<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Abolude Peter on Medium]]></title>
        <description><![CDATA[Stories by Abolude Peter on Medium]]></description>
        <link>https://medium.com/@aboludepeter?source=rss-32e4c85f7529------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*jWCh-W4y-_fNqL4LP8YcKQ.jpeg</url>
            <title>Stories by Abolude Peter on Medium</title>
            <link>https://medium.com/@aboludepeter?source=rss-32e4c85f7529------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 14:03:21 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@aboludepeter/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[When Azure Virtual Desktop Deployment Fails: Troubleshooting the Network Before Rebuilding…]]></title>
            <link>https://medium.com/@aboludepeter/when-azure-virtual-desktop-deployment-fails-troubleshooting-the-network-before-rebuilding-c372c518e757?source=rss-32e4c85f7529------2</link>
            <guid isPermaLink="false">https://medium.com/p/c372c518e757</guid>
            <category><![CDATA[troubleshooting]]></category>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[azure]]></category>
            <category><![CDATA[microsoft]]></category>
            <dc:creator><![CDATA[Abolude Peter]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 14:34:42 GMT</pubDate>
            <atom:updated>2026-10-27T14:34:42.010Z</atom:updated>
            <content:encoded><![CDATA[<h3>When Azure Virtual Desktop Deployment Fails: Troubleshooting the Network Before Rebuilding Everything</h3><p>There is a particular kind of frustration that comes with deploying cloud infrastructure. Everything looks right in the Azure portal, the resources appear to have been created, and then one small component fails and brings the entire deployment to a halt.</p><p>Recently, I encountered this while deploying Azure Virtual Desktop (AVD) in a new Azure subscription.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/763/0*aQeHu-DBRriH5I2Q.png" /></figure><p>The virtual machines were created. The host pool existed. However, the deployment failed during configuration, leaving me with two running VMs and an unsuccessful deployment.</p><p>What followed was a reminder of something I continue to learn as an engineer: <strong>a failed deployment does not always mean you need to start over. Sometimes, you just need to understand why it failed.</strong></p><h3>Before we begin: Microsoft documentation</h3><p>The following official Microsoft resources provide additional context for the services and troubleshooting techniques discussed in this article.</p><ul><li><a href="https://learn.microsoft.com/en-us/azure/virtual-desktop/required-fqdn-endpoint">Azure Virtual Desktop required URL list</a></li><li><a href="https://learn.microsoft.com/en-us/azure/nat-gateway/nat-overview">Azure NAT Gateway overview</a></li><li><a href="https://learn.microsoft.com/en-us/azure/virtual-machines/extensions/troubleshoot">Troubleshoot Azure VM extensions</a></li><li><a href="https://learn.microsoft.com/en-us/azure/virtual-desktop/manage-app-groups">Manage Azure Virtual Desktop application groups</a></li><li><a href="https://learn.microsoft.com/en-us/azure/virtual-desktop/workspace-management">Manage Azure Virtual Desktop workspaces</a></li></ul><h3>A quick glossary for beginners</h3><p>If some of these terms sound unfamiliar, here’s a simple explanation before we get into the troubleshooting.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fdocs.google.com%2Fspreadsheets%2Fd%2F1jDocZJOh9yLu9_3K8NUklvri3gNw6MVLZV99D8PJCeg%2Fpreview%3Fembedded%3Dtrue&amp;display_name=Google+Docs&amp;url=https%3A%2F%2Fdocs.google.com%2Fspreadsheets%2Fd%2F1jDocZJOh9yLu9_3K8NUklvri3gNw6MVLZV99D8PJCeg%2Fedit%3Fusp%3Dsharing&amp;image=https%3A%2F%2Flh7-us.googleusercontent.com%2Fdocs%2FAHkbwyLfQlCEd9K59dKQLGeOpMliIsrZY_rsyIeFu1E4I2hihMnfpfSQtoufNlZAIjye31IlESBW0NBVARBqlaMUHc_a5G7pFo768bNWpZe-0RZR4TZ_Grfp%3Dw1200-h630-p&amp;type=text%2Fhtml&amp;scroll=auto&amp;schema=google" width="700" height="530" frameborder="0" scrolling="no"><a href="https://medium.com/media/cd07eb85bb7458e81d66361ff6cff966/href">https://medium.com/media/cd07eb85bb7458e81d66361ff6cff966/href</a></iframe><h3>The problem: The VMs were running, but the deployment had failed</h3><p>During the AVD deployment, Azure created two session-host virtual machines.