Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Run shell suites and Windows browser checks in parallel (#12899)

oobabooga06b90151aff4
diff --git a/.github/workflows/studio-backend-ci.yml b/.github/workflows/studio-backend-ci.yml
index 7a30863fc7c..8b719f36281 100644
--- a/.github/workflows/studio-backend-ci.yml
+++ b/.github/workflows/studio-backend-ci.yml
@@ -613,17 +613,42 @@ jobs:
# Skipped:
# test_install_rollback_lifecycle.sh: covered by cross-platform-parity-ci.yml.
run: |
- set -e
+ set -euo pipefail
+ shopt -s nullglob
skip="test_install_rollback_lifecycle.sh"
- found=0
+ suites=()
for s in tests/sh/test_*.sh; do
case " $skip " in
*" $(basename "$s") "*) echo "skipping $s (see workflow comment)"; continue ;;
esac
- found=$((found + 1))
+ suites+=("$s")
+ done
+ found=${#suites[@]}
+ [ "$found" -gt 0 ] || { echo "::error::no shell tests discovered under tests/sh"; exit 1; }
+ export suite_logs
+ suite_logs=$(mktemp -d "$RUNNER_TEMP/shell-suites.XXXXXX")
+ # Each suite already owns its fixtures. Keep its output and exit status
+ # separate, then collect every result even when another suite fails.
+ failed=0
+ printf '%s\0' "${suites[@]}" | xargs -0 -P 4 -n 1 bash -c '
+ s=$1
+ name=${s##*/}
+ echo "Starting $s"
+ rc=0
+ timeout --signal=TERM --kill-after=10s 480s bash "$s" > "$suite_logs/$name.log" 2>&1 || rc=$?
+ printf "%s\n" "$rc" > "$suite_logs/$name.status"
+ echo "Finished $s (exit $rc)"
+ ' _ || failed=1
+ for s in "${suites[@]}"; do
+ name=${s##*/}
echo "::group::$s"
- bash "$s"
+ cat "$suite_logs/$name.log" || failed=1
+ rc=$(cat "$suite_logs/$name.status" 2>/dev/null) || rc=1
+ if [ "$rc" != 0 ]; then
+ echo "::error::$s exited $rc (124 means the suite timed out)"
+ failed=1
+ fi
echo "::endgroup::"
done
- [ "$found" -gt 0 ] || { echo "::error::no shell tests discovered under tests/sh"; exit 1; }
echo "ran $found shell installer test files"
+ exit "$failed"
diff --git a/.github/workflows/studio-frontend-ci.yml b/.github/workflows/studio-frontend-ci.yml
index 254d736b1e9..1f09e3ad01d 100644
--- a/.github/workflows/studio-frontend-ci.yml
+++ b/.github/workflows/studio-frontend-ci.yml
@@ -700,16 +700,28 @@ jobs:
timeout-minutes: 3
run: python tests/studio/playwright_composer_icons.py chromium firefox
- - name: Data settings deletion choices in Chromium
+ - name: Data settings deletion choices in Chromium and Firefox
working-directory: ${{ github.workspace }}
- run: python tests/studio/playwright_data_settings.py
-
- - name: Data settings deletion choices in Firefox
- working-directory: ${{ github.workspace }}
- env:
- PW_ENGINE: firefox
- PW_OUT: logs/data_settings_firefox_report.json
- run: python tests/studio/playwright_data_settings.py
+ run: |
+ mkdir -p logs
+ # each driver owns its Vite server, optimizer cache, and report
+ PW_ENGINE=chromium PW_PORT=5420 PW_OUT=logs/data_settings_report.json \
+ VITE_TEST_CACHE_DIR="$RUNNER_TEMP/vite-data-chromium" \
+ python tests/studio/playwright_data_settings.py > logs/data_settings_chromium.log 2>&1 &
+ chromium_pid=$!
+ PW_ENGINE=firefox PW_PORT=5421 PW_OUT=logs/data_settings_firefox_report.json \
+ VITE_TEST_CACHE_DIR="$RUNNER_TEMP/vite-data-firefox" \
+ python tests/studio/playwright_data_settings.py > logs/data_settings_firefox.log 2>&1 &
+ firefox_pid=$!
+ failed=0
+ wait "$chromium_pid" || failed=1
+ wait "$firefox_pid" || failed=1
+ for engine in chromium firefox; do
+ echo "::group::Data settings: $engine"
+ cat "logs/data_settings_$engine.log" || failed=1
+ echo "::endgroup::"
+ done
+ exit "$failed"
- name: Upload alignment failure artifacts
if: failure()
diff --git a/.github/workflows/workflow-trigger-lint.yml b/.github/workflows/workflow-trigger-lint.yml
index 1d677be6032..e7a5a106d57 100644
--- a/.github/workflows/workflow-trigger-lint.yml
+++ b/.github/workflows/workflow-trigger-lint.yml
@@ -141,6 +141,7 @@ jobs:
tests/studio/test_ui_shard_engines.py \
tests/studio/test_zoo_suite_parallel_isolation.py \
tests/studio/test_core_setup_zoo_revision_notice.py \
+ tests/studio/test_ci_parallel_browser_results.py \
tests/studio/test_ci_shell_suite_coverage.py \
tests/studio/test_pip_cache_naming.py \
tests/studio/test_compile_caches_are_per_worker.py \
diff --git a/studio/frontend/vite.config.ts b/studio/frontend/vite.config.ts
index 0b24777958f..bf8e6cd80e8 100644
--- a/studio/frontend/vite.config.ts
+++ b/studio/frontend/vite.config.ts
@@ -24,13 +24,13 @@ function smokeModuleDelay(): Plugin {
};
}
-// https://vite.dev/config/
export default defineConfig({
+ // concurrent browser checks use separate dependency optimizer caches.
+ cacheDir: process.env.VITE_TEST_CACHE_DIR || "node_modules/.vite",
// Reasoning's highlighter loads only the grammar it needs in its module worker.
worker: { format: "es" },
plugins: [react(), tailwindcss(), smokeModuleDelay()],
- // Keep an unrelated PostCSS config in an ancestor directory from leaking
- // into Unsloth installs. Tailwind is provided by its dedicated Vite plugin.
+ // prevent ancestor PostCSS configs from leaking into installs; Tailwind uses its Vite plugin.
css: {
postcss: {
plugins: [],
diff --git a/tests/studio/playwright_data_settings.py b/tests/studio/playwright_data_settings.py
index 1628326b6cb..17a10ca2a22 100644
--- a/tests/studio/playwright_data_settings.py
+++ b/tests/studio/playwright_data_settings.py
@@ -850,7 +850,8 @@ def run_restore_notifications(page):
expect(dialog.get_by_role("heading")).to_have_text(f"{heading} ({count})")
confirm = "Unarchive" if action == "restore" else "Delete"
dialog.get_by_role("button", name = confirm, exact = True).click()
- expect(dialog).to_have_count(0)
+ # allow shared runners 30 seconds for 23 sequential mocked HTTP requests
+ expect(dialog).to_have_count(0, timeout = 30_000)
completed = count if failure_index is None else failure_index
expected_events = [kind] if completed and action == "restore" else []
observed = page.evaluate("window.__dataFixture.notifications")
diff --git a/tests/studio/test_ci_parallel_browser_results.py b/tests/studio/test_ci_parallel_browser_results.py
new file mode 100644
index 00000000000..a47626e29f6
--- /dev/null
+++ b/tests/studio/test_ci_parallel_browser_results.py
@@ -0,0 +1,64 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import os
+from pathlib import Path
+import shutil
+import subprocess
+
+import pytest
+import yaml
+
+
+WORKFLOW = Path(__file__).resolve().parents[2] / ".github/workflows/studio-frontend-ci.yml"
+
+
+def _workflow_bash():
+ if os.name != "nt":
+ return shutil.which("bash")
+ git = shutil.which("git")
+ if not git:
+ return None
+ candidate = Path(git).resolve().parent.parent / "bin" / "bash.exe"
+ return str(candidate) if candidate.is_file() else None
+
+
+BASH = _workflow_bash()
+pytestmark = pytest.mark.skipif(not BASH, reason = "requires the workflow's bash shell")
+
+
[email protected]("failing", ["", "chromium", "firefox"])
+def test_browser_children_are_isolated_and_both_results_gate(tmp_path, failing):
+ steps = yaml.safe_load(WORKFLOW.read_text(encoding = "utf-8"))["jobs"]["windows"]["steps"]
+ script = next(
+ s["run"]
+ for s in steps
+ if s.get("name") == "Data settings deletion choices in Chromium and Firefox"
+ )
+ stub = r"""
+python() {
+ printf '%s %s %s %s\n' "$PW_ENGINE" "$PW_PORT" "$PW_OUT" "$VITE_TEST_CACHE_DIR"
+ touch "$PW_ENGINE.ready"
+ for attempt in $(seq 1 100); do
+ [ -e chromium.ready ] && [ -e firefox.ready ] && break
+ sleep 0.02
+ done
+ [ -e chromium.ready ] && [ -e firefox.ready ] || return 9
+ [ "$PW_ENGINE" != "$FAIL_ENGINE" ] || return 7
+}
+"""
+ result = subprocess.run(
+ [BASH, "-e", "-o", "pipefail", "-c", stub + script],
+ cwd = tmp_path,
+ env = {**os.environ, "RUNNER_TEMP": str(tmp_path), "FAIL_ENGINE": failing},
+ text = True,
+ capture_output = True,
+ timeout = 15,
+ )
+ assert result.returncode == bool(failing), result.stdout + result.stderr
+ assert "chromium 5420 logs/data_settings_report.json" in result.stdout
+ assert "firefox 5421 logs/data_settings_firefox_report.json" in result.stdout
+ assert f"{tmp_path}/vite-data-chromium" in result.stdout
+ assert f"{tmp_path}/vite-data-firefox" in result.stdout
+ assert (tmp_path / "chromium.ready").exists()
+ assert (tmp_path / "firefox.ready").exists()
diff --git a/tests/studio/test_ci_shell_suite_coverage.py b/tests/studio/test_ci_shell_suite_coverage.py
index 1d76cea0567..040f5675290 100644
--- a/tests/studio/test_ci_shell_suite_coverage.py
+++ b/tests/studio/test_ci_shell_suite_coverage.py
@@ -1,28 +1,12 @@
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
-"""Guards that the installer test suites actually run on a PR.
