fix(compose): forward full profile-b provider and extraction config i…

docker-compose.v4.yml

@@ -27,17 +27,25 @@ x-mesa-v4-runtime: &mesa-v4-runtime
     MESA_LOAD_DOTENV: "false"
     MESA_MODEL_ENABLED: ${MESA_MODEL_ENABLED:?MESA_MODEL_ENABLED must be explicitly enabled for v4}
     MESA_EXTERNAL_PROVIDER_ENABLED: ${MESA_EXTERNAL_PROVIDER_ENABLED:?MESA_EXTERNAL_PROVIDER_ENABLED must be explicit}
+    MESA_LLM_PROVIDER: ${MESA_LLM_PROVIDER:-openai_compatible}
     LLM_BASE_URL: ${LLM_BASE_URL:-}
     LLM_API_KEY: ${LLM_API_KEY:-}
     LLM_MODEL_NAME: ${LLM_MODEL_NAME:-llama-3.1-8b-instant}
+    LLM_TIMEOUT_SECONDS: ${LLM_TIMEOUT_SECONDS:-20}
     MESA_LOCAL_EMBEDDING_MODEL: ${MESA_LOCAL_EMBEDDING_MODEL:-magibu/embeddingmagibu-200m}
     MESA_EMBEDDING_DIMENSION: ${MESA_EMBEDDING_DIMENSION:-768}
+    MESA_EMBEDDING_VERSION: ${MESA_EMBEDDING_VERSION:-v1}
+    MESA_EMBEDDING_MODEL_REVISION: ${MESA_EMBEDDING_MODEL_REVISION:-}
     MESA_EMBEDDING_NORMALIZED: ${MESA_EMBEDDING_NORMALIZED:-true}
     MESA_EMBEDDING_PROVIDER: ${MESA_EMBEDDING_PROVIDER:-openai_compatible}
     MESA_EXTERNAL_EMBEDDING_MODEL: ${MESA_EXTERNAL_EMBEDDING_MODEL:-text-embedding-3-small}
+    MESA_EMBEDDING_BASE_URL: ${MESA_EMBEDDING_BASE_URL:-}
+    MESA_EMBEDDING_API_KEY: ${MESA_EMBEDDING_API_KEY:-}
     MESA_EXTRACTION_PROVIDER: ${MESA_EXTRACTION_PROVIDER:-ollama}
     MESA_EXTRACTION_MODEL: ${MESA_EXTRACTION_MODEL:-qwen3:1.7b}
     MESA_EXTRACTION_THINKING: ${MESA_EXTRACTION_THINKING:-false}
+    MESA_EXTRACTION_MAX_TOKENS: ${MESA_EXTRACTION_MAX_TOKENS:-4096}
+    MESA_EXTRACTION_LANG: ${MESA_EXTRACTION_LANG:-tr}
     MESA_TIER3_MODE: ${MESA_TIER3_MODE:-}
     MESA_TIER3_LLM_PROVIDER_A: ${MESA_TIER3_LLM_PROVIDER_A:-}
     MESA_TIER3_LLM_MODEL_A: ${MESA_TIER3_LLM_MODEL_A:-}

tests/test_deployment_assets.py

@@ -122,13 +122,26 @@ def test_full_cognitive_compose_forwards_provider_and_tier3_contract() -> None:
     )
     environment = compose["services"]["mesa-v4"]["environment"]
     assert {
+        "MESA_LLM_PROVIDER",
         "LLM_BASE_URL",
         "LLM_API_KEY",
         "LLM_MODEL_NAME",
+        "LLM_TIMEOUT_SECONDS",
+        "MESA_EXTERNAL_PROVIDER_ENABLED",
         "MESA_LOCAL_EMBEDDING_MODEL",
         "MESA_EMBEDDING_PROVIDER",
+        "MESA_EXTERNAL_EMBEDDING_MODEL",
+        "MESA_EMBEDDING_BASE_URL",
+        "MESA_EMBEDDING_API_KEY",
         "MESA_EMBEDDING_DIMENSION",
+        "MESA_EMBEDDING_VERSION",
+        "MESA_EMBEDDING_MODEL_REVISION",
         "MESA_EMBEDDING_NORMALIZED",
+        "MESA_EXTRACTION_PROVIDER",
+        "MESA_EXTRACTION_MODEL",
+        "MESA_EXTRACTION_THINKING",
+        "MESA_EXTRACTION_MAX_TOKENS",
+        "MESA_EXTRACTION_LANG",
         "MESA_TIER3_LLM_PROVIDER_A",
         "MESA_TIER3_LLM_MODEL_A",
         "MESA_TIER3_LLM_PROVIDER_B",
@@ -141,6 +154,109 @@ def test_full_cognitive_compose_forwards_provider_and_tier3_contract() -> None:
     assert '"${wheel}[ml,adapters]"' in dockerfile
 
 
+def test_profile_b_rendered_compose_environment_parity() -> None:
