Skip to content

docs: add IPv6 readiness guidance for Compute - #118

Open
dlotterman wants to merge 8 commits into
mainfrom
dlott_agents+ipv6
Open

dlotterman wants to merge 8 commits into
mainfrom
dlott_agents+ipv6

Conversation

@dlotterman

@dlotterman dlotterman commented Nov 3, 2026 •

Copy link
Copy Markdown
Contributor

Stacked on #120. This branch sits on top of the docs/private-dns-guides commits from #120, so the first six commits here are Scot's and belong to that PR. The only commit specific to this PR is the last one, docs: add IPv6 readiness guidance for Compute (6 files, +31/−4). Merge #120 first and this PR's diff shrinks to that one commit.

Summary

Compute and gVPC are IPv6 only. Agents and users keep running into problems because apps bind to IPv4 (0.0.0.0, listen 80;) or connect to IPv4 addresses, and the docs don't say so until a deploy fails.

This PR puts the guidance where people and agents look first, without adding a new page.

  • Compute troubleshooting: new "IPv6 readiness" section with the pre-deploy checklist, bind examples for nginx, Node.js, Python, and Go, a local ss check, and steps for agents.
  • Compute overview: short "IPv6 readiness" section that links to the troubleshooting section, and a note that Instances are IPv6 only.
  • First-workload tutorial ("What's next") and ALB prerequisites: link to the readiness section. The tutorial itself is unchanged for readers who just follow it.
  • Deploy workloads (datumctl): tells agents to search the app's code and config for IPv4 addresses before they deploy.
  • Compute configuration: notes that environment values should use hostnames, not IPv4 literals.

Testing

I compared this branch with its parent on seven IPv6-related agent tasks, four reps each, across Haiku, Luna, DeepSeek and GLM (28 runs per model per condition). Accuracy with this change versus without: Haiku 100% vs 93%, Luna 93% vs 82%, DeepSeek 96% vs 93%, GLM 96% vs 96%. Most of the gain is on the "sweep an app for IPv4 assumptions" task, where agents without this change told users to use IPv6 literals because they thought private DNS names were unavailable. Keyword scoring is crude and samples are small, so treat the numbers as directional.

Review notes

  • The server examples come from general knowledge, not from a live Datum deploy. Please check them, especially the Node.js, Python, and Go snippets.
  • Nothing here changes the Compute API or behavior. It only documents existing IPv6-only networking.

@mintlify

mintlify Bot commented Nov 3, 2026 •

Copy link
Copy Markdown
Contributor

Preview deployment for your docs. Learn more about Mintlify Previews.

Project Status Preview Updated
Datum 🟢 Ready View Preview Nov 11, 2026, 11:50 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

Add a quickstart that reaches an Instance by its private name and then
by a custom name, a guide to private zones, records, and network
associations, a limits and behavior reference, and Private DNS
troubleshooting. Stop describing Private DNS as planned.
Use an AAAA record in the quickstart, since CNAMEs into another zone
are not followed yet. Add the Network Admin role, use made-up names in
examples, and correct the troubleshooting reasons.
Private zones no longer need a zone class.
@cla-assistant

cla-assistant Bot commented Nov 4, 2026 •

Copy link
Copy Markdown

CLA assistant check
All committers have signed the CLA.

scotwells and others added 3 commits November 4, 2026 14:09
Describe private zone visibility only as association-based, and say
what happens past 256 zones on a network, since that isn't refused yet.
Compute and gVPC are IPv6 only. Apps that bind to IPv4 (0.0.0.0,
listen 80;) or connect to IPv4 addresses fail, and the docs didn't say
so until a deploy failed.

- Add an "IPv6 readiness" section to Compute troubleshooting with the
  pre-deploy checklist, bind examples, a local check, and steps for agents.
- Add a short "IPv6 readiness" section to the Compute overview that links
  to it, and note that Instances are IPv6 only.
- Link the first-workload tutorial's "What's next" and the ALB
  prerequisites to the readiness section.
- Tell agents on the deploy-workloads page to search the app's code and
  config for IPv4 addresses before they deploy, and note in the
  configuration page that env values should use hostnames.
@dlotterman

Copy link
Copy Markdown
Contributor Author

Theres some real improvement here:

  ┌──────────┬──────┬──────┐                                                                                                                                                                                                                  
  │  Model   │ base │ ours │                                                                                                                                                                                                                  
  ├──────────┼──────┼──────┤                                                                                                                                                                                                                  
  │ Haiku    │ 93%  │ 100% │                                                                                                                                                                                                                  
  ├──────────┼──────┼──────┤                                                                                                                                                                                                                  
  │ Luna     │ 82%  │ 93%  │                                                                                                                                                                                                                  
  ├──────────┼──────┼──────┤                                                                                                                                                                                                                  
  │ DeepSeek │ 93%  │ 96%  │                                                                                           
  ├──────────┼──────┼──────┤                                                                                                                                                                                                                  
  │ GLM      │ 96%  │ 96%  │                                                                                                                                                                                                                  
  ├──────────┼──────┼──────┤                                                                                                                                                                                                                  
  │ DeepSeek │ 93%  │ 96%  │                                                                                                                                                                                                                  
  ├──────────┼──────┼──────┤                                                                                                                                                                                                                  
  │ GLM      │ 96%  │ 96%  │                                                                                                                                                                                                                  
  └──────────┴──────┴──────┘ 

I think this has enough merit to merge and anything that gets missed we can include in #157 or other cleanup

@dlotterman
dlotterman requested a review from scotwells November 12, 2026 00:02

This branch was successfully deployed

1 active deployment
staging — bb6712d5 Deployed Nov 11, 2026 by mintlify[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants