I have been having this exact problem for months. Changing networks, pcs etc. has no effect. My backup file size is 4.3 GB, could that be the issue?
If the destination file system is FAT32 the answer is yes.
NTFS, Win 11. The error message I get when it fails is generic - ‘network issue’.
I think you have the same issue as us. Can you tell us how long it takes for the backup download to fail? For me, it seems to be always around 30s. Could be because of some token expiring.
The .part file sizes for two attempts are 1.1 and 1.9 GB of 4.3 GB total so 26% and 44%.
I don’t think you’re going to get very far on this. If you’re trying to solve the problem of getting backups downloaded to your computer, I’d recommend using scp/rsync. Another way to solve it would be to push backups to some S3-like thing and get them from there.
I was going to see if I could figure anything out, but I can’t find a site that’s got a huge backup that’s not on S3.
If you want to learn about the intricacies of Discourse, Nnginx, and reverse proxies and you’re having a good time, carry on.
I’ve been using scp since this started, just chimed in to document another instance of the issue. In case anyone wants to follow this up another possible clue is that the server space is getting cramped, “70.1% of 46.79GB”, making it difficult to even keep a few backups in rotation on it. If I succeed in talking the client into expanding the droplet I’ll post here again about the results.
I’d push the backups to S3. If you configure it in the yml file, you can restore to a new server with just the yml file. Digital Ocean spaces is just $5/month. I think Backblaze may even have a free tier.
And I’m not saying it’s not a bug, but it doesn’t look like one that anyone is interested in fixing (unless there was a response from a Discourse team member that I missed).
What’s the point in reporting bugs if they won’t be addressed? Thanks for letting us know, have a great day!
I did miss that Sam had responded and seemed to suggest a fix, though it didn’t sound like he’d assigned anyone to enact it. Maybe there’s other discussion going on that’s not visible here.
This issue appears to have been resolved. I updated to 2026.01.0-latest Friday evening and just got around to the UI backup download email link via the browser. Worked just fine.
I confirm the bug seems fixed ![]()