Merge pull request #151 from lsst-dm/tickets/DM-48081

DM-48081 : Refactor Server Daemon

.dockerignore

@@ -4,5 +4,7 @@
 # Except
 !pyproject.toml
 !uv.lock
+!alembic.ini
 !src/lsst/
 !examples/
+!alembic/

.github/workflows/ci.yaml

@@ -1,5 +1,6 @@
 name: "CI"
-"on":
+
+on:
   push:
   pull_request:
     branches:
@@ -42,7 +43,7 @@ jobs:
           python-version: ${{ matrix.python-version }}
 
       - name: Install packages for testing
-        run: uv sync --dev
+        run: uv sync --dev --frozen
 
       - name: Run tests
         run: |
@@ -66,39 +67,8 @@ jobs:
           python-version: "3.11"
 
       - name: Install packages for testing
-        run: uv sync --dev
+        run: uv sync --dev --frozen
 
       - name: Run tests
         run: |
           uv run make typing
-
-  build:
-    runs-on: ubuntu-latest
-    needs: [test]
-    timeout-minutes: 20
-
-    # Only do Docker builds of tagged releases and pull requests from ticket
-    # branches. This will still trigger on pull requests from untrusted
-    # repositories whose branch names match our tickets/* branch convention,
-    # but in this case the build will fail with an error since the secret
-    # won't be set.
-    if: >
-      github.event_name != 'merge_group'
-      && (startsWith(github.ref, 'refs/tags/')
-          || startsWith(github.head_ref, 'tickets/'))
-
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - uses: lsst-sqre/build-and-push-to-ghcr@v1
-        id: build-service
-        with:
-          dockerfile: docker/Dockerfile
-          image: ${{ github.repository }}
-          github_token: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Report result
-        run: |
-          echo Pushed ghcr.io/${{ github.repository }}:${{ steps.build-service.outputs.tag }}

.gitignore

@@ -50,3 +50,4 @@ test_cm.db
 
 build/
 bruno/
+prod_area/

Makefile

@@ -98,7 +98,7 @@ psql: run-compose
 test: PGPORT=$(shell docker compose port postgresql 5432 | cut -d: -f2)
 test: export DB__URL=postgresql://cm-service@localhost:${PGPORT}/cm-service
 test: export DB__PASSWORD=INSECURE-PASSWORD
-test: export DB__SCHEMA=cm_service_test
+test: export DB__TABLE_SCHEMA=cm_service_test
 test: run-compose
 	alembic upgrade head
 	pytest -vvv --asyncio-mode=auto --cov=lsst.cmservice --cov-branch --cov-report=term --cov-report=html ${PYTEST_ARGS}
@@ -110,7 +110,7 @@ run: export DB__PASSWORD=INSECURE-PASSWORD
 run: export DB__ECHO=true
 run: run-compose
 	alembic upgrade head
-	python3 -m lsst.cmservice.cli.server run
+	python3 -m lsst.cmservice.main
 
 .PHONY: run-worker
 run-worker: PGPORT=$(shell docker compose port postgresql 5432 | cut -d: -f2)
@@ -148,7 +148,7 @@ run-sqlite: export DB__URL=sqlite+aiosqlite://///test_cm.db
 run-sqlite: export DB__ECHO=true
 run-sqlite:
 	alembic -x cm_database_url=sqlite:///test_cm.db upgrade head
-	python3 -m lsst.cmservice.cli.server run
+	python3 -m lsst.cmservice.main
 
 .PHONY: run-worker-sqlite
 run-worker-sqlite: export DB__URL=sqlite+aiosqlite://///test_cm.db
@@ -191,7 +191,7 @@ run-usdf-dev: export DB__PASSWORD=$(shell kubectl --cluster=usdf-cm-dev -n cm-se
 run-usdf-dev: export DB__ECHO=true
 run-usdf-dev:
 	alembic upgrade head
-	python3 -m lsst.cmservice.cli.server run
+	python3 -m lsst.cmservice.main
 
 .PHONY: run-worker-usdf-dev
 run-worker-usdf-dev: DB__HOST=$(shell kubectl --cluster=usdf-cm-dev -n cm-service get svc/cm-pg-lb -o jsonpath='{..ingress[0].ip}')