-
-Two ways coverage went missing without anyone noticing:
-
-1. Backend CI ran a hardcoded list of tests/sh/*.sh files. New tests were added
- to the directory and never to the list, so by the time this was written the
- list was seven files behind -- including test_strixhalo_wsl_reroute.sh, the
- only shell coverage of the ROCm WSL reroute, which had never run on a PR.
- tests/run_all.sh, the local entrypoint, had drifted the other way.
-
-2. Backend CI's path filter did not include install.sh / install.ps1, while a
- large share of the suites it runs (tests/sh/*, tests/studio/install/*) assert
- against exactly those two files. An install-only change -- the shape most
- AMD/ROCm routing fixes take, e.g. #7277 / #7293 / #7300 -- skipped the
- workflow that tests it.
-
-Both are now discovery-based. These tests fail if either reverts to a list, if a
-shell test lands somewhere the discovery cannot see it, or if a skip is added
-without a reason next to it.
-"""
+"""guard installer shell-suite discovery, path triggers, and parallel result reporting."""
+import os
import re
+import shutil
+import subprocess
from pathlib import Path
import pytest
@@ -37,7 +21,6 @@ _PARITY_CI = _WORKFLOWS / "cross-platform-parity-ci.yml"
_RUN_ALL = REPO_ROOT / "tests" / "run_all.sh"
_SH_DIR = REPO_ROOT / "tests" / "sh"
-# Files deliberately not run by the auto-discovered Backend CI step.
_EXPECTED_CI_SKIPS = {
"test_install_rollback_lifecycle.sh": "runs on both platforms in cross-platform-parity-ci.yml",
}
@@ -370,7 +353,7 @@ class TestWindowsPowerShellStepsAreGated:
)
def test_the_rule_finds_the_steps_it_is_about(self):
- """Without this, a walk that matched nothing would leave the rule above vacuous."""
+ """prevent a vacuous pass when workflow traversal finds no matching steps."""
found = 0
for workflow in sorted(_WORKFLOWS.glob("*.yml")):
doc = yaml.safe_load(workflow.read_text(encoding = "utf-8"))
@@ -382,3 +365,60 @@ class TestWindowsPowerShellStepsAreGated:
f"only {found} Windows PowerShell steps found on multi-OS jobs; the walk above is "
f"looking in the wrong place"
)
+
+
+ os.name != "posix" or not all(shutil.which(x) for x in ("bash", "xargs", "timeout")),
+ reason = "the Linux shell-suite runner needs bash, xargs and timeout",
+)
+class TestParallelShellResults:
+ def run_step(self, tmp_path, suites):
+ directory = tmp_path / "tests" / "sh"
+ directory.mkdir(parents = True)
+ for name, body in suites.items():
+ (directory / name).write_text(body)
+ return subprocess.run(
+ ["bash", "-c", _shell_step_script()],
+ cwd = tmp_path,
+ env = {**os.environ, "RUNNER_TEMP": str(tmp_path)},
+ capture_output = True,
+ text = True,
+ timeout = 15,
+ )
+
+ def test_a_failure_does_not_hide_other_suite_results(self, tmp_path):
+ result = self.run_step(
+ tmp_path,
+ {
+ "test_a.sh": "echo first-failure; exit 7",
+ "test_b.sh": "echo second-success",
+ "test_c with spaces.sh": "echo third-success",
+ "test_install_rollback_lifecycle.sh": "echo should-not-run; exit 99",
+ },
+ )
+ assert result.returncode == 1
+ assert "first-failure" in result.stdout
+ assert "second-success" in result.stdout
+ assert "third-success" in result.stdout
+ assert "test_a.sh exited 7" in result.stdout
+ assert "should-not-run" not in result.stdout
+ assert "ran 3 shell installer test files" in result.stdout
+
+ def test_empty_discovery_fails(self, tmp_path):
+ result = self.run_step(tmp_path, {})
+ assert result.returncode != 0
+ assert "no shell tests discovered under tests/sh" in result.stdout
+
+ def test_four_suites_can_reach_a_barrier_together(self, tmp_path):
+ body = """touch "${0##*/}.ready"
+for attempt in $(seq 1 100); do
+ count=$(find . -maxdepth 1 -name '*.ready' | wc -l)
+ [ "$count" -eq 4 ] && exit 0
+ sleep 0.02
+done
+echo "the other suites never reached the barrier"
+exit 1
+"""
+ result = self.run_step(tmp_path, {f"test_{i}.sh": body for i in range(4)})
+ assert result.returncode == 0, result.stdout + result.stderr
+ assert "ran 4 shell installer test files" in result.stdout