and application pool, and the OAuth dependency it leans on harder |
| One mailbox on one database | The mailbox, not the server |
| Started right after a cumulative or security update | Elevation, virtual directory configuration, or the framework version |
| Started right after a certificate change | The binding on the Exchange Back End site |
| External users only | The reverse proxy or the published URL |
Three event IDs with nothing published behind them
Event 41 from the MSExchange OWA source, event 140 from the same source and ASP.NET event 1310 all appear in these investigations, and Microsoft publishes a description for none of them. That is not a reason to ignore them: the entries carry the application path and the exception type, and reading three entries written in the same second usually resolves a vague page into one specific component. It is a reason not to look up the number and act on what a search result claims it means.
Event ID 41 is documented by Microsoft for a completely different source – the Windows Kernel-Power event recorded when a system reboots without shutting down cleanly. If a search leads you there, you are reading about a different event on a different log.
The auth certificate rollover, in order
| Step | Command | What it does |
|---|---|---|
| Create | New-ExchangeCertificate |
Produces the replacement. This role uses a self-signed certificate, not a public one |
| Stage | Set-AuthConfig -NewCertificateThumbprint <t> -NewCertificateEffectiveDate <date> |
Records the next certificate and when it should take over |
| Publish | Set-AuthConfig -PublishCertificate |
Rolls the staged certificate over immediately and deploys it |
| Clean up | Set-AuthConfig -ClearPreviousCertificate |
Clears the certificate saved as the previous one |
Do this only when you have a reason to. For the specific symptom of OWA and the admin centre becoming unreachable because the OAuth certificate has expired, Microsoft’s published resolution is the cumulative update, and applying it is both quicker and less likely to leave you with a half-rolled configuration.
Recreating a virtual directory discards every setting on it, including internal and external URLs, authentication choices and any branding. Export the current settings to a file first so you can put them back, and use the -Role parameter so you recreate the copy you actually meant to.
Log locations worth having open
- The Application log, filtered to the second the failure happened rather than to a source.
C:\inetpub\logs\LogFilesfor the IIS request itself, which tells you the status and substatus the front end returned.- The HttpProxy folders under the Exchange Logging tree, which show whether the request died in the front end or after it was proxied to the back end.
C:\ExchangeSetupLogs\ExchangeSetup.logif the failure began after an update.- The System log, for application pool crashes recorded by the Windows Process Activation Service.
When everything checks out and it still fails
- Confirm the Exchange Back End site still has a certificate bound to port 444. A certificate replacement that missed that binding produces exactly this.
- Confirm the application pool identities and .NET versions have not been altered from what Exchange configured.
- Check whether a hardening exercise removed an authentication method or added a header that the application does not expect.
- Test from the server console as well as from a workstation, so you can see whether a reverse proxy is contributing.
- Confirm the pools are still running an hour later, which distinguishes a worker process that starts and dies from one that never started.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
Event ID 41 |
Recorded by the OWA source when the application threw before rendering. Microsoft publishes no description; read the event text and the entries logged in the same second | not published by the vendor |
Event ID 140 |
A companion entry from the same source for the same failed request. No published description | not published by the vendor |
Event ID 1310 |
The ASP.NET health-monitoring entry that carries the application path and exception type. No published description | not published by the vendor |
Confirm the fix worked
- OWA and the Exchange admin centre both render for a test mailbox, from the server console and from a workstation.
Get-AuthConfigreturns a thumbprint whose certificate has a NotAfter date in the future.- The OWA and ECP application pools are still running an hour later.
- Reproduce the original action and confirm no new event 41 or 1310 entries appear.
- Free space on every Exchange volume is back above whatever threshold your monitoring uses.
Questions people ask about this
Does this need a licence to fix?
No. Every cause here is configuration, capacity or a mailbox problem on a server you already license. If the server turns out to be out of support that is worth addressing on its own merits, but it is not what this event is telling you.
Why does the admin centre fail while OWA still works?
They are separate applications with separate pools and separate virtual directories, and the admin centre leans harder on OAuth. An expired auth certificate takes it down while ordinary mailbox access carries on.
Is it safe to recreate the OWA virtual directory on a production server?
Yes, provided you record the existing settings first and do it in a change window. It is quick, but it drops every customisation and needs an IIS restart, which disconnects clients briefly. Use the -Role parameter so you recreate the front end or back end copy you intended.
Do I have to reboot the server?
Rarely. Restarting the affected application pool is usually enough and iisreset covers the rest. Reboot when a framework installation asks for one, or when you have reinstalled a security update, because Microsoft’s repair procedure includes a restart.
Can I trust what a search result says event 41 means?
Not without checking the source. Microsoft publishes no description for event 41 from the OWA source, and it does publish one for a completely unrelated Kernel-Power event with the same number. Read your own event’s text instead.