</p><p>Both appeared to be running, but the deployment reported a failure involving the Microsoft.PowerShell.DSC extension.</p><p>The important part of the error was:</p><blockquote><em>The DSC Extension failed to execute: Error downloading the AVD configuration package after 17 attempts. Unable to connect to the remote server.</em></blockquote><p>The configuration package was hosted at: <a href="https://wvdportalstorageblob.blob.core.windows.net/galleryartifacts/Configuration_1.0.03575.1577.zip">https://wvdportalstorageblob.blob.core.windows.net/galleryartifacts/Configuration_1.0.03575.1577.zip</a></p><p>This was the first important clue.</p><p>The VM was not necessarily failing because of its size, Windows image, or host-pool configuration. The extension was failing while trying to download a package required during setup.</p><p>Instead of deleting the resources and starting again, I decided to investigate the actual failure.</p><h3>Step 1: Check DNS resolution</h3><p>The first question was simple: could the VM resolve the hostname?</p><p>I used Azure Run Command to execute:</p><pre>Resolve-DnsName wvdportalstorageblob.blob.core.windows.net</pre><p>The hostname resolved successfully to an Azure Storage endpoint.</p><p>This established that DNS resolution was working.</p><p>However, resolving a hostname does not prove that the VM can establish a connection to the service. DNS tells the computer where to go, not whether it can get there.</p><p>That meant I needed to test the connection itself.</p><h3>Step 2: Test outbound connectivity</h3><p>Next, I tested TCP connectivity to port 443:</p><pre>Test-NetConnection wvdportalstorageblob.blob.core.windows.net -Port 443</pre><p>The initial test took an unusually long time to return through Azure Run Command. After adapting the test to provide clearer output, the result showed:</p><pre>TcpTestSucceeded : False</pre><p>Now I had a more specific finding.</p><p>DNS worked, but the VM could not establish a TCP connection to the storage endpoint on port 443.</p><p>This pointed toward an outbound connectivity problem.</p><p>Rather than guessing which network component was responsible, I moved to the Azure network configuration.</p><h3>Step 3: Inspect the subnet configuration</h3><p>I opened the virtual network and inspected the subnet used by the AVD session hosts.</p><p>The configuration revealed:</p><ul><li>Private subnet enabled.</li><li>Default outbound access disabled.</li><li>No Network Security Group associated with the subnet.</li><li>No NAT Gateway associated with the subnet.</li><li>No route table associated with the subnet.</li><li>No Azure Firewall deployed in the subscription.</li></ul><p>This was a new subscription without existing centralized network infrastructure.</p><p>The combination of disabled default outbound access and the absence of another configured outbound connectivity path explained why the VM could not reach the external storage endpoint.</p><blockquote>One important distinction: a subnet without an NSG is not automatically unable to access the internet. An NSG controls traffic through security rules. It does not, by itself, provide internet connectivity.</blockquote><blockquote>Likewise, a private subnet does not necessarily mean that every form of network communication is blocked.</blockquote><p>In this case, the critical issue was that default outbound access had been disabled and no alternative outbound path was configured.</p><p><strong>A VM can be running perfectly well in Azure while still being unable to reach the services it needs during provisioning.</strong></p><h3>Step 4: Configure outbound connectivity with a NAT Gateway</h3><p>Because this was a production environment, I wanted an explicit outbound connectivity solution rather than making an unplanned network change.