matrice-compute 0.1.36__tar.gz → 0.1.38__tar.gz

{matrice_compute-0.1.36 → matrice_compute-0.1.38}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: matrice_compute
-Version: 0.1.36
+Version: 0.1.38
 Summary: Common server utilities for Matrice.ai services
 Author-email: "Matrice.ai" <dipendra@matrice.ai>
 License-Expression: MIT

{matrice_compute-0.1.36 → matrice_compute-0.1.38}/matrice_compute.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: matrice_compute
-Version: 0.1.36
+Version: 0.1.38
 Summary: Common server utilities for Matrice.ai services
 Author-email: "Matrice.ai" <dipendra@matrice.ai>
 License-Expression: MIT

{matrice_compute-0.1.36 → matrice_compute-0.1.38}/src/matrice_compute/action_instance.py

@@ -26,6 +26,10 @@ from matrice_common.utils import log_errors
 class ActionInstance:
     """Base class for tasks that run in Action containers."""
 
+    # Class-level dictionary to track deployed services and their ports
+    # Key: _idService, Value: {"triton_ports": "port1,port2,port3"}
+    _deployed_services = {}
+
     def __init__(self, scaling: Scaling, action_info: dict):
         """Initialize an action instance.
 
@@ -85,6 +89,52 @@ class ActionInstance:
             raise ValueError(f"Unknown action type: {self.action_type}")
         self.task = self.actions_map[self.action_type]
 
+    @classmethod
+    def get_or_create_triton_ports(cls, service_id, scaling_instance):
+        """Get existing TRITON_PORTS for a service or create new ones.
+
+        Args:
+            service_id (str): Service ID (_idService)
+            scaling_instance: Scaling instance to get open ports
+
+        Returns:
+            str: Comma-separated string of 3 port numbers (e.g., "8001,8002,8003")
+        """
+        if not service_id:
+            # No service_id, generate new ports
+            port1 = scaling_instance.get_open_port()
+            port2 = scaling_instance.get_open_port()
+            port3 = scaling_instance.get_open_port()
+            return f"{port1},{port2},{port3}"
+
+        # Check if ports already exist for this service
+        if service_id in cls._deployed_services:
+            triton_ports = cls._deployed_services[service_id]["triton_ports"]
+            logging.info(
+                "Reusing TRITON_PORTS for service %s: %s",
+                service_id,
+                triton_ports
+            )
+            return triton_ports
+
+        # First deployment: generate new ports and store them
+        port1 = scaling_instance.get_open_port()
+        port2 = scaling_instance.get_open_port()
+        port3 = scaling_instance.get_open_port()
+        triton_ports = f"{port1},{port2},{port3}"
+
+        # Store for future use
+        cls._deployed_services[service_id] = {
+            "triton_ports": triton_ports,
+        }
+
+        logging.info(
+            "First deployment for service %s - generated TRITON_PORTS: %s",
+            service_id,
+            triton_ports
+        )
+        return triton_ports
+
     @log_errors(default_return={}, raise_exception=True, log_error=False)
     def _init_credentials(self):
         """Initialize Matrice credentials.
@@ -346,6 +396,7 @@ class ActionInstance:
         destination_workspace_path: str = "/usr/src/workspace",
         docker_workdir: str = "",
         extra_pkgs: list = [],
+        container_name: str = "",
     ):
         """Build base Docker command with common options.
 
@@ -360,6 +411,7 @@ class ActionInstance:
             destination_workspace_path (str): Container workspace path
             docker_workdir (str): Docker working directory
             extra_pkgs (list): List of extra packages to install
+            container_name (str): Docker container name (format: {action_type}_{action_id})
         Returns:
             str: Base Docker command
         """
@@ -427,14 +479,19 @@ class ActionInstance:
         # if the service provider is local, then put --restart unless-stopped
         if os.environ.get("SERVICE_PROVIDER") in ("local", "LOCAL"):
             env_exports += " && export DOCKER_RESTART_POLICY='--restart unless-stopped' "
+            use_restart_policy = "--restart unless-stopped"
+        else:
+            use_restart_policy = ""
+
+        # Build container name option if provided
+        name_option = f"--name {container_name}" if container_name else ""
 