docker-compose.yaml

@@ -11,17 +11,20 @@ services:
     build:
       context: .
       dockerfile: docker/Dockerfile
+      target: cmservice
     entrypoint:
-      - python3
-      - -m
+      - alembic
     command:
-      - lsst.cmservice.cli.server
-      - init
+      - upgrade
+      - head
     environment: &cmenv
       DB__HOST: postgresql
       DB__ECHO: true
       DB__URL: postgresql://cm-service@postgresql:5432/cm-service
       DB__PASSWORD: INSECURE-PASSWORD
+      DB__TABLE_SCHEMA: public
+    networks:
+      - cmservice
     depends_on:
       postgresql:
         condition: service_healthy
@@ -33,17 +36,15 @@ services:
     build:
       context: .
       dockerfile: docker/Dockerfile
-    entrypoint:
-      - uvicorn
-    command:
-      - lsst.cmservice.main:app
-      - --host
-      - "0.0.0.0"
-      - --port
-      - "8080"
+      target: cmservice
+    env_file:
+      - path: .env
+        required: false
     environment: *cmenv
     ports:
       - "8080:8080"
+    networks:
+      - cmservice
     depends_on:
       init-db:
         condition: service_completed_successfully
@@ -55,12 +56,15 @@ services:
     build:
       context: .
       dockerfile: docker/Dockerfile
-    entrypoint:
-      - /opt/venv/bin/python3
-      - -m
-    command:
-      - lsst.cmservice.daemon
+      target: cmworker
+    env_file:
+      - path: .env
+        required: false
     environment: *cmenv
+    volumes:
+      - "./prod_area:/prod_area"
+    networks:
+      - cmservice
     depends_on:
       init-db:
         condition: service_completed_successfully
@@ -73,7 +77,9 @@ services:
       POSTGRES_USER: "cm-service"
       POSTGRES_DB: "cm-service"
     ports:
-      - "5432"
+      - "65432:5432"
+    networks:
+      - cmservice
     volumes:
       - "pgsql:/var/lib/postgresql/data"
     healthcheck:
@@ -87,3 +93,7 @@ services:
 
 volumes:
   pgsql:
+
+networks:
+  cmservice:
+    driver: bridge

docker/Dockerfile

@@ -1,16 +1,29 @@
 # syntax=docker/dockerfile:1
 
-ARG PYTHON_VERSION=3.11
-ARG UV_VERSION=0.5
+ARG PYTHON_VERSION="3.11"
+ARG UV_VERSION="0.5"
+ARG ASGI_PORT="8080"
 
 #==============================================================================
 # UV SOURCE IMAGE
 FROM ghcr.io/astral-sh/uv:${UV_VERSION} AS uv
 
 
+#==============================================================================
+# HTCONDOR SOURCE IMAGE
+# - The easiest way to obtain an architecture-appropriate set of htcondor client
+#   binaries is to install them into an otherwise bare conda environment.
+FROM continuumio/miniconda3:latest AS htcondor
+RUN conda create -p /opt/htcondor -c conda-forge htcondor-utils --no-default-packages
+
+
 #==============================================================================
 # BASE IMAGE
+# - The base image is an updated debian python container with a nonroot user
+#   created. This can be extended into include any *runtime* system dependencies
+#   during the apt-get step.
 FROM python:${PYTHON_VERSION}-slim-bookworm AS base-image
+ARG ASGI_PORT
 
 # Upgrade base packages
 RUN <<ENDRUN
@@ -34,6 +47,9 @@ groupadd --gid 3967 lsstsvc1
 useradd lsstsvc1 --uid 17951 --no-user-group --gid gu --groups rubin-users,lsst,lsstsvc1 --create-home --shell /bin/bash
 ENDRUN
 
+# Expose the API port
+EXPOSE ${ASGI_PORT}
+
 
 #==============================================================================
 # BUILDER IMAGE
@@ -69,11 +85,8 @@ ENDRUN
 
 
 #==============================================================================
-# RUNTIME IMAGE
-FROM base-image AS runtime-image
-
-# Expose the API port
-EXPOSE 8080
+# RUNTIME IMAGE - CM-Service
+FROM base-image AS cmservice
 
 # Make sure frontend output won't go into the layer store
 VOLUME /output
@@ -91,19 +104,19 @@ WORKDIR /home/lsstsvc1
 ENV CM_CONFIGS=/home/lsstsvc1/examples
 COPY examples ./examples
 COPY src/lsst ./lsst
+COPY alembic ./alembic
+COPY alembic.ini .
 
-# Run the Server
-# TODO refactor module main to start unicorn instead of using a bash script
-#      after eliminating db migration requirement here
-COPY --chmod=755 <<EOF ./docker-entrypoint.sh
-#!/bin/bash
-python -m lsst.cmservice.cli.server init
-uvicorn "\$@"
-EOF
-
-# Both the entrypoint and cmd can be disambiguated in the Helm chart / Compose file
-ENTRYPOINT ["./docker-entrypoint.sh"]
-CMD ["lsst.cmservice.main:app", "--host", "0.0.0.0", "--port", "8080"]
-
-# ENTRYPOINT ["/opt/venv/bin/python3", "-m"]
-# CMD ["lsst.cmservice.daemon"]
+ENTRYPOINT ["/opt/venv/bin/python3", "-m"]
+CMD ["lsst.cmservice.main"]
+
+
+#==============================================================================
+# RUNTIME IMAGE - Daemon
+FROM cmservice AS cmworker
+
+USER ROOT
+COPY --from=htcondor /opt/htcondor /opt/htcondor
+
+USER lsstsvc1
+CMD ["lsst.cmservice.daemon"]

