Find the API node when TAILSCALE_HOSTNAME names the desk, not a number

[david-hummingbot]

tailnet_api_peers (added in #231) matched hummingbot-api plus a numeric suffix, because that is the only suffix Tailscale itself appends when a name is taken. But TAILSCALE_HOSTNAME is a setting, and naming the API per desk — hummingbot-api-cornell, hummingbot-api-eu — is the ordinary way to run more than one.

For those the search found nothing, and nothing is the branch that warns:

! No hummingbot-api node is visible on this tailnet yet.
→ Deploy it on that machine first (its setup joins the tailnet), or
→ enter the name/address it is reachable at.

…about a node sitting in plain sight in tailscale status, and then asks the operator to type the name it had just declined to recognise.

A lone match is not a confirmed match

Widening the suffix also widens what reaches the count == 1 branch, and that branch selected silently (thanks @greptile-apps for catching it). A stale node, a colleague's staging box, another desk's API — each is alone on a tailnet the intended machine has not joined yet, and each would be accepted and written into config.yml with only a ✓ Found it to show for it. The wrong node is very likely running the same software, so it answers, and the mistake arrives as 401 Incorrect username or password against a password that was never wrong — the exact failure #231 set out to stop.

So count == 1 splits:

found	behaviour
hummingbot-api (unsuffixed)	taken silently — it is the name hummingbot-api's own setup asks for, and there is nothing to choose between
hummingbot-api-cornell (suffixed)	shown and confirmed, pre-filled so Enter accepts it

It costs one keystroke in the common case, and no name the operator never chose lands in config.yml unseen.

"Pick the one you just deployed" was wrong guidance

The multi-match message asserted:

→ Tailscale adds -1, -2 ... when a name is taken, so these are
→ different machines. Pick the one you just deployed.

Running several APIs on purpose — hummingbot-api-1, hummingbot-api-2 alongside the default — is an ordinary deployment, and there those suffixes are chosen names, not collision artifacts. That wording told such an operator their own naming was an accident. And "the one you just deployed" is the wrong instruction regardless when Condor is being pointed at an instance that has been up for weeks. Now:

! More than one hummingbot-api node on this tailnet:
      • hummingbot-api
      • hummingbot-api-1
      • hummingbot-api-2
→ These are separate machines -- deployments named on purpose, or
→ names Tailscale suffixed because one was already taken.
→ Pick the one this Condor should talk to.

Verification

Driven end to end through the real wizard in a container, against a tailnet whose only API node was hummingbot-api-cornell (tailscale stubbed, a stand-in API bound off-loopback so the localhost probe could not see it). Before:

! No hummingbot-api node is visible on this tailnet yet.
  hummingbot-api host [hummingbot-api]: hummingbot-api-cornell

After:

✓ Found one hummingbot-api node on your tailnet: hummingbot-api-cornell
→ That is not the default name, so it is worth a look before it
→ goes in config.yml. Press Enter to use it, or type another.
  hummingbot-api host [hummingbot-api-cornell]:

…and Enter is the whole answer. Both runs end with ✓ API reachable and credentials accepted and the same config.yml. All four branches were also rendered through the real decision block with stubbed prompts.

tests/test_setup_tailnet_peers.py adds 16 assertions over the helper. They source the real function out of setup-environment.sh with a stubbed tailscale ahead of it on PATH, rather than restating the regex in Python — a copy of the regex asserts only that the copy matches itself. Four of them fail against the old -[0-9]+, confirmed by reverting it. They cover the numeric fleet (hummingbot-api + -1 + -2, which must reach the pick-one branch and never auto-select), desk names, and the nodes that must not be claimed (condor-hackathon, condor1, hummingbot-apis, my-hummingbot-api).

Full suite: 5109 passed, 23 skipped.

(An earlier revision of this description claimed 13 stubbed assertions; they were ad-hoc and never committed. They are in the tree now.)

🤖 Generated with Claude Code

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Changes how the setup wizard finds the API node on the tailnet.

The PR appears safe to merge.

Summary

The PR expands Tailscale API-node discovery to recognize operator-defined suffixes and prevents a lone suffixed match from being selected without being shown to the operator.

• Matches names such as hummingbot-api-cornell while retaining anchored filtering.
• Prompts before accepting a lone suffixed candidate, with the discovered hostname pre-filled.
• Clarifies guidance when multiple API nodes are visible.
• Adds focused shell-backed tests for matching and exclusion behavior.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Tailscale peer names] --> B[Filter hummingbot-api or suffixed names]
  B --> C{Number of matches}
  C -->|None| D[Prompt for hostname]
  C -->|One unsuffixed| E[Use default API node]
  C -->|One suffixed| F[Display candidate and request confirmation]
  C -->|Multiple| G[Ask operator to select a node]
  D --> H[Probe and save API configuration]
  E --> H
  F --> H
  G --> H

Loading

_{Reviews (3) · Last reviewed commit: "Run black over the new test file"}