         cmd_parts = [
-            f"docker run {use_gpu} ",
+            f"docker run -d {use_gpu} {use_restart_policy} ",
             network_config,
             *[f"-e {key}={shlex.quote(str(value))}" for key, value in env_vars.items()],
             *volumes,
             # Container configuration and startup commands
-            f"--cidfile ./{self.action_record_id}.cid ",
             f"--shm-size=30G --pull=always {shlex.quote(self.docker_container)}",
             f'/bin/bash -c "cd {docker_workdir} && '
             f"{env_exports} && "
@@ -836,55 +893,50 @@ class ActionInstance:
         self.cmd = cmd
         self.log_path = f"{self.get_log_path()}/{log_name}_{self.action_record_id}.txt"
 
-        with open(self.log_path, "wb") as out:
-            self.process = subprocess.Popen(
-                shlex.split(self.cmd),
-                stdout=out,
-                stderr=out,
-                env={**os.environ},
-                start_new_session=True,
-            )
+        # Run docker with -d flag to get container ID from stdout
+        process = subprocess.Popen(
+            shlex.split(self.cmd),
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            env={**os.environ},
+        )
         
-        self.container_id = None
-
-        cid_file_path = f"./{self.action_record_id}.cid"
-        max_retries = 5
-        retry_delay = 1  # seconds
-        for attempt in range(max_retries):
-            try:
-                with open(cid_file_path, "r") as cid_file:
-                    container_id = cid_file.read().strip()
-                    self.container_id = container_id
-                    logging.info(
-                        "Started process for action %s with container ID: %s",
-                        self.action_record_id,
-                        self.container_id,
-                    )
-                    break
-            except FileNotFoundError:
-                logging.warning(
-                    "CID file not found for action %s, attempt %d/%d",
-                    self.action_record_id,
-                    attempt + 1,
-                    max_retries,
-                )
-                time.sleep(retry_delay)
-            except Exception as e:
-                logging.error(
-                    "Error reading CID file for action %s: %s",
-                    self.action_record_id,
-                    str(e),
-                )
-                time.sleep(retry_delay)
-        else:
+        # Use a longer timeout for docker run since --pull=always may need to
+        # download large images on first run. Default: 30 minutes (1800 seconds)
+        # Can be configured via DOCKER_START_TIMEOUT_SECONDS environment variable
+        docker_start_timeout = int(os.environ.get("DOCKER_START_TIMEOUT_SECONDS", 1800))
+        logging.info(
+            "Waiting for docker container to start for action %s (timeout: %d seconds)",
+            self.action_record_id,
+            docker_start_timeout,
+        )
+        stdout, stderr = process.communicate(timeout=docker_start_timeout)
+        
+        if process.returncode != 0:
             logging.error(
-                "Failed to read CID file for action %s after %d attempts",
+                "Docker run failed for action %s: %s",
                 self.action_record_id,
-                max_retries,
+                stderr,
             )
-            raise Exception("Failed to start process: CID file not found")
+            raise RuntimeError(f"Docker run failed: {stderr}")
+        
+        self.container_id = stdout.strip()
+        logging.info(
+            "Started container for action %s with ID: %s",
+            self.action_record_id,
+            self.container_id,
+        )
         
-        # report container id to scaling service
+        # Start following container logs in background
+        self.process = subprocess.Popen(
+            ["docker", "logs", "-f", self.container_id],
+            stdout=open(self.log_path, "wb"),
+            stderr=subprocess.STDOUT,
+            start_new_session=True,
+        )
+        
+        # Report container id to scaling service
         self.scaling.update_action_container_id(
             action_record_id=self.action_record_id,
             container_id=self.container_id,
@@ -1050,7 +1102,8 @@ def data_preparation_execute(
             "Started pulling Docker image with PID: %s",
             process.pid,
         )
-    cmd = f'{self.get_base_docker_cmd(work_fs, destination_workspace_path="/usr/src/app/workspace", docker_workdir="/usr/src/app/workspace", extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/data_preparation.py {self.action_record_id} "'
+    container_name = f"data_prep_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, destination_workspace_path="/usr/src/app/workspace", docker_workdir="/usr/src/app/workspace", extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/data_preparation.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "data_preparation_log")
 
