Fix it now
0x80D02003 is DO_E_JOB_NOT_FOUND: the job was not found when one was expected. 0x80D0000A has no published meaning at all. Neither is a peer caching error, so start by proving whether peering is involved rather than assuming it.
Get-DODownloadMode
Get-DeliveryOptimizationStatus
Set-DODownloadMode -DownloadMode 0
Delete-DeliveryOptimizationCache -Force
Restart-Service DoSvc
- Retry the download from Windows Update. If it now completes, peering was involved and the rest of this article is about deciding how much of it you want back.
- If it still fails with peering off, the job itself is not being created or is being lost. Stop wuauserv and bits, rename
C:\Windows\SoftwareDistribution, restart the services and rescan so a fresh job is created. - To keep peering on the local network only, set
Set-DODownloadMode -DownloadMode 1and allow TCP 7680 inbound on the clients. - For most organisations Microsoft recommends Group mode:
Set-DODownloadMode -DownloadMode 2, which also needs UDP 3544 for Teredo.
Do not use download mode 100. Bypass is deprecated starting in Windows 11 and Microsoft says configuring it can cause content to fail to download.
If downloads complete and the mode is the one you meant to leave in place, you are done. The next section covers what these codes do and do not say.
Why it happens
A Delivery Optimization download is a job: the client registers it, the service tracks it, and content arrives against it from HTTP sources, a Microsoft Connected Cache server, or peers. 0x80D02003 is DO_E_JOB_NOT_FOUND, whose published wording is that the job was not found when one was expected to be present. That is a lost or never-created job, not a peer serving something the client could not use.
This is worth being blunt about, because this pair of codes is widely described as peer caching failures and that description sends people to open firewall ports for a fault that has nothing to do with peers. 0x80D0000A is not in Microsoft’s published Delivery Optimization return value list at all, and no Microsoft page carries a meaning for it. A code with no published meaning cannot support a diagnosis, and the honest answer is to work from what the client is doing rather than from the number.
Range failures are real, and Microsoft does publish codes for them. They are DO_E_INVALID_RANGE (0x80D05010), DO_E_INSUFFICIENT_RANGE_SUPPORT (0x80D05011), DO_E_OVERLAPPING_RANGES (0x80D05012) and DO_E_HTTP_BLOCKSIZE_MISMATCH (0x80D05001), which is an HTTP server returning a response whose data size does not match what was requested. If a middlebox is mangling byte ranges, those are the codes you will see.
Peering does still fail for ordinary reasons, and they are worth knowing because they explain the pattern people describe as random. A client has to be reachable inbound on TCP 7680 to serve content, and it has to be in the same peer scope as the machine that holds the content. Peer selection is not arbitrary: Microsoft documents that the group is chosen automatically from the AD DS site, then the authenticated domain SID, then the Microsoft Entra tenant ID, with GroupID and GroupIDSource available to override it. A machine that ends up in a group of one behaves exactly like a machine with peering switched off.
The job was lost or never created
You have this one if 0x80D02003, and turning peering off changes nothing.
- Run
Delete-DeliveryOptimizationCache -ForceandRestart-Service DoSvc. - Stop wuauserv and bits, rename
C:\Windows\SoftwareDistribution, restart the services and rescan so the download is requested again from scratch. - Check free space on the system volume before retrying.
- Collect
Get-DeliveryOptimizationLogAnalysisoutput if it recurs, so you can see which stage lost the job.
The peer port is blocked inbound
You have this one if Errors clear the moment you set download mode 0, and peer bytes in Get-DeliveryOptimizationStatus are always zero on the machines that fail.
- On a failing client, open Windows Defender Firewall with Advanced Security and confirm the Delivery Optimization inbound rules are enabled for the profile the machine is actually on.
- Allow TCP 7680 inbound for the domain and private profiles if the rules are missing.
- Ask the network team to allow 7680 between client subnets. Peering across VLANs is blocked by default in many designs.
- Retest with download mode 1 and confirm peer bytes are non-zero on at least one machine.
