fix(deployment): Use CLP Package's Docker Compose network for Presto deployments (fixes #1498). (#1501)

Author: junhaoliao

components/clp-package-utils/clp_package_utils/controller.py

@@ -503,8 +503,8 @@ def _set_up_env_for_webui(self, container_clp_config: CLPConfig) -> EnvVarsDict:
 
         query_engine = self._clp_config.package.query_engine
         if QueryEngine.PRESTO == query_engine:
-            server_settings_json_updates["PrestoHost"] = self._clp_config.presto.host
-            server_settings_json_updates["PrestoPort"] = self._clp_config.presto.port
+            server_settings_json_updates["PrestoHost"] = container_clp_config.presto.host
+            server_settings_json_updates["PrestoPort"] = container_clp_config.presto.port
         else:
             server_settings_json_updates["PrestoHost"] = None
             server_settings_json_updates["PrestoPort"] = None

components/clp-py-utils/clp_py_utils/clp_config.py

@@ -32,6 +32,7 @@
 RESULTS_CACHE_COMPONENT_NAME = "results_cache"
 COMPRESSION_SCHEDULER_COMPONENT_NAME = "compression_scheduler"
 QUERY_SCHEDULER_COMPONENT_NAME = "query_scheduler"
+PRESTO_COORDINATOR_COMPONENT_NAME = "presto-coordinator"
 COMPRESSION_WORKER_COMPONENT_NAME = "compression_worker"
 QUERY_WORKER_COMPONENT_NAME = "query_worker"
 WEBUI_COMPONENT_NAME = "webui"
@@ -591,9 +592,15 @@ class GarbageCollector(BaseModel):
 
 
 class Presto(BaseModel):
+    DEFAULT_PORT: ClassVar[int] = 8080
+
     host: DomainStr
     port: Port
 
+    def transform_for_container(self):
+        self.host = PRESTO_COORDINATOR_COMPONENT_NAME
+        self.port = self.DEFAULT_PORT
+
 
 def _get_env_var(name: str) -> str:
     value = os.getenv(name)
@@ -815,6 +822,8 @@ def transform_for_container(self):
         self.results_cache.transform_for_container()
         self.query_scheduler.transform_for_container()
         self.reducer.transform_for_container()
+        if self.package.query_engine == QueryEngine.PRESTO and self.presto is not None:
+            self.presto.transform_for_container()
 
 
 class WorkerConfig(BaseModel):

docs/src/user-docs/guides-using-presto.md

@@ -43,23 +43,6 @@ Using Presto with CLP requires:
         query_engine: "presto"
       ```
 
-    * Set the `database.host` key to a non-localhost hostname/IP.
-
-      ```yaml
-        database:
-        #  type: "mariadb"
-          host: "<non-local-ip-address>"
-        #  port: 3306
-        #  name: "clp-db"
-      ```
-
-      :::{note}
-      This change is necessary because the Presto containers run on a Docker network, and CLP's
-      database runs on the host network. `localhost` will refer to a different entity in each of
-      those contexts. This limitation will be addressed in the future when we unify Presto and CLP's
-      deployment infrastructure.
-      :::
-
     * Set the `results_cache.retention_period` key to `null` since the CLP + Presto integration
       doesn't yet support garbage collection.
 

tools/deployment/presto-clp/docker-compose.yaml

@@ -11,7 +11,7 @@ services:
       - "./coordinator/scripts:/scripts:ro"
       - "coordinator-config:/opt/presto-server/etc"
     networks:
-      - "presto"
+      - "clp-package"
     ports:
       - "8889:8080"
     healthcheck:
@@ -40,7 +40,7 @@ services:
       - "./worker/scripts:/scripts:ro"
       - "worker-config:/opt/presto-server/etc"
     networks:
-      - "presto"
+      - "clp-package"
 
 volumes:
   # Dummy volume to use when a bind mount is not desired.
@@ -50,5 +50,6 @@ volumes:
   worker-config:
 
 networks:
-  presto:
-    driver: "bridge"
+  clp-package:
+    name: "${CLP_PACKAGE_NETWORK_NAME:-clp-package_default}"
+    external: true

tools/deployment/presto-clp/scripts/init.py

@@ -7,6 +7,11 @@
 import yaml
 from dotenv import dotenv_values
 
+# Database endpoint inside the CLP Package Docker network. Must match the constants defined in
+# `components/clp-py-utils/clp_py_utils/clp_config.py`.
+DATABASE_COMPONENT_NAME = "database"
+DATABASE_DEFAULT_PORT = 3306
+
 # Set up console logging
 logging_console_handler = logging.StreamHandler()
 logging_formatter = logging.Formatter(
@@ -101,11 +106,9 @@ def _add_clp_env_vars(
         )
         return False
 
-    database_host = _get_config_value(clp_config, "database.host", "localhost")
-    database_port = _get_config_value(clp_config, "database.port", 3306)
     database_name = _get_config_value(clp_config, "database.name", "clp-db")
     env_vars["PRESTO_COORDINATOR_CLPPROPERTIES_METADATA_DATABASE_URL"] = (
-        f"jdbc:mysql://{database_host}:{database_port}"
+        f"jdbc:mysql://{DATABASE_COMPONENT_NAME}:{DATABASE_DEFAULT_PORT}"
     )
     env_vars["PRESTO_COORDINATOR_CLPPROPERTIES_METADATA_DATABASE_NAME"] = database_name
 
@@ -163,6 +166,11 @@ def _add_clp_env_vars(
     env_vars["PRESTO_COORDINATOR_CLPPROPERTIES_METADATA_DATABASE_USER"] = database_user
     env_vars["PRESTO_COORDINATOR_CLPPROPERTIES_METADATA_DATABASE_PASSWORD"] = database_password
 
+    instance_id = _get_clp_package_instance_id(clp_config, clp_package_dir)
+    if instance_id is None:
+        return False
+    env_vars["CLP_PACKAGE_NETWORK_NAME"] = f"clp-package-{instance_id}_default"
+
     return True
 
 
@@ -274,6 +282,46 @@ def _generate_worker_clp_properties(
     return True
 
 
+def _get_clp_package_instance_id(
+    clp_config: Dict[str, Any], clp_package_dir: Path
+) -> Optional[str]:
+    """
+    Retrieves the CLP package instance ID from the logs directory.
+
+    :param clp_config:
+    :param clp_package_dir:
+    :return: The instance ID if it could be read, otherwise `None`.
+    """
+
+    logs_directory = _get_path_clp_config_value(
+        clp_config, "logs_directory", Path("var") / "log", clp_package_dir
+    )
+    instance_id_path = logs_directory / "instance-id"
+    if not instance_id_path.exists():
+        logger.error(
+            "Cannot determine the CLP package Docker network because '%s' does not exist."
+            " Start the CLP package at least once before configuring Presto.",
+            instance_id_path,
+        )
+        return None
+
+    try:
+        instance_id = instance_id_path.read_text(encoding="utf-8").strip()
+    except OSError:
+        logger.exception("Failed to read the CLP package instance ID from '%s'.", instance_id_path)
+        return None
+
+    if not instance_id:
+        logger.error(
+            "Instance ID file '%s' is empty. Restart the CLP package to regenerate the instance"
+            " ID.",
+            instance_id_path,
+        )
+        return None
+
+    return instance_id
+
+
 def _get_path_clp_config_value(
     clp_config: Dict[str, Any], key: str, default_value: Path, clp_package_dir: Path
 ) -> Path: