Skip to content

Commit 7902d58

Browse files
cdoerncdoern
committed
feat: add OpenResponses conformance CI job with replay recordings
Adds an optional, informational CI job that runs the OpenResponses compliance test suite (https://github.com/openresponses/openresponses) against llama-stack's Responses API implementation, tracking progress toward full spec conformance per issue #4818. Key design decisions: - The job is non-blocking (continue-on-error: true) because failures are expected while conformance gaps remain. It exists to make regressions and progress visible, not to gate merges. - Inference calls are replayed from checked-in recordings rather than hitting the live OpenAI API on every run. This keeps CI free, fast, and deterministic. The replay mechanism intercepts outbound OpenAI SDK calls at the client level and serves pre-recorded responses keyed by a SHA-256 hash of the normalized request. - Recordings are the PR author's responsibility to generate and commit, following the same convention as all other llama-stack integration tests. A repeatable shell script (scripts/record-openresponses-conformance.sh) handles the full local workflow: installing the ci-tests distro deps, starting the server in record-if-missing mode, running the compliance CLI, and reporting what was written. - The CI workflow uses the ci-tests distro (llama stack run ci-tests) for consistency with the rest of the integration test infrastructure. Also includes: - Initial set of 7 replay recordings covering all 6 compliance tests (basic text, streaming, system prompt, tool calling, image input, multi-turn conversation) plus the models list call - CONFORMANCE_GAPS.md documenting the 14 specific schema/serialization issues causing the current failures, with file locations and suggested fixes for each - README explaining the replay mechanism and how to update recordings Signed-off-by: Charlie Doern <cdoern@redhat.com>
1 parent eddaf1e commit 7902d58

12 files changed

Lines changed: 4537 additions & 0 deletions

‎.github/workflows/README.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -15,6 +15,7 @@ Llama Stack uses GitHub Actions for Continuous Integration (CI). Below is a tabl
1515
| Integration Tests (Replay) | [integration-tests.yml](integration-tests.yml) | Run the integration test suites from tests/integration in replay mode |
1616
| Vector IO Integration Tests | [integration-vector-io-tests.yml](integration-vector-io-tests.yml) | Run the integration test suite with various VectorIO providers |
1717
| OpenAPI Generator SDK Validation | [openapi-generator-validation.yml](openapi-generator-validation.yml) | Validate OpenAPI Generator SDK generation |
18+
| OpenResponses Conformance Tests | [openresponses-conformance.yml](openresponses-conformance.yml) | Run OpenResponses conformance tests against llama-stack Responses API |
1819
| Pre-commit | [pre-commit.yml](pre-commit.yml) | Run pre-commit checks |
1920
| Test Llama Stack Build | [providers-build.yml](providers-build.yml) | Test llama stack build |
2021
| Test llama stack list-deps | [providers-list-deps.yml](providers-list-deps.yml) | Test llama stack list-deps |