+    import os
+    import shutil
+    import subprocess
+
+    import pytest
+
+    from mesa_memory.config import MesaConfig
+
+    if not shutil.which("docker"):
+        pytest.skip("docker binary unavailable in this environment")
+
+    env_overrides = {
+        "MESA_API_KEY": "test-mesa-api-key",
+        "MESA_PRINCIPAL_ID": "test-principal",
+        "MESA_MODEL_ENABLED": "true",
+        "MESA_EXTERNAL_PROVIDER_ENABLED": "true",
+        "MESA_LLM_PROVIDER": "openai_compatible",
+        "LLM_BASE_URL": "https://integrate.api.nvidia.com/v1",
+        "LLM_MODEL_NAME": "openai/gpt-oss-20b",
+        "LLM_TIMEOUT_SECONDS": "35",
+        "LLM_API_KEY": "test-llm-secret-sentinel",
+        "MESA_EMBEDDING_PROVIDER": "openai_compatible",
+        "MESA_EXTERNAL_EMBEDDING_MODEL": "nvidia/nemotron-3-embed-1b",
+        "MESA_EMBEDDING_DIMENSION": "2048",
+        "MESA_EMBEDDING_VERSION": "nemotron-qpass-v1",
+        "MESA_EMBEDDING_NORMALIZED": "true",
+        "MESA_EMBEDDING_BASE_URL": "https://integrate.api.nvidia.com/v1",
+        "MESA_EMBEDDING_API_KEY": "test-embed-secret-sentinel",
+        "MESA_EMBEDDING_MODEL_REVISION": "nemotron-rev-1",
+        "MESA_EXTRACTION_PROVIDER": "openai_compatible",
+        "MESA_EXTRACTION_MODEL": "openai/gpt-oss-20b",
+        "MESA_EXTRACTION_LANG": "tr",
+        "MESA_EXTRACTION_THINKING": "false",
+        "MESA_EXTRACTION_MAX_TOKENS": "4096",
+    }
+
+    cmd = ["docker", "compose", "-f", str(ROOT / "docker-compose.v4.yml"), "config"]
+    res = subprocess.run(
+        cmd,
+        cwd=str(ROOT),
+        env={**os.environ, **env_overrides},
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    rendered = yaml.safe_load(res.stdout)
+    container_env = rendered["services"]["mesa-v4"]["environment"]
+
+    assert container_env["MESA_LLM_PROVIDER"] == "openai_compatible"
+    assert container_env["LLM_BASE_URL"] == "https://integrate.api.nvidia.com/v1"
+    assert container_env["LLM_MODEL_NAME"] == "openai/gpt-oss-20b"
+    assert container_env["LLM_TIMEOUT_SECONDS"] == "35"
+    assert container_env["LLM_API_KEY"] == "test-llm-secret-sentinel"
+
+    assert container_env["MESA_EMBEDDING_PROVIDER"] == "openai_compatible"
+    assert (
+        container_env["MESA_EXTERNAL_EMBEDDING_MODEL"]
+        == "nvidia/nemotron-3-embed-1b"
+    )
+    assert container_env["MESA_EMBEDDING_DIMENSION"] == "2048"
+    assert container_env["MESA_EMBEDDING_VERSION"] == "nemotron-qpass-v1"
+    assert container_env["MESA_EMBEDDING_NORMALIZED"] == "true"
+    assert (
+        container_env["MESA_EMBEDDING_BASE_URL"]
+        == "https://integrate.api.nvidia.com/v1"
+    )
+    assert container_env["MESA_EMBEDDING_API_KEY"] == "test-embed-secret-sentinel"
+    assert container_env["MESA_EMBEDDING_MODEL_REVISION"] == "nemotron-rev-1"
+
+    assert container_env["MESA_EXTRACTION_PROVIDER"] == "openai_compatible"
+    assert container_env["MESA_EXTRACTION_MODEL"] == "openai/gpt-oss-20b"
