From d1521547ef7e00c796b14e40b8219372c8122379 Mon Sep 17 00:00:00 2001
From: Michael Piatek <michael.piatek@databricks.com>
Date: Wed, 28 Oct 2026 06:34:02 -0700
Subject: [PATCH] Raise the `databricks --version` probe timeout to 30s

On a loaded arca box the dbexec-wrapped CLI can take more than 10s to
answer `--version`. The timeout reads as an unparseable version, so the
launch fails with "Could not parse Databricks CLI version" even though the
re-probe for the error message returns a valid one.

Co-authored-by: Isaac <no-reply@databricks.com>
---
 src/ucode/databricks.py  | 14 ++++++++++++--
 tests/test_databricks.py | 11 +++++++++++
 2 files changed, 23 insertions(+), 2 deletions(-)

diff --git a/src/ucode/databricks.py b/src/ucode/databricks.py
index 69c0267f2..7358b1cf3 100644
--- a/src/ucode/databricks.py
+++ b/src/ucode/databricks.py
@@ -66,6 +66,10 @@
 MIN_DATABRICKS_CLI_VERSION = (1, 0, 0)
 # v1.11.0 fixes `fs cp` (create -> finalize), which the skills MCP uploads rely on.
 SKILLS_MCP_MIN_DATABRICKS_CLI_VERSION = (1, 11, 0)
+# `databricks --version` is normally instant, but a wrapper binary (e.g. dbexec on
+# arca) under concurrent startups can take several seconds; a timeout reads as an
+# unparseable version and fails the launch.
+_DATABRICKS_CLI_VERSION_TIMEOUT_SECONDS = 30
 TOKEN_REFRESH_INTERVAL_SECONDS = 1800
 # Substrings the Databricks CLI emits when it loses the token-cache write lock
 # to a concurrent `databricks auth token` (e.g. another ucode helper process
@@ -789,7 +793,13 @@ def _read_databricks_cli_version(path: str) -> tuple[int, int, int] | None:
     place — every other caller reads a version through ``databricks_cli_path()``.
     """
     try:
-        result = run([path, "--version"], check=False, capture_output=True, text=True, timeout=10)
+        result = run(
+            [path, "--version"],
+            check=False,
+            capture_output=True,
+            text=True,
+            timeout=_DATABRICKS_CLI_VERSION_TIMEOUT_SECONDS,
+        )
     except (OSError, subprocess.TimeoutExpired):
         return None
     raw = result.stdout or result.stderr or ""
@@ -980,7 +990,7 @@ def ensure_databricks_cli_version(
                 check=False,
                 capture_output=True,
                 text=True,
-                timeout=10,
+                timeout=_DATABRICKS_CLI_VERSION_TIMEOUT_SECONDS,
             )
             raw = result.stdout or result.stderr or ""
             output = (raw if isinstance(raw, str) else raw.decode(errors="replace")).strip()
diff --git a/tests/test_databricks.py b/tests/test_databricks.py
index b0efa989f..4959f9b8e 100644
--- a/tests/test_databricks.py
+++ b/tests/test_databricks.py
@@ -4280,6 +4280,17 @@ def boom(*a, **kw):
         monkeypatch.setattr(db_mod, "run", boom)
         assert db_mod._read_databricks_cli_version("/usr/bin/databricks") is None
 
+    def test_read_version_allows_slow_cli(self, monkeypatch):
+        seen = {}
+
+        def fake_run(*a, **kw):
+            seen["timeout"] = kw["timeout"]
+            return subprocess.CompletedProcess(a, 0, "Databricks CLI v1.20.0", "")
+
+        monkeypatch.setattr(db_mod, "run", fake_run)
+        db_mod._read_databricks_cli_version("/usr/bin/databricks")
+        assert seen["timeout"] == db_mod._DATABRICKS_CLI_VERSION_TIMEOUT_SECONDS >= 30
+
     def test_read_version_parses_output(self, monkeypatch):
         monkeypatch.setattr(
             db_mod,