Check which firewall profile the adapter is using before you edit rules. A machine that has classified the corporate network as Public ignores rules you added to the Domain profile.
Peer selection is scoped wrongly for your network
You have this one if Peering works within a site but never between sites, or never works at all in an estate behind one shared public address.
- Decide the boundary you actually want. LAN mode matches peers on the same network; Group mode peers across internal subnets within a group, and is Microsoft’s recommended option for most organisations.
- Set Group mode with
Set-DODownloadMode -DownloadMode 2and allow UDP 3544 for Teredo, which Group and Internet modes require. - Leave the group source alone unless you have a reason to change it. Without GroupID or GroupIDSource configured, the default order is AD site, then authenticated domain SID, then Microsoft Entra tenant ID.
- Confirm the result with
Get-DeliveryOptimizationStatuson two machines that should be able to see each other.
Cache limits mean there is nothing to share
You have this one if Peering is configured correctly but peer bytes stay at zero across the whole estate.
- Check the Delivery Optimization policies under Computer Configuration – Administrative Templates – Windows Components – Delivery Optimization for a maximum cache size or age set very low.
- Check the minimum file size to cache. Below that threshold nothing is peered at all, which is normal rather than a fault.
- Confirm there is free space on the system volume. Delivery Optimization reduces or abandons caching as the disk fills.
- Leave one machine downloading a large update and watch whether others then take content from it.
A VPN or virtual switch is in the way
You have this one if Peering fails only when the VPN is connected, or only on virtual machines.
- Test with the VPN disconnected. If peering works, the fix belongs in the VPN client’s routing and filtering rules rather than in Windows.
- On virtual machines, confirm the virtual switch is not isolating guests from each other.
- For roaming laptops that spend most of their time off the corporate network, mode 0 or mode 1 is usually the sensible permanent setting.
- Confirm the adapter has not classified the network as Public, which changes which firewall rules apply.
Full reference
What is published and what is not
| Code | Status | Reading |
|---|---|---|
| 0x80D02003 | Published | DO_E_JOB_NOT_FOUND: the job was not found, expecting a job to be present |
| 0x80D0000A | Not published | Absent from Microsoft’s Delivery Optimization return value list. Diagnose from the client’s behaviour, not from the number |
| 0x80D02005 | Not published | Absent from the same list. The neighbouring documented values are 0x80D02004 DO_E_JOB_EMPTY and 0x80D0200D DO_E_LOCALPATH_NOT_SET |
| 0x80D01009 | Not published | Absent from the same list. The only documented 0x80D010xx value is 0x80D01001 DO_E_NO_SERVICE |
Three of the four codes in this article have no published meaning. That is not a failure of research; vendors do not document every internal value, and the correct response is to describe them functionally and diagnose from the surrounding evidence. It does mean you should be wary of any source that confidently tells you what 0x80D0000A means.
The range codes, for when it really is byte ranges
| Code | Published name | What it says |
|---|---|---|
| 0x80D05010 | DO_E_INVALID_RANGE | The specified byte range is invalid |
| 0x80D05011 | DO_E_INSUFFICIENT_RANGE_SUPPORT | The server does not support the necessary HTTP protocol |
| 0x80D05012 | DO_E_OVERLAPPING_RANGES | The list of byte ranges contains overlapping ranges, which are not supported |
| 0x80D05001 | DO_E_HTTP_BLOCKSIZE_MISMATCH | The HTTP server returned a response with a data size not equal to what was requested |
If a caching proxy or an inspection appliance is rewriting partial content, these are the codes to expect. Seeing one of them is a concrete instruction: find the device in the path that is not honouring range requests and take it out of the path for Microsoft’s content endpoints.