</p><p>I created an Azure NAT Gateway in the same Azure region as the virtual network.</p><p>The configuration included:</p><ul><li>A Standard NAT Gateway.</li><li>A dedicated static public IP address.</li><li>An association with the virtual network containing the AVD subnet.</li><li>An association with the specific subnet used by the session hosts.</li></ul><p>I selected the existing AVD subnet rather than associating the gateway with unrelated subnets.</p><p>A NAT Gateway provides outbound source network address translation, allowing resources in the associated subnet to initiate outbound connections using its public IP address.</p><p>It does not allow unsolicited inbound connections to the VMs, and it is not a firewall. Production environments still need appropriate security controls and any required outbound restrictions.</p><p>After the NAT Gateway deployment succeeded, I did not immediately assume the problem was fixed. I went back to the VM and tested connectivity again.</p><h3>Step 5: Verify the fix with a single diagnostic script</h3><p>I wanted to verify three things:</p><ol><li>DNS resolution.</li><li>TCP connectivity to port 443.</li><li>An actual HTTPS request to the AVD configuration package.</li></ol><pre>$hostName = &#39;wvdportalstorageblob.blob.core.windows.net&#39;<br>$url = &quot;https://$hostName/galleryartifacts/Configuration_1.0.03575.1577.zip&quot;<br><br>Write-Output &quot;=== 1. DNS Resolution ===&quot;<br><br>try {<br>    Resolve-DnsName $hostName -ErrorAction Stop |<br>        Where-Object Type -eq &#39;A&#39; |<br>        Select-Object -First 1 Name, IPAddress |<br>        Format-Table -AutoSize<br>} catch {<br>    Write-Output &quot;DNS FAILED: $($_.Exception.Message)&quot;<br>}<br><br>Write-Output &quot;`n=== 2. TCP 443 Connectivity ===&quot;<br><br>$tcp = [System.Net.Sockets.TcpClient]::new()<br><br>try {<br>    $connect = $tcp.ConnectAsync($hostName, 443)<br>    if ($connect.Wait(5000) -and $tcp.Connected) {<br>        Write-Output &quot;TCP 443: SUCCESS&quot;<br>    } else {<br>        Write-Output &quot;TCP 443: FAILED or TIMED OUT&quot;<br>    }<br>} catch {<br>    Write-Output &quot;TCP 443: FAILED - $($_.Exception.Message)&quot;<br>} finally {<br>    $tcp.Dispose()<br>}<br><br>Write-Output &quot;`n=== 3. HTTPS Request ===&quot;<br><br>try {<br>    $request = [System.Net.HttpWebRequest]::Create($url)<br>    $request.Method = &#39;GET&#39;<br>    $request.Timeout = 15000<br>    $request.ReadWriteTimeout = 15000<br>    $request.AllowAutoRedirect = $true<br>    $response = $request.GetResponse()<br>    Write-Output &quot;HTTPS: SUCCESS (HTTP $([int]$response.StatusCode))&quot;<br>    $response.Close()<br>} catch {<br>    $ex = $_.Exception<br>    if ($ex.Response) {<br>        Write-Output &quot;HTTPS: HTTP $([int]$ex.Response.StatusCode) $($ex.Response.StatusDescription)&quot;<br>        $ex.Response.Close()<br>    } else {<br>        Write-Output &quot;HTTPS FAILED: $($ex.Message)&quot;<br>    }<br>}<br><br>Write-Output &quot;`n=== Diagnostics Complete ===&quot;</pre><p>The results were:</p><pre>=== 1. DNS Resolution ===<br>DNS resolution succeeded.<br><br>=== 2. TCP 443 Connectivity ===<br>TCP 443: SUCCESS<br><br>=== 3. HTTPS Request ===<br>HTTPS: SUCCESS (HTTP 200)<br><br>=== Diagnostics Complete ===</pre><p>The endpoint that had previously been unreachable was now accessible.</p><p>This was a useful turning point because I had verified the result from the VM itself rather than relying solely on the NAT Gateway’s successful deployment status.</p><h3>Step 6: Recover the failed DSC extension</h3><p>Restoring connectivity did not automatically rerun the extension that had already failed.</p><p>I checked the extension status of both VMs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/588/1*ORH-Z06s5u7XkmwVAU_vIw.png" /><figcaption>Extension status on both VMs</figcaption></figure><p>One session host had completed provisioning successfully, while the other still reported the original error.