@@ -1079,7 +1132,8 @@ def data_processing_execute(self: ActionInstance):
         service="bg-job-scheduler",
         job_params=action["jobParams"],
     )
-    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/main.py {self.action_record_id} "'
+    container_name = f"data_processing_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/main.py {self.action_record_id} "'
     logging.info("cmd: %s", cmd)
     self.start(cmd, "data_processing_log")
 
@@ -1092,7 +1146,8 @@ def data_split_execute(self: ActionInstance):
     if not action_details:
         return
     self.setup_action_requirements(action_details, work_fs, model_family="")
-    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/data_split.py {self.action_record_id} "'
+    container_name = f"data_split_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/data_split.py {self.action_record_id} "'
     logging.info("cmd: %s", cmd)
     self.start(cmd, "data_split")
 
@@ -1107,7 +1162,8 @@ def dataset_annotation_execute(
     if not action_details:
         return
     self.setup_action_requirements(action_details, work_fs)
-    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/dataset_annotation.py {self.action_record_id} "'
+    container_name = f"dataset_annotation_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/dataset_annotation.py {self.action_record_id} "'
     logging.info("cmd: %s", cmd)
     self.start(cmd, "dataset_annotation")
 
@@ -1122,7 +1178,8 @@ def dataset_augmentation_execute(
     if not action_details:
         return
     self.setup_action_requirements(action_details, work_fs)
-    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/data_augmentation.py {self.action_record_id} "'
+    container_name = f"dataset_augmentation_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/data_augmentation.py {self.action_record_id} "'
     logging.info("cmd: %s", cmd)
     self.start(cmd, "dataset_augmentation")
 
@@ -1138,7 +1195,8 @@ def augmentation_server_creation_execute(
     if not action_details:
         return
     self.setup_action_requirements(action_details, work_fs)
-    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/aug_server.py {self.action_record_id} {external_port} "'
+    container_name = f"augmentation_setup_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/aug_server.py {self.action_record_id} {external_port} "'
     logging.info("cmd: %s", cmd)
     self.start(cmd, "augmentation_setup")
 
@@ -1159,32 +1217,34 @@ def database_setup_execute(self: ActionInstance):
 
     project_id = action_details["_idProject"]
 
+    # Define container names with action_record_id for uniqueness
+    mongodb_container_name = f"database_setup_{self.action_record_id}"
+    qdrant_container_name = f"qdrant_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
-            "Using existing container ID for inference tracker: %s",
+            "Using existing container ID for database setup: %s",
             action_details["actionDetails"]["containerId"],
         )
         self.docker_container = action_details["actionDetails"]["containerId"]
         cmd = "docker restart " + self.docker_container 
-        self.start(cmd, "qdrant_setup")
+        self.start(cmd, "database_setup")
 
-        #qdrant restart
-        qdrant_cmd = "docker restart qdrant"
-        self.start(qdrant_cmd, 'qdrant_setup')
+        # qdrant restart
+        qdrant_cmd = f"docker restart {qdrant_container_name}"
+        self.start(qdrant_cmd, "qdrant_setup")
 
         return
-    
-
-    dbPath =action_details["jobParams"].get("dbPath","/host/data/path/mongodb_data")
 
+    dbPath = action_details["jobParams"].get("dbPath", "/host/data/path/mongodb_data")
 
     # MongoDB container with --net=host (Port: 27020:27017)
     cmd = (
-        f"docker run --pull=always --net=host "
+        f"docker run -d --pull=always --net=host "
+        f"--name {mongodb_container_name} "
+        f"-v matrice_myvol:/matrice_data "
         f"-v {dbPath}:{dbPath} "
-        f"--name database_setup_{self.action_record_id} "
         f"-v /var/run/docker.sock:/var/run/docker.sock "
-        f"--cidfile ./{self.action_record_id}.cid "
         f"-e ACTION_RECORD_ID={self.action_record_id} "
         f"-e MATRICE_ACCESS_KEY_ID={self.matrice_access_key_id} "
         f"-e MATRICE_SECRET_ACCESS_KEY={self.matrice_secret_access_key} "
@@ -1194,6 +1254,23 @@ def database_setup_execute(self: ActionInstance):
     )
     logging.info("Starting DB container (Port: 27020:27017): %s", cmd)
 