‎.github/workflows/openresponses-conformance.yml‎

Lines changed: 229 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,229 @@
1+
name: OpenResponses Conformance Tests
2+
3+
run-name: Run OpenResponses conformance tests against llama-stack Responses API
4+
5+
# This job is OPTIONAL and informational — it tracks progress toward full
6+
# conformance with the OpenResponses spec (https://openresponses.org).
7+
# Failures are expected while gaps remain in the Responses API implementation.
8+
# See: https://github.com/llamastack/llama-stack/issues/4818
9+
#
10+
# Inference calls are replayed from checked-in recordings under
11+
# tests/integration/openresponses/recordings/. To add or update recordings,
12+
# run the server locally with LLAMA_STACK_TEST_INFERENCE_MODE=record-if-missing
13+
# pointed at that directory, run the compliance tests, and commit the results.
14+
15+
on:
16+
push:
17+
branches:
18+
- main
19+
- 'release-[0-9]+.[0-9]+.x'
20+
paths:
21+
- 'src/llama_stack/providers/inline/agents/**'
22+
- 'src/llama_stack/apis/agents/**'
23+
- 'tests/integration/openresponses/**'
24+
- '.github/workflows/openresponses-conformance.yml'
25+
pull_request:
26+
branches:
27+
- main
28+
- 'release-[0-9]+.[0-9]+.x'
29+
paths:
30+
- 'src/llama_stack/providers/inline/agents/**'
31+
- 'src/llama_stack/apis/agents/**'
32+
- 'tests/integration/openresponses/**'
33+
- '.github/workflows/openresponses-conformance.yml'
34+
workflow_dispatch:
35+
36+
concurrency:
37+
group: ${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_id || github.ref }}
38+
cancel-in-progress: true
39+
40+
permissions:
41+
contents: read
42+
43+
jobs:
44+
openresponses-conformance:
45+
name: OpenResponses Conformance (Informational)
46+
runs-on: ubuntu-latest
47+
# Failures are expected — this job does NOT block PRs or merges.
48+
# It exists solely to track progress toward OpenResponses API conformance.
49+
continue-on-error: true
50+
51+
env:
52+
# Dummy key satisfies config parsing; actual inference is replayed from recordings
53+
OPENAI_API_KEY: "dummy-key-for-replay-mode"
54+
INFERENCE_MODEL: "openai/gpt-4o-mini"
55+
LLAMA_STACK_PORT: 8321
56+
LLAMA_STACK_TEST_INFERENCE_MODE: "replay"
57+
LLAMA_STACK_TEST_RECORDING_DIR: ${{ github.workspace }}/tests/integration/openresponses
58+
59+
steps:
60+
- name: Checkout repository
61+
uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
62+
63+
- name: Check for recordings
64+
id: check-recordings
65+
run: |
66+
RECORDINGS_DIR="${{ github.workspace }}/tests/integration/openresponses/recordings"
67+
COUNT=$(find "$RECORDINGS_DIR" -name "*.json" 2>/dev/null | wc -l)
68+
echo "recordings_count=$COUNT" >> $GITHUB_OUTPUT
69+
if [ "$COUNT" -eq 0 ]; then
70+
echo "::warning::No recordings found in tests/integration/openresponses/recordings/."
71+
echo "::warning::Run the server locally with LLAMA_STACK_TEST_INFERENCE_MODE=record-if-missing"
72+
echo "::warning::and commit the resulting recordings to enable conformance testing in CI."
73+
else
74+
echo "Found $COUNT recording(s)"
75+
fi
76+
77+
- name: Install dependencies
78+
uses: ./.github/actions/setup-runner
79+
with:
80+
python-version: '3.12'
81+
82+
- name: Install ci-tests distro provider dependencies
83+
run: uv run llama stack list-deps ci-tests --format uv | sh
84+
85+
- name: Start Llama Stack server
86+
run: |
87+
mkdir -p /tmp/llama-stack-conformance
88+
export LLAMA_STACK_LOG_WIDTH=200
89+
nohup uv run llama stack run ci-tests --port $LLAMA_STACK_PORT \
90+
> /tmp/llama-stack-conformance/server.log 2>&1 &
91+
echo "Server PID: $!"
92+
93+
- name: Wait for Llama Stack server to be ready
94+
run: |
95+
echo "Waiting for Llama Stack server..."
96+
for i in {1..60}; do
97+
if curl -s http://localhost:$LLAMA_STACK_PORT/v1/health | grep -q "OK"; then
98+
echo "Llama Stack server is ready!"
99+
exit 0
100+
fi
101+
sleep 2
102+
done
103+
echo "Llama Stack server failed to start in 120 seconds"
104+
cat /tmp/llama-stack-conformance/server.log
105+
exit 1
106+
107+
- name: Setup Bun
108+
uses: oven-sh/setup-bun@3d267786b128fe76c2f16a390aa2448b815359f3 #v2.0.0
109+
110+
- name: Clone OpenResponses repository
111+
run: |
112+
git clone --depth=1 https://github.com/openresponses/openresponses.git /tmp/openresponses
113+