</p><p>I inspected the failed extension using Azure PowerShell:</p><pre>Get-AzVMExtension `<br>  -ResourceGroupName &quot;&lt;RESOURCE_GROUP&gt;&quot; `<br>  -VMName &quot;&lt;FAILED_SESSION_HOST&gt;&quot; `<br>  -Name &quot;Microsoft.PowerShell.DSC&quot; |<br>  Format-List *</pre><p>The output showed that the extension was still in a Failed provisioning state. Its configuration referenced the AVD configuration package and the relevant host pool.</p><p>I then reapplied the existing extension configuration using Set-AzVMExtension and forced the extension handler to run again.</p><blockquote>The first attempt encountered a PowerShell parameter-type error because the extension’s public settings were returned as a JSON string, while the command expected a hashtable.</blockquote><p>I corrected the settings by converting the JSON string:</p><pre>$settings = $ext.PublicSettings | ConvertFrom-Json -AsHashtable</pre><p>I then reapplied the extension using the existing settings:</p><pre>Set-AzVMExtension `<br>  -ResourceGroupName &quot;&lt;RESOURCE_GROUP&gt;&quot; `<br>  -VMName &quot;&lt;FAILED_SESSION_HOST&gt;&quot; `<br>  -Name &quot;Microsoft.PowerShell.DSC&quot; `<br>  -Location $ext.Location `<br>  -Publisher $ext.Publisher `<br>  -ExtensionType $ext.ExtensionType `<br>  -TypeHandlerVersion $ext.TypeHandlerVersion `<br>  -Settings $settings `<br>  -ForceRerun ([guid]::NewGuid().ToString())</pre><p>The extension provisioning subsequently succeeded.</p><p>This step required care because the DSC extension participates in AVD session-host configuration and registration. Existing protected settings and registration credentials should not be overwritten casually. The command above illustrates the corrected public-settings approach used in this investigation, not a universal replacement for every AVD extension configuration.</p><p>The important point was that I could recover the failed session host without deleting the VM or recreating the host pool.</p><h3>Step 7: Investigate why the desktop was missing from Windows App</h3><p>With both VMs running and both DSC extensions successfully provisioned, I moved to the user experience.</p><p>I expected users in the assigned Entra ID group to see the desktop in Windows App, but it was not appearing.</p><p>I checked the desktop application group’s assignments. The intended user group was already assigned.</p><p>I also noticed that the Assigned VM tab showed zero. This was not, by itself, evidence of a problem. In a pooled host pool, users generally do not need to be permanently assigned to individual session-host VMs.</p><p>The next check was the application’s workspace association.</p><p>The desktop application group was not associated with a workspace.</p><p>I associated the application group with the appropriate AVD workspace and checked the result again.</p><p>The desktop appeared in Windows App.</p><p>This was a separate configuration issue from the original networking failure, but it was essential to completing the deployment.</p><p>A successfully provisioned VM does not automatically mean users can discover and access its published desktop. The host pool, application group, user assignments, workspace association, and client feed all play their respective roles.</p><h3>What I learned from this incident</h3><h3>1. Read the actual error message</h3><p>The deployment error explicitly identified a failed download from an Azure Storage endpoint.</p><p>That gave me a starting point and helped avoid wasting time investigating unrelated VM settings.</p><p>When an extension fails, I want to understand what it was doing when it failed, not just which resource has a Failed status.</p><h3>2. Separate DNS resolution from network connectivity</h3><p>DNS resolution succeeded while TCP connectivity failed.</p><p>If I had stopped after the successful DNS lookup, I might have incorrectly concluded that the network was working.</p><p>Testing progressively helped narrow down the issue.</p><h3>3. Understand the subnet’s outbound connectivity</h3><p>Deploying a VM in Azure does not guarantee that it has internet access.