pyproject.toml

@@ -22,6 +22,7 @@ dynamic = ["version"]
 
 dependencies = [
     "alembic==1.14.*",
+    "anyio==4.7.*",
     "asyncpg==0.30.*",
     "click==8.1.*",
     "fastapi==0.115.*",
@@ -37,9 +38,8 @@ dependencies = [
     "pause==0.3.*",
     "psycopg2-binary==2.9.*",
     "pydantic==2.10.*",
-    "pydantic-settings==2.6.*",
+    "pydantic-settings==2.7.*",
     "python-multipart==0.0.*",
-    "starlette==0.41.*",
     "structlog==24.4.*",
     "tabulate==0.9.*",
     "sqlalchemy[asyncio]==2.0.*",
@@ -61,7 +61,7 @@ dev = [
     "pytest-cov>=6.0.0",
     "pytest-playwright>=0.5.2",
     "pytest-timeout>=2.3.1",
-    "ruff>=0.7.4",
+    "ruff>=0.8.0",
     "python-semantic-release==9.14.*",
     "sqlalchemy[mypy]>=2.0.36",
     "types-pyyaml>=6.0.12.20240917",

src/lsst/cmservice/cli/campaign.py

@@ -134,8 +134,6 @@ def action() -> None:
 
 action_retry_script = wrappers.get_element_retry_script_command(action_command, sub_client)
 
-get_sleep_time = wrappers.get_element_estimate_sleep_time_command(action_command, sub_client)
-
 get_wms_task_reports = wrappers.get_element_wms_task_reports_command(get_command, sub_client)
 
 get_tasks = wrappers.get_element_tasks_command(get_command, sub_client)

src/lsst/cmservice/cli/group.py

@@ -131,8 +131,6 @@ def action() -> None:
 
 action_retry_script = wrappers.get_element_retry_script_command(action_command, sub_client)
 
-get_sleep_time = wrappers.get_element_estimate_sleep_time_command(action_command, sub_client)
-
 get_wms_task_reports = wrappers.get_element_wms_task_reports_command(get_command, sub_client)
 
 get_tasks = wrappers.get_element_tasks_command(get_command, sub_client)

src/lsst/cmservice/cli/job.py

@@ -129,8 +129,6 @@ def action() -> None:
 
 action_retry_script = wrappers.get_element_retry_script_command(action_command, sub_client)
 
-get_sleep_time = wrappers.get_element_estimate_sleep_time_command(action_command, sub_client)
-
 get_wms_task_reports = wrappers.get_element_wms_task_reports_command(get_command, sub_client)
 
 get_tasks = wrappers.get_element_tasks_command(get_command, sub_client)

src/lsst/cmservice/cli/server.py

@@ -1,5 +1,6 @@
+import warnings
+
 import click
-import uvicorn
 from safir.asyncio import run_with_asyncio
 
 from .. import __version__
@@ -9,7 +10,11 @@
 @click.group()
 @click.version_option(version=__version__)
 def server() -> None:
-    """Administrative command-line interface for cm-service."""
+    """Administrative command-line interface for cm-service.
+
+    .. deprecated:: v0.2.0
+        The `server` command is deprecated in v0.2.0
+    """
 
 
 @server.command(deprecated=True)
@@ -18,17 +23,22 @@ def server() -> None:
 async def init(*, reset: bool) -> None:  # pragma: no cover
     """Initialize the service database.
 
-    .. deprecated:: v1.5.0
+    .. deprecated:: v0.2.0
         The `init` command is deprecated in v0.2.0; it is replaced by alembic.
     """
-    print("Use `alembic upgrade head` instead.")
+    warnings.warn("Use `alembic upgrade head` instead.", DeprecationWarning)
 
 
 @server.command()
 @click.option("--port", default=8080, type=int, help="Port to run the application on.")
 def run(port: int) -> None:  # pragma: no cover
-    """Run the service application (for testing only)."""
-    uvicorn.run("lsst.cmservice.main:app", host="0.0.0.0", port=port, reload=True, reload_dirs=["src"])
+    """Run the service application (for testing only).
+
+    .. deprecated:: v0.2.0
+        The `run` command is deprecated in v0.2.0. Launch the server with a
+        module entrypoint instead.
+    """
+    warnings.warn("Use `python3 -m lsst.cmservice.main` instead.", DeprecationWarning)
 
 
 # Build the client CLI

src/lsst/cmservice/cli/step.py

@@ -130,8 +130,6 @@ def action() -> None:
 
 action_retry_script = wrappers.get_element_retry_script_command(action_command, sub_client)
 
-get_sleep_time = wrappers.get_element_estimate_sleep_time_command(action_command, sub_client)
-
 get_wms_task_reports = wrappers.get_element_wms_task_reports_command(get_command, sub_client)
 
 get_tasks = wrappers.get_element_tasks_command(get_command, sub_client)