+    assert container_env["MESA_EXTRACTION_LANG"] == "tr"
+    assert container_env["MESA_EXTRACTION_THINKING"] == "false"
+    assert container_env["MESA_EXTRACTION_MAX_TOKENS"] == "4096"
+
+    # Container-inside config validation smoke
+    parsed_config = MesaConfig(**container_env)
+    assert parsed_config.mesa_llm_provider == "openai_compatible"
+    assert parsed_config.llm_base_url == "https://integrate.api.nvidia.com/v1"
+    assert parsed_config.llm_model_name == "openai/gpt-oss-20b"
+    assert parsed_config.llm_timeout_seconds == 35.0
+    assert parsed_config.llm_api_key == "test-llm-secret-sentinel"
+    assert parsed_config.embedding_provider == "openai_compatible"
+    assert (
+        parsed_config.external_embedding_model == "nvidia/nemotron-3-embed-1b"
+    )
+    assert parsed_config.embedding_dimension == 2048
+    assert parsed_config.embedding_version == "nemotron-qpass-v1"
+    assert parsed_config.embedding_normalized is True
+    assert (
+        parsed_config.embedding_base_url
+        == "https://integrate.api.nvidia.com/v1"
+    )
+    assert parsed_config.embedding_api_key == "test-embed-secret-sentinel"
+    assert parsed_config.embedding_model_revision == "nemotron-rev-1"
+    assert parsed_config.extraction_provider == "openai_compatible"
+    assert parsed_config.extraction_model == "openai/gpt-oss-20b"
+    assert parsed_config.extraction_lang == "tr"
+    assert parsed_config.extraction_thinking is False
+    assert parsed_config.extraction_max_tokens == 4096
+
+
 def test_dockerfile_uses_exact_base_nonroot_health_and_bounded_entrypoint() -> None:
     dockerfile = (ROOT / "Dockerfile").read_text(encoding="utf-8")
     assert (

tests/test_profile_b_compose_env_parity.py

@@ -0,0 +1,181 @@
+"""Regression test suite for Profile B Docker Compose environment parity.
+
+Ensures all provider, model, embedding, and extraction runtime settings
+are forwarded from host to container environment without schema or contract drift.
+"""
+
+from __future__ import annotations
+
+import os
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+import yaml
+
+from mesa_memory.config import MesaConfig
+
+ROOT = Path(__file__).parents[1]
+COMPOSE_V4_PATH = ROOT / "docker-compose.v4.yml"
+
+REQUIRED_PROFILE_B_ENV_KEYS = {
+    "MESA_LLM_PROVIDER",
+    "LLM_BASE_URL",
+    "LLM_API_KEY",
+    "LLM_MODEL_NAME",
+    "LLM_TIMEOUT_SECONDS",
+    "MESA_EXTERNAL_PROVIDER_ENABLED",
+    "MESA_EMBEDDING_PROVIDER",
+    "MESA_EXTERNAL_EMBEDDING_MODEL",
+    "MESA_EMBEDDING_BASE_URL",
+    "MESA_EMBEDDING_API_KEY",
+    "MESA_EMBEDDING_DIMENSION",
+    "MESA_EMBEDDING_VERSION",
+    "MESA_EMBEDDING_MODEL_REVISION",
+    "MESA_EMBEDDING_NORMALIZED",
+    "MESA_EXTRACTION_PROVIDER",
+    "MESA_EXTRACTION_MODEL",
+    "MESA_EXTRACTION_THINKING",
+    "MESA_EXTRACTION_MAX_TOKENS",
+    "MESA_EXTRACTION_LANG",
+}
+
+
+def test_static_compose_v4_contains_all_profile_b_env_keys() -> None:
+    """Verify statically that docker-compose.v4.yml defines all Profile B runtime keys."""