</p><p>In this incident, default outbound access was disabled, and no other outbound connectivity path had been configured.</p><p>That combination was central to the failure.</p><h3>4. Fix the cause before retrying the operation</h3><p>Repeatedly retrying the DSC extension before correcting the network configuration would not have solved the underlying issue.</p><p>The better sequence was to restore connectivity, verify it, and then rerun the failed extension.</p><h3>5. Avoid rebuilding resources unnecessarily</h3><p>Both VMs had been created. One extension had already succeeded, and the host pool did not need to be recreated.</p><p>Rebuilding everything would have added more work without addressing the original network configuration.</p><h3>6. Treat deployment and user access as separate milestones</h3><p>Even after both VMs and their DSC extensions succeeded, the desktop was not visible to users because the application group lacked a workspace association.</p><p>Infrastructure provisioning and end-user accessibility need to be verified separately.</p><h3>7. Make troubleshooting manageable</h3><p>I also found that investigating one thing at a time made it easier to follow the evidence.</p><p>Check DNS. Then check connectivity. Inspect the subnet. Fix the network. Verify the connection. Recover the failed extension. Finally, check the user-facing configuration.</p><p>Each result informed the next step.</p><p>This is more effective than making several changes at once and then trying to work out which one solved the problem.</p><h3>Final thoughts</h3><p>This incident reinforced an important principle for me: troubleshooting should be evidence-driven.</p><p>A failed deployment is a symptom, not a complete diagnosis.</p><p>By examining the extension error, checking DNS, testing outbound connectivity, inspecting the subnet, configuring a NAT Gateway, recovering the failed extension, and verifying the workspace association, I was able to get the environment working without rebuilding the VMs or host pool.</p><p>The biggest takeaway was not simply learning how to configure a NAT Gateway. It was learning to slow down, follow the evidence, and make one change at a time.</p><p><strong>Understand the failure before rebuilding the solution.</strong></p><p>And, as always, document what you learn. The next time a similar problem appears, the investigation will start from a much better place.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c372c518e757" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[NEVER SKIP THE DOCUMENTATION.]]></title>
            <link>https://medium.com/@aboludepeter/never-skip-the-documentation-1d1725642042?source=rss-32e4c85f7529------2</link>
            <guid isPermaLink="false">https://medium.com/p/1d1725642042</guid>
            <category><![CDATA[azure-functions]]></category>
            <category><![CDATA[azure-devops]]></category>
            <category><![CDATA[troubleshooting]]></category>
            <category><![CDATA[microsoft-azure]]></category>
            <dc:creator><![CDATA[Abolude Peter]]></dc:creator>
            <pubDate>Mon, 23 Mar 2026 17:44:07 GMT</pubDate>
            <atom:updated>2026-03-24T01:16:35.780Z</atom:updated>
            <content:encoded><![CDATA[<p>This was one of those rabbit holes; the kind where you check the time and it’s 3 a.m. for the second night straight, and the trigger still hasn’t fired.</p><p>Debugging a non-triggering Blob Storage Function (Python, EP1 Linux, Azure DevOps).</p><p>I got pulled into this one halfway through. A friend had a Function App that <em>“deploys fine but doesn’t do anything”</em>. Classic.</p><p>The requirement was straightforward:</p><ul><li>Blob lands in storage</li><li>Python Azure Function triggers</li><li>Data gets written to SQL Server</li></ul><figure><img alt="An illustration of the architecture" src="https://cdn-images-1.medium.com/max/1024/1*1TX0cgHVRWUvDQzNcSQfUg.png" /><figcaption>An Illustration of the Architecture.