+    # Qdrant container with --net=host (Port: 6334)
+    qdrant_cmd = (
+        f"docker run -d --pull=always --net=host "
+        f"--name {qdrant_container_name} "
+        f"-v matrice_myvol:/matrice_data "
+        f"qdrant/qdrant:latest "
+    )
+    logging.info("Starting Qdrant container (Port: 6334): %s", qdrant_cmd)
+
+    # Start Qdrant container
+    qdrant_process = subprocess.Popen(
+        qdrant_cmd,
+        shell=True,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+    )
+    logging.info("Qdrant container started successfully")
 
     # Docker Command run
     self.start(cmd, "database_setup") 
@@ -1213,6 +1290,8 @@ def facial_recognition_setup_execute(self: ActionInstance):
 
     self.setup_action_requirements(action_details)
 
+    container_name = f"facial_recognition_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
             "Using existing container ID for facial recognition worker: %s",
@@ -1226,15 +1305,13 @@ def facial_recognition_setup_execute(self: ActionInstance):
     # Facial recognition worker container with --net=host (Port: 8081)
     worker_cmd = (
         f"docker run -d --pull=always --net=host "
-        f"--name worker "
-         f"--cidfile ./{self.action_record_id}.cid "
+        f"--name {container_name} "
         f"-v matrice_myvol:/matrice_data "
-        f"--cidfile ./{self.action_record_id}.cid "
         f'-e ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
         f'-e ACTION_ID="{self.action_record_id}" '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image}"
     )
     logging.info("Starting facial recognition worker (Port: 8081): %s", worker_cmd)
@@ -1256,6 +1333,8 @@ def lpr_setup_execute(self: ActionInstance):
 
     self.setup_action_requirements(action_details)
 
+    container_name = f"lpr_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
             "Using existing container ID for LPR worker: %s",
@@ -1269,15 +1348,14 @@ def lpr_setup_execute(self: ActionInstance):
     # LPR worker container with --net=host (Port: 8082)
     worker_cmd = (
         f"docker run -d --net=host --pull=always "
-        f"--name lpr-worker "
-        f"--cidfile ./{self.action_record_id}.cid "
+        f"--name {container_name} "
         f"-v matrice_myvol:/matrice_data "
         f'-e ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
         f'-e ACTION_ID="{self.action_record_id}" '
         f'-e PORT=8082 '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image}"
     )
     logging.info("Starting LPR worker (Port: 8082): %s", worker_cmd)
@@ -1308,6 +1386,8 @@ def inference_ws_server_execute(self: ActionInstance):
     
     logging.info(f"Inference WebSocket server will use IP: {ws_host} on port 8102 (use_host_network={use_host_network})")
 
+    container_name = f"inference_ws_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
             "Using existing container ID for inference WebSocket server: %s",
@@ -1321,12 +1401,11 @@ def inference_ws_server_execute(self: ActionInstance):
     # Inference WebSocket server with --net=host (Port: 8102)
     worker_cmd = (
         f"docker run -d --pull=always --net=host "
-        f"--name inference "
-        f"--cidfile ./{self.action_record_id}.cid "
+        f"--name {container_name} "
         f'-e ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image} "
         f"./app "
         f"{self.action_record_id} "
@@ -1357,6 +1436,8 @@ def fe_fs_streaming_execute(self: ActionInstance):
     
     logging.info(f"Frontend streaming will connect to WebSocket at: {ws_url}")
 