Download modes and what each one costs you
| Mode | Name | Behaviour |
|---|---|---|
| 0 | HTTP Only | No peering. Still uses cloud service metadata for an efficient download from the original source or a Connected Cache server |
| 1 | LAN | The default. Peer sharing on the same network |
| 2 | Group | Peering across internal subnets within a group. Microsoft’s recommended option for most organisations |
| 3 | Internet | Peer sources across the internet, plus HTTP |
| 99 | Simple | Cloud services disabled entirely, for offline environments. The client falls back to this automatically when the cloud services are unreachable |
| 100 | Bypass | Deprecated starting in Windows 11. Do not configure it |
The ports, precisely
| Port | Protocol | Purpose |
|---|---|---|
| 7680 | TCP/IP | Peer-to-peer content sharing between devices on the network |
| 3544 | UDP | Teredo NAT traversal, required when using download mode Group (2) or Internet (3) |
| 443 | HTTPS, TLS 1.2 | Client to Delivery Optimization cloud service |
| 67 and 68 | UDP | DHCP client, for DHCP-based groups and cache server discovery |
DoSvc shuts down when idle and stops listening on 7680, so a port scan that finds nothing listening is not evidence of a blocked port. Test while a download is actually running.
How the group is chosen when you have not chosen it
Group mode does not need a manually assigned identifier for most estates. Where neither GroupID nor GroupIDSource is configured, the documented default order is the AD DS site, then the authenticated domain SID, then the Microsoft Entra tenant ID. GroupIDSource can select one of those explicitly, or DHCP option 234 or the DNS suffix, and GroupID takes a GUID you set on each device for groups that cross those boundaries.
The failure mode worth watching for is inconsistency. Mixing sources across an estate produces groups of one, which behaves exactly like no peering at all while looking configured. Pick one method and apply it everywhere, then confirm with Get-DeliveryOptimizationStatus on two machines that should be able to see each other.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
0x80D0000A |
Seen against failing Delivery Optimization downloads. Microsoft’s published return value list does not carry it, so diagnose from the download mode, the peer ports and the service log rather than from this code | not published by the vendor |
0x80D02003 |
DO_E_JOB_NOT_FOUND: the job was not found when one was expected to be present | Microsoft Learn |
0x80D02005 |
Seen alongside job failures. Not in the published return value list; the documented neighbours are 0x80D02004 DO_E_JOB_EMPTY and 0x80D0200D DO_E_LOCALPATH_NOT_SET | not published by the vendor |
0x80D01009 |
Seen during Delivery Optimization downloads. Not in the published return value list; the only documented 0x80D010xx value is 0x80D01001 DO_E_NO_SERVICE | not published by the vendor |
Confirm the fix worked
- Run
Get-DODownloadModeand confirm the mode is the one you intended. - Start a download on two machines and run
Get-DeliveryOptimizationStatuson both, confirming peer bytes are non-zero if you meant to keep peering. - Confirm the update completes in Windows Update with no code against it.
- Check the firewall rules survived a restart and a network profile change.
- Open the Delivery Optimization activity monitor in Settings and confirm the sources match what you configured.
Questions people ask about this
Is 0x80D02003 a peer caching error?
No. It is DO_E_JOB_NOT_FOUND: the job was not found when one was expected to be present. That is a lost or never-created download job. If peers really are serving unusable byte ranges, the codes for that are 0x80D05010, 0x80D05011, 0x80D05012 and 0x80D05001.
What does 0x80D0000A mean?
Microsoft does not publish a meaning for it, and neither should anyone else. It appears against failing Delivery Optimization downloads, so work from what the client is doing: the download mode, whether the peer ports are open, and the service’s own log.
Is peering worth configuring at all?
On a site with many machines and a constrained internet link, yes. It moves most of a cumulative update across the LAN once rather than pulling it per machine. On a small office with fast internet, mode 0 is simpler. Microsoft recommends Group mode for most organisations that do want it.
Do I need to set a group identifier?
Usually not. Where neither GroupID nor GroupIDSource is configured, the group is chosen automatically from the AD DS site, then the authenticated domain SID, then the Microsoft Entra tenant ID. Set one only when you need a boundary those do not give you, and then set it consistently.
Does any of this require a paid product?
No. Delivery Optimization is part of Windows and the settings above are free. Central policy control is easier on editions that include Group Policy, but the peering behaviour itself is not a licensed feature.