</figcaption></figure><p>In reality:</p><ul><li>Function deployed ✅</li><li>App running ✅</li><li>Blob uploaded ✅</li><li>Trigger fired ❌</li></ul><p>No logs, no execution, nothing.</p><p>So, this turned into a full investigation. I’ll walk through what I found, what I changed, and more importantly, <strong>how each issue revealed itself</strong>.</p><h3>1. First assumption: trigger misconfiguration</h3><p>When a Blob trigger doesn’t fire, I don’t look at networking first. I check whether the runtime even registered the trigger.</p><p>What I noticed:</p><ul><li>No “Listening on blob container…” logs</li><li>No indexing errors shown clearly</li><li>Function app was “healthy”</li></ul><p>That usually means the function <strong>wasn’t indexed correctly</strong>.</p><h4>Fix: simplify and correct trigger</h4><p>I reverted everything to a <strong>pure Blob trigger</strong> (Python v1 model).</p><p>function.json</p><pre>{<br>  &quot;scriptFile&quot;: &quot;__init__.py&quot;,<br>  &quot;bindings&quot;: [<br>    {<br>      &quot;name&quot;: &quot;inputBlob&quot;,<br>      &quot;type&quot;: &quot;blobTrigger&quot;,<br>      &quot;direction&quot;: &quot;in&quot;,<br>      &quot;path&quot;: &quot;blob-storage/folder/{name}&quot;,<br>      &quot;connection&quot;: &quot;AzureWebJobsStorage&quot;<br>    }<br>  ]<br>}</pre><p><strong>Thought process:</strong></p><p>Remove all ambiguity. No Event Grid, no hybrid logic. If this doesn’t trigger, the problem is not configuration complexity, it’s runtime.</p><h3>2. Extension bundle mismatch.</h3><p>Next thing I checked was the runtime compatibility.</p><p>The app was on Functions v4, but extension configuration wasn’t aligned.</p><p>If the extension bundle is wrong, Blob triggers don’t register, and you don’t always get explicit errors.</p><p>host.json</p><pre>{<br>  &quot;version&quot;: &quot;2.0&quot;,<br>  &quot;extensionBundle&quot;: {<br>    &quot;id&quot;: &quot;Microsoft.Azure.Functions.ExtensionBundle&quot;,<br>    &quot;version&quot;: &quot;[4.0.0, 5.0.0)&quot;<br>  }<br>}</pre><p><strong>Thought process:</strong></p><p>This is one of those “if you skip docs, nothing works” cases. Blob trigger support depends on the correct bundle version.</p><h3>3. Folder structure (this matters more than people think)</h3><p>The project structure was slightly off. Azure Functions is strict here.</p><h4>Final structure:</h4><pre>APP_FOLDER_NAME/<br>├── function_name/ # All files related to making the function run.<br>│   ├── __init__.py<br>│   ├── function.json<br>│   └── function_script.py<br>├── host.json<br>├── requirements.txt<br>├── azure-pipelines.yml</pre><p>__init__.py</p><pre>import logging<br>from .function_script import main_logic<br><br>def main(inputBlob):<br>    logging.info(f&quot;Processing blob: {inputBlob.name}&quot;)<br>    main_logic(inputBlob)</pre><p><strong>Thought process:</strong></p><p>If indexing fails, sometimes it’s not obvious. Structure must match exactly what the runtime expects.</p><h3>4. App settings cleanup</h3><p>At this point, I checked App Settings.</p><h4>Found issues:</h4><ul><li>Missing AzureWebJobsStorage</li><li>Inconsistent storage references across code and config</li></ul><h4>Fix:</h4><ul><li>Added AzureWebJobsStorage</li><li>Removed conflicting storage references in code and bindings</li><li>Updated DB-related settings to correct PROD values</li></ul><p><strong>Thought process:</strong></p><p>Blob trigger relies on AzureWebJobsStorage for leases/checkpoints. Missing it = trigger won’t function properly, even if everything else is correct.</p><h3>5. Deployment was the real problem</h3><p>This was the biggest issue.</p><p>Pipeline was deploying <strong>source files only</strong>, not dependencies.