bug: FailedToOpenSocket when workspace lacks AI Gateway

[dgokeeffe]

Problem

get_gateway_host() auto-constructs an AI Gateway URL from DATABRICKS_WORKSPACE_ID (always set on Databricks Apps) without checking if the gateway actually exists. Workspaces without AI Gateway enabled get a FailedToOpenSocket error from Claude Code (Bun runtime) when it tries to connect to the non-existent host.

The fallback to {DATABRICKS_HOST}/serving-endpoints/anthropic never triggers because the gateway URL is returned unconditionally.

Affected code

• utils.py:get_gateway_host() — returns unreachable URL without probing
• All setup scripts (setup_claude.py, setup_codex.py, setup_gemini.py, setup_opencode.py, setup_proxy.py) — inherit the bad URL

Fix

• Probe auto-discovered gateway URLs with a 2s timeout GET before using them
• Cache the result in _GATEWAY_RESOLVED env var so subprocesses skip re-probing
• Add resolve_and_cache_gateway() helper called once at boot

PR: #114

[datasciencemonkey]

Fixed in #123 (v0.17.1) — AI Gateway URLs are now probed for reachability with a 2s timeout. Workspaces without AI Gateway gracefully fall back to {DATABRICKS_HOST}/serving-endpoints. Credit: @dgokeeffe.