114+
- name: Install OpenResponses dependencies
115+
working-directory: /tmp/openresponses
116+
run: bun install
117+
118+
- name: Run OpenResponses conformance tests
119+
id: run-tests
120+
run: |
121+
cd /tmp/openresponses
122+
bun run bin/compliance-test.ts \
123+
--base-url "http://localhost:$LLAMA_STACK_PORT/v1" \
124+
--api-key "llama-stack" \
125+
--model "$INFERENCE_MODEL" \
126+
--json > /tmp/openresponses-results.json 2>/dev/null || true
127+
128+
echo "=== OpenResponses Conformance Test Output ==="
129+
bun run bin/compliance-test.ts \
130+
--base-url "http://localhost:$LLAMA_STACK_PORT/v1" \
131+
--api-key "llama-stack" \
132+
--model "$INFERENCE_MODEL" \
133+
--verbose 2>&1 | tee /tmp/openresponses-output.txt || true
134+
135+
- name: Generate conformance report
136+
if: always()
137+
run: |
138+
{
139+
echo "## OpenResponses Conformance Test Results"
140+
echo ""
141+
echo "> **Note:** These tests are **informational only** and track progress toward full"
142+
echo "> [OpenResponses](https://www.openresponses.org) API conformance."
143+
echo "> Failures are expected while gaps remain in the Responses API implementation."
144+
echo "> See [#4818](https://github.com/llamastack/llama-stack/issues/4818)."
145+
echo ""
146+
echo "**Model:** \`$INFERENCE_MODEL\`"
147+
echo "**Recordings:** ${{ steps.check-recordings.outputs.recordings_count }} file(s) in \`tests/integration/openresponses/recordings/\`"
148+
echo ""
149+
} >> $GITHUB_STEP_SUMMARY
150+
151+
if [ "${{ steps.check-recordings.outputs.recordings_count }}" -eq 0 ]; then
152+
{
153+
echo "### No Recordings Found"
154+
echo ""
155+
echo "To enable conformance testing, generate recordings locally:"
156+
echo '```bash'
157+
echo "OPENAI_API_KEY=\$YOUR_KEY bash scripts/record-openresponses-conformance.sh"
158+
echo ""
159+
echo "# Then commit the recordings"
160+
echo "git add tests/integration/openresponses/recordings/"
161+
echo "git commit -m 'chore: add OpenResponses conformance recordings'"
162+
echo '```'
163+
echo "Commit the resulting \`tests/integration/openresponses/recordings/*.json\` files."
164+
} >> $GITHUB_STEP_SUMMARY
165+
elif [ -f /tmp/openresponses-results.json ] && jq -e '.summary' /tmp/openresponses-results.json > /dev/null 2>&1; then
166+
PASSED=$(jq -r '.summary.passed' /tmp/openresponses-results.json)
167+
FAILED=$(jq -r '.summary.failed' /tmp/openresponses-results.json)
168+
TOTAL=$(jq -r '.summary.total' /tmp/openresponses-results.json)
169+
170+
{
171+
echo "### Summary"
172+
echo ""
173+
echo "| Metric | Count |"
174+
echo "|--------|-------|"
175+
echo "| ✅ Passed | $PASSED |"
176+
echo "| ❌ Failed | $FAILED |"
177+
echo "| **Total** | **$TOTAL** |"
178+
echo ""
179+
echo "### Test Details"
180+
echo ""
181+
echo "| Test | Status | Duration |"
182+
echo "|------|--------|----------|"
183+
jq -r '.results[] | "| \(.name) | \(if .status == "passed" then "✅ Pass" else "❌ Fail" end) | \(.duration // "—")ms |"' \
184+
/tmp/openresponses-results.json
185+
echo ""
186+
} >> $GITHUB_STEP_SUMMARY
187+
188+
if [ "$FAILED" -gt 0 ]; then
189+
{
190+
echo "### Failure Details"
191+
echo ""
192+
jq -r '.results[] | select(.status == "failed") | "**\(.name)**\n" + ((.errors // []) | map("- \(.)") | join("\n")) + "\n"' \
193+
/tmp/openresponses-results.json
194+
} >> $GITHUB_STEP_SUMMARY
195+
fi
196+
else
197+
{
198+
echo "### Results"
199+
echo ""
200+
echo "Could not parse structured results."
201+
if [ -f /tmp/openresponses-output.txt ]; then
202+
echo ""
203+
echo "<details><summary>Raw output</summary>"
204+
echo ""
205+
echo '```'
206+
cat /tmp/openresponses-output.txt
207+
echo '```'
208+
echo ""
209+
echo "</details>"
210+
fi
211+
} >> $GITHUB_STEP_SUMMARY
212+
fi
213+
214+
- name: Upload conformance test results
215+
if: always()
216+
uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
217+
with:
218+
name: openresponses-conformance-results-${{ github.run_id }}
219+
path: |
220+
/tmp/openresponses-results.json
221+
/tmp/openresponses-output.txt
222+
retention-days: 30
223+
if-no-files-found: warn
224+
225+
- name: Print server log on failure
226+
if: failure()
227+
run: |
228+
echo "=== Llama Stack server log (last 200 lines) ==="
229+
tail -200 /tmp/llama-stack-conformance/server.log || true

