[bugfix] Fix test env defaults and server dependencies (#1901)

The distributed_setup test fixture now sets MASTER_ADDR and MASTER_PORT when
they are not set, with a new port for each test, and restores them after the
test. pyproject.toml now declares orjson and python-multipart, which the
OpenAI-compatible server needs. Two documents about the SSIM reference
bootstrap are corrected.

docs/contributing/testing.md

@@ -233,10 +233,11 @@ test fails. For new-model PRs, CI can run SSIM in bootstrap mode so missing
 references are uploaded as draft artifacts for review instead of immediately
 blocking on a missing canonical reference.
 
-Buildkite enables SSIM bootstrap mode when either condition is true:
-
-- the PR title or Buildkite message contains `[new-model]`;
-- `FASTVIDEO_SSIM_BOOTSTRAP_MODE=1` is set for the Buildkite job.
+The active Slurm SSIM lane (`.buildkite/scripts/lanes/ssim.sh`) enables
+bootstrap mode only when `FASTVIDEO_SSIM_BOOTSTRAP_MODE=1` is set for the
+Buildkite job. A `[new-model]` PR title does not enable it on this path (only
+the retired Modal script `.buildkite/scripts/pr_test.sh` checked the title), so
+ask a maintainer to start the SSIM job with that variable set.
 
 Bootstrap mode passes `--ssim-bootstrap-mode` to pytest. When a generated
 artifact is available, the test uploads it under the `drafts/...` namespace in

fastvideo/tests/conftest.py

@@ -1,12 +1,15 @@
 # SPDX-License-Identifier: Apache-2.0
 import contextlib
+import os
 
 import pytest
 
 import torch
 import numpy as np
 
+from fastvideo import envs
 from fastvideo.distributed import (maybe_init_distributed_environment_and_model_parallel, cleanup_dist_env_and_memory)
+from fastvideo.utils import get_open_port
 
 
 @pytest.fixture(scope="function")
@@ -18,10 +21,18 @@ def distributed_setup():
     """
     torch.manual_seed(42)
     np.random.seed(42)
-    maybe_init_distributed_environment_and_model_parallel(1, 1)
-    yield
-
-    cleanup_dist_env_and_memory()
+    with contextlib.ExitStack() as stack:
+        # The env:// rendezvous needs these outside the CI runner. Keep values the
+        # launcher provides; otherwise use a fresh port per test and restore on exit.
+        if os.environ.get("MASTER_ADDR") is None:
+            stack.enter_context(envs.override_external("MASTER_ADDR", "127.0.0.1"))
+        if os.environ.get("MASTER_PORT") is None:
+            stack.enter_context(envs.override_external("MASTER_PORT", str(get_open_port())))
+        maybe_init_distributed_environment_and_model_parallel(1, 1)
+        try:
+            yield
+        finally:
+            cleanup_dist_env_and_memory()
 
 
 @pytest.fixture

fastvideo/tests/ssim/README.md

@@ -89,8 +89,9 @@ into one subprocess per model id by setting `FASTVIDEO_SSIM_MODEL_ID`.
 ## Bootstrapping New References
 
 Normal SSIM runs are strict: missing references fail the test. For a new model
-PR, add `[new-model]` to the PR title or set `FASTVIDEO_SSIM_BOOTSTRAP_MODE=1`
-for the Buildkite job. CI then passes `--ssim-bootstrap-mode` to pytest.
+PR, a maintainer sets `FASTVIDEO_SSIM_BOOTSTRAP_MODE=1` for the Buildkite SSIM
+job (the Slurm lane does not read the PR title). CI then passes
+`--ssim-bootstrap-mode` to pytest.
 
 In bootstrap mode, a missing pixel `.mp4` or latent `.pt` reference is treated
 as an expected draft-reference case after the generated artifact has been

pyproject.toml

@@ -105,6 +105,8 @@ dependencies = [
 
     # Job Runner
     "fastapi>=0.129.0",
+    "orjson>=3.9.0",  # fastapi ORJSONResponse in the OpenAI-compatible server
+    "python-multipart>=0.0.18",  # fastapi Form/UploadFile parsing
     "uvicorn>=0.41.0"
 ]