</p><p>So:</p><ul><li>Function deployed</li><li>Runtime tried to resolve dependencies</li><li>Failed silently</li><li>Trigger never registered</li></ul><h4>Fix: Package dependencies explicitly</h4><p>Install dependencies into: .python_packages/lib/site-packages</p><h4>Pipeline Fix:</h4><pre>- script: |<br>    python -m pip install --upgrade pip<br>    pip install -r requirements.txt \<br>      --target=&quot;.python_packages/lib/site-packages&quot;<br>  displayName: Install dependencies</pre><p>Then package and deploy that artifact:</p><pre>- task: AzureFunctionApp@1<br>  inputs:<br>    azureSubscription: &#39;&lt;service-connection&gt;&#39;<br>    appType: functionAppLinux<br>    appName: &#39;&lt;function-app-name&gt;&#39;<br>    package: $(Build.ArtifactStagingDirectory)/functionapp.zip</pre><p><strong>Thought process:</strong></p><p>I initially assumed runtime build would handle this. It didn’t. Once I controlled the packaging, the behavior became predictable.</p><h3>6. Service connection mismatch</h3><p>Another subtle but important issue.</p><ul><li>New Function App was created</li><li>Pipeline still used an old service connection</li></ul><h4>Result:</h4><ul><li>Deployment “succeeded”</li><li>But wasn’t consistently targeting the right resource</li></ul><h4>Fix:</h4><ul><li>Created new Azure Resource Manager service connection</li><li>Scoped to correct subscription/resource group</li><li>Updated pipeline to use it</li></ul><p><strong>Thought process:</strong></p><p>This explained why some fixes didn’t seem to apply. Deployment wasn’t reliably hitting the intended app.</p><h3>7. Dependency issue: cryptography + Python 3.10</h3><p>After fixing deployment, I started seeing import errors.</p><h4>Root cause:</h4><ul><li>azure-identity depends on cryptography</li><li>Python 3.10 + Linux EP1 has a <strong>GLIBC compatibility issue</strong></li></ul><h4>Fix:</h4><p>requirements.txt</p><pre>azure-functions<br>azure-storage-blob<br>azure-identity<br>pyodbc<br>cryptography==43.0.3</pre><h4>Reference:</h4><p><a href="https://github.com/pyca/cryptography/issues/12064#issuecomment-2506717233">Cryptography 44.0.0 error with GlibC 2.33 on Azure function · Issue #12064 · pyca/cryptography</a></p><p><strong>Thought process:</strong></p><p>This wasn’t obvious from Azure docs. Had to trace the import error and verify against known issues.</p><h3>8. EP1 Linux limitation (startup.sh)</h3><p>There was an attempt to use startup.sh to install system packages.</p><p>That does not work on EP1 Linux.</p><ul><li>No root access</li><li>No apt-get</li><li>Startup scripts ignored for this purpose</li></ul><p><strong>Thought process:</strong></p><p>Important to accept platform constraints early. Otherwise you waste time trying to force something unsupported.</p><h3>9. Finally: trigger works, next failure appears</h3><p>Once everything above was fixed:</p><ul><li>Blob trigger fired ✅</li><li>Function executed ✅</li></ul><p>Then I hit:</p><pre>SQLConnect login timeout</pre><p>At this point, the system was behaving correctly.</p><p><strong>Thought process:</strong></p><p>This is where debugging becomes normal again.</p><p>The problem is no longer infrastructure; it’s application logic or connectivity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SEj_0H0YB6NvYwBk" /><figcaption>Photo by <a href="https://unsplash.com/@william_rudolph?utm_source=medium&amp;utm_medium=referral">William Rudolph</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>What actually fixed the system</h3><p>Not one thing. A sequence:</p><ol><li>Correct trigger configuration</li><li>Correct extension bundle</li><li>Proper folder structure</li><li>Valid app settings (AzureWebJobsStorage)</li><li>Correct dependency packaging</li><li>Correct service connection</li><li>Dependency compatibility fix (cryptography)</li></ol><h3>Final takeaway</h3><p>Nothing here was “advanced”.</p><p>Every issue came from:</p><ul><li>Skipping documentation</li><li>Assuming defaults would work</li><li>Not validating deployment artifacts</li></ul><blockquote><em>Azure Functions don’t fail loudly when misconfigured. They just don’t run.</em></blockquote><p>That’s what makes this tricky.</p><blockquote><strong>“If your Blob trigger isn’t firing, don’t start with networking. Start with whether the function was actually built and indexed correctly.”</strong></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1d1725642042" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>