‎scripts/record-openresponses-conformance.sh‎

Lines changed: 126 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,126 @@
1+
#!/bin/bash
2+
# Copyright (c) Meta Platforms, Inc. and affiliates.
3+
# All rights reserved.
4+
#
5+
# This source code is licensed under the terms described in the LICENSE file in
6+
# the root directory of this source tree.
7+
8+
set -euo pipefail
9+
10+
# Records OpenResponses conformance test interactions against a local llama-stack
11+
# server so that CI can replay them without a live API key.
12+
#
13+
# Run this script whenever you add new compliance tests or the openresponses
14+
# test suite changes, then commit the resulting recordings:
15+
#
16+
# git add tests/integration/openresponses/recordings/
17+
# git commit -m "chore: update OpenResponses conformance recordings"
18+
#
19+
# Requirements:
20+
# - OPENAI_API_KEY must be set
21+
# - uv must be available (https://github.com/astral-sh/uv)
22+
# - bun will be installed automatically if missing
23+
24+
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
25+
REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
26+
RECORDING_DIR="$REPO_ROOT/tests/integration/openresponses"
27+
PORT="${PORT:-8321}"
28+
INFERENCE_MODEL="${INFERENCE_MODEL:-openai/gpt-4o-mini}"
29+
OPENRESPONSES_DIR="${OPENRESPONSES_DIR:-/tmp/openresponses}"
30+
LOG_FILE="/tmp/openresponses-server.log"
31+
32+
# ── Cleanup ────────────────────────────────────────────────────────────────────
33+
SERVER_PID=""
34+
cleanup() {
35+
if [[ -n "$SERVER_PID" ]]; then
36+
echo ""
37+
echo "Stopping llama-stack server (PID $SERVER_PID)..."
38+
kill "$SERVER_PID" 2>/dev/null || true
39+
fi
40+
}
41+
trap cleanup EXIT
42+
43+
# ── Preflight checks ───────────────────────────────────────────────────────────
44+
if [[ -z "${OPENAI_API_KEY:-}" ]]; then
45+
echo "Error: OPENAI_API_KEY must be set to record conformance test interactions."
46+
exit 1
47+
fi
48+
49+
cd "$REPO_ROOT"
50+
51+
# ── Bun ────────────────────────────────────────────────────────────────────────
52+
if ! command -v bun &>/dev/null; then
53+
echo "=== Installing bun ==="
54+
curl -fsSL https://bun.sh/install | bash
55+
export PATH="$HOME/.bun/bin:$PATH"
56+
fi
57+
58+
# ── OpenResponses CLI ──────────────────────────────────────────────────────────
59+
if [[ -d "$OPENRESPONSES_DIR/.git" ]]; then
60+
echo "=== Updating openresponses ==="
61+
git -C "$OPENRESPONSES_DIR" pull --ff-only
62+
else
63+
echo "=== Cloning openresponses ==="
64+
git clone --depth=1 https://github.com/openresponses/openresponses.git "$OPENRESPONSES_DIR"
65+
fi
66+
echo "=== Installing openresponses dependencies ==="
67+
(cd "$OPENRESPONSES_DIR" && bun install)
68+
69+
# ── llama-stack provider dependencies ─────────────────────────────────────────
70+
echo "=== Installing ci-tests distro dependencies ==="
71+
llama stack list-deps ci-tests --format uv | sh
72+
73+
# ── Start server ───────────────────────────────────────────────────────────────
74+
echo "=== Starting llama-stack server (record-if-missing) ==="
75+
mkdir -p "$(dirname "$LOG_FILE")"
76+
77+
LLAMA_STACK_TEST_INFERENCE_MODE=record-if-missing \
78+
LLAMA_STACK_TEST_RECORDING_DIR="$RECORDING_DIR" \
79+
LLAMA_STACK_LOG_WIDTH=200 \
80+
nohup llama stack run ci-tests --port "$PORT" \
81+
> "$LOG_FILE" 2>&1 &
82+
SERVER_PID=$!
83+
echo "Server PID: $SERVER_PID"
84+
85+
# ── Wait for health ────────────────────────────────────────────────────────────
86+
echo "Waiting for llama-stack server to be ready..."
87+
for i in {1..60}; do
88+
if curl -sf "http://localhost:$PORT/v1/health" 2>/dev/null | grep -q "OK"; then
89+
echo "Server is ready!"
90+
break
91+
fi
92+
if [[ $i -eq 60 ]]; then
93+
echo "Server failed to start within 120 seconds. Log:"
94+
cat "$LOG_FILE"
95+
exit 1
96+
fi
97+
sleep 2
98+
done
99+
100+
# ── Run compliance tests ───────────────────────────────────────────────────────
101+
echo ""
102+
echo "=== Running OpenResponses compliance tests ==="
103+
echo ""
104+
(
105+
cd "$OPENRESPONSES_DIR"
106+
bun run bin/compliance-test.ts \
107+
--base-url "http://localhost:$PORT/v1" \
108+
--api-key "llama-stack" \
109+
--model "$INFERENCE_MODEL" \
110+
--verbose
111+
) || true # continue-on-error: failures here are expected while the implementation has gaps
112+
113+
# ── Summary ────────────────────────────────────────────────────────────────────
114+
echo ""
115+
echo "=== Recordings written to: $RECORDING_DIR/recordings/ ==="
116+
RECORDING_COUNT=$(find "$RECORDING_DIR/recordings" -name "*.json" 2>/dev/null | wc -l | tr -d ' ')
117+
echo "Total recording files: $RECORDING_COUNT"
118+
echo ""
119+
if [[ "$RECORDING_COUNT" -gt 0 ]]; then
120+
echo "Commit the recordings to include them in CI:"
121+
echo ""
122+
echo " git add tests/integration/openresponses/recordings/"
123+
echo " git commit -m 'chore: add OpenResponses conformance recordings'"
124+
else
125+
echo "No recordings were created. Check $LOG_FILE for server errors."
126+
fi

0 commit comments

Comments
 (0)