+    raw_compose = yaml.safe_load(COMPOSE_V4_PATH.read_text(encoding="utf-8"))
+    runtime_env = raw_compose["x-mesa-v4-runtime"]["environment"]
+    service_env = raw_compose["services"]["mesa-v4"]["environment"]
+
+    missing_runtime = REQUIRED_PROFILE_B_ENV_KEYS - set(runtime_env.keys())
+    assert not missing_runtime, f"Missing in x-mesa-v4-runtime: {missing_runtime}"
+
+    missing_service = REQUIRED_PROFILE_B_ENV_KEYS - set(service_env.keys())
+    assert not missing_service, f"Missing in mesa-v4 service environment: {missing_service}"
+
+
+def test_effective_compose_config_forwards_profile_b_environment() -> None:
+    """Verify that docker compose config renders synthetic Profile B env vars identically."""
+    if not shutil.which("docker"):
+        pytest.skip("docker binary unavailable in this environment")
+
+    synthetic_env = {
+        "MESA_API_KEY": "test-mesa-api-key",
+        "MESA_PRINCIPAL_ID": "test-principal-id",
+        "MESA_MODEL_ENABLED": "true",
+        "MESA_EXTERNAL_PROVIDER_ENABLED": "true",
+        "MESA_LLM_PROVIDER": "openai_compatible",
+        "LLM_BASE_URL": "https://integrate.api.nvidia.com/v1",
+        "LLM_MODEL_NAME": "openai/gpt-oss-20b",
+        "LLM_TIMEOUT_SECONDS": "25",
+        "LLM_API_KEY": "test-llm-secret-sentinel",
+        "MESA_EMBEDDING_PROVIDER": "openai_compatible",
+        "MESA_EXTERNAL_EMBEDDING_MODEL": "nvidia/nemotron-3-embed-1b",
+        "MESA_EMBEDDING_DIMENSION": "2048",
+        "MESA_EMBEDDING_VERSION": "nemotron-qpass-v1",
+        "MESA_EMBEDDING_NORMALIZED": "true",
+        "MESA_EMBEDDING_BASE_URL": "https://integrate.api.nvidia.com/v1",
+        "MESA_EMBEDDING_API_KEY": "test-embed-secret-sentinel",
+        "MESA_EMBEDDING_MODEL_REVISION": "nemotron-rev-42",
+        "MESA_EXTRACTION_PROVIDER": "openai_compatible",
+        "MESA_EXTRACTION_MODEL": "openai/gpt-oss-20b",
+        "MESA_EXTRACTION_LANG": "tr",
+        "MESA_EXTRACTION_THINKING": "false",
+        "MESA_EXTRACTION_MAX_TOKENS": "4096",
+    }
+
+    cmd = ["docker", "compose", "-f", str(COMPOSE_V4_PATH), "config"]
+    res = subprocess.run(
+        cmd,
+        cwd=str(ROOT),
+        env={**os.environ, **synthetic_env},
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    rendered = yaml.safe_load(res.stdout)
+    container_env = rendered["services"]["mesa-v4"]["environment"]
+
+    # Verify every synthetic Profile B key forwarded to container
+    assert container_env["MESA_LLM_PROVIDER"] == "openai_compatible"
+    assert container_env["LLM_BASE_URL"] == "https://integrate.api.nvidia.com/v1"
+    assert container_env["LLM_MODEL_NAME"] == "openai/gpt-oss-20b"
+    assert container_env["LLM_TIMEOUT_SECONDS"] == "25"
+    assert container_env["LLM_API_KEY"] == "test-llm-secret-sentinel"
+
+    assert container_env["MESA_EMBEDDING_PROVIDER"] == "openai_compatible"
+    assert container_env["MESA_EXTERNAL_EMBEDDING_MODEL"] == "nvidia/nemotron-3-embed-1b"
+    assert container_env["MESA_EMBEDDING_DIMENSION"] == "2048"
+    assert container_env["MESA_EMBEDDING_VERSION"] == "nemotron-qpass-v1"
+    assert container_env["MESA_EMBEDDING_NORMALIZED"] == "true"
+    assert container_env["MESA_EMBEDDING_BASE_URL"] == "https://integrate.api.nvidia.com/v1"
+    assert container_env["MESA_EMBEDDING_API_KEY"] == "test-embed-secret-sentinel"
+    assert container_env["MESA_EMBEDDING_MODEL_REVISION"] == "nemotron-rev-42"
+
+    assert container_env["MESA_EXTRACTION_PROVIDER"] == "openai_compatible"
+    assert container_env["MESA_EXTRACTION_MODEL"] == "openai/gpt-oss-20b"
+    assert container_env["MESA_EXTRACTION_LANG"] == "tr"
+    assert container_env["MESA_EXTRACTION_THINKING"] == "false"
+    assert container_env["MESA_EXTRACTION_MAX_TOKENS"] == "4096"
+
+    # Container-inside config parsing smoke test
+    cfg = MesaConfig(**container_env)
+    assert cfg.mesa_llm_provider == "openai_compatible"
+    assert cfg.llm_base_url == "https://integrate.api.nvidia.com/v1"
+    assert cfg.llm_model_name == "openai/gpt-oss-20b"
+    assert cfg.llm_timeout_seconds == 25.0
+    assert cfg.llm_api_key == "test-llm-secret-sentinel"
+    assert cfg.embedding_provider == "openai_compatible"
+    assert cfg.external_embedding_model == "nvidia/nemotron-3-embed-1b"
+    assert cfg.embedding_dimension == 2048
+    assert cfg.embedding_version == "nemotron-qpass-v1"
+    assert cfg.embedding_normalized is True
+    assert cfg.embedding_base_url == "https://integrate.api.nvidia.com/v1"
+    assert cfg.embedding_api_key == "test-embed-secret-sentinel"
+    assert cfg.embedding_model_revision == "nemotron-rev-42"
+    assert cfg.extraction_provider == "openai_compatible"
+    assert cfg.extraction_model == "openai/gpt-oss-20b"
+    assert cfg.extraction_lang == "tr"
+    assert cfg.extraction_thinking is False
+    assert cfg.extraction_max_tokens == 4096
+
+
+def test_effective_compose_config_defaults_match_mesaconfig_contract() -> None:
+    """Verify that when optional host env vars are omitted, rendered defaults align with MesaConfig."""
+    if not shutil.which("docker"):
+        pytest.skip("docker binary unavailable in this environment")
+
+    minimal_env = {
+        "MESA_API_KEY": "test-mesa-api-key",
+        "MESA_PRINCIPAL_ID": "test-principal-id",
+        "MESA_MODEL_ENABLED": "true",
+        "MESA_EXTERNAL_PROVIDER_ENABLED": "false",
+    }
+
+    cmd = ["docker", "compose", "-f", str(COMPOSE_V4_PATH), "config"]
+    res = subprocess.run(
+        cmd,
+        cwd=str(ROOT),
+        env={**os.environ, **minimal_env},
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    rendered = yaml.safe_load(res.stdout)
+    container_env = rendered["services"]["mesa-v4"]["environment"]
+
+    # Verify defaults
+    assert container_env["MESA_LLM_PROVIDER"] == "openai_compatible"
+    assert container_env["LLM_TIMEOUT_SECONDS"] == "20"
+    assert container_env["MESA_EMBEDDING_VERSION"] == "v1"
+    assert container_env["MESA_EXTRACTION_MAX_TOKENS"] == "4096"
+    assert container_env["MESA_EXTRACTION_LANG"] == "tr"
+
+    # Container-inside config smoke
+    cfg = MesaConfig(**container_env)
+    assert cfg.mesa_llm_provider == "openai_compatible"
+    assert cfg.llm_timeout_seconds == 20.0
+    assert cfg.embedding_version == "v1"
+    assert cfg.extraction_max_tokens == 4096
+    assert cfg.extraction_lang == "tr"