+    container_name = f"fe_streaming_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
             "Using existing container ID for frontend streaming: %s",
@@ -1370,15 +1451,14 @@ def fe_fs_streaming_execute(self: ActionInstance):
     # Frontend streaming with --net=host (Port: 3000)
     worker_cmd = (
         f"docker run -d --pull=always --net=host "
-        f"--name fe_streaming "
-        f"--cidfile ./{self.action_record_id}.cid "
+        f"--name {container_name} "
         f"-v matrice_myvol:/matrice_data "
         f'-e ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
         f"-e PORT=3000 "
         f'-e WS_HOST="{ws_url}" '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image}"
     )
     logging.info("Starting frontend streaming (Port: 3000) with WS_HOST=%s: %s", ws_url, worker_cmd)
@@ -1403,6 +1483,8 @@ def fe_analytics_service_execute(self: ActionInstance):
 
     project_id = action_details["_idProject"]
 
+    container_name = f"fe_analytics_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
             "Using existing container ID for frontend analytics service: %s",
@@ -1416,15 +1498,14 @@ def fe_analytics_service_execute(self: ActionInstance):
     # Frontend analytics service with --net=host (Port: 3001)
     worker_cmd = (
         f"docker run -d --pull=always --net=host "
-        f"--name fe-analytics "
-        f"--cidfile ./{self.action_record_id}.cid "
+        f"--name {container_name} "
         f'-e NEXT_PUBLIC_DEPLOYMENT_ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
         f'-e ACTION_ID="{self.action_record_id}" '
         f"-e PORT=3001 "
         f'-e PROJECT_ID="{project_id}" '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image}"
     )
     logging.info("Starting frontend analytics service (Port: 3001): %s", worker_cmd)
@@ -1449,7 +1530,8 @@ def synthetic_dataset_generation_execute(self: ActionInstance):
     else:
         return
     use_gpu = self.get_gpu_config(action_details)
-    cmd = f'{self.get_base_docker_cmd(work_fs=work_fs, use_gpu=use_gpu, extra_env_vars=extra_env_vars, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/synthetic_dataset_generation.py {self.action_record_id} "'
+    container_name = f"dataset_generation_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs=work_fs, use_gpu=use_gpu, extra_env_vars=extra_env_vars, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/synthetic_dataset_generation.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "dataset_generation")
 
@@ -1470,7 +1552,8 @@ def synthetic_data_setup_execute(self: ActionInstance):
     else:
         return
     use_gpu = self.get_gpu_config(action_details)
-    cmd = f'{self.get_base_docker_cmd(work_fs=work_fs, use_gpu=use_gpu, extra_env_vars=extra_env_vars, extra_pkgs=["matrice_dataset"])} python3 /usr/src/app/data_generation.py {self.action_record_id} {external_port} "'
+    container_name = f"synthetic_data_setup_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs=work_fs, use_gpu=use_gpu, extra_env_vars=extra_env_vars, extra_pkgs=["matrice_dataset"], container_name=container_name)} python3 /usr/src/app/data_generation.py {self.action_record_id} {external_port} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "synthetic_data_setup")
 
@@ -1507,6 +1590,8 @@ def redis_setup_execute(self: ActionInstance):
 
     redis_image = action_details["actionDetails"].get("redis_image", "redis:latest")
 
+    # Define container names with action_record_id for uniqueness
+    redis_container_name = f"redis_{self.action_record_id}"
 
     if action_details["actionDetails"].get("containerId"):
         logging.info(
@@ -1518,18 +1603,34 @@ def redis_setup_execute(self: ActionInstance):
         self.start(cmd, "redis_setup")
 
         # Redis container restart
-        redis_restart_cmd = "docker restart redis_container"
+        redis_restart_cmd = f"docker restart {redis_container_name}"
         self.start(redis_restart_cmd, "redis")
 
         return
     
-    # Redis container with --net=host (Port: 6379)
+    # Redis container with --net=host (Port: 6379) with optimized configuration
     redis_cmd = (
         f"docker run -d --net=host "
-        f"--name redis_container "
+        f"--name {redis_container_name} "
         f"--restart unless-stopped "
         f"{redis_image} "
-        f"redis-server --bind 0.0.0.0 --appendonly yes --requirepass {redis_password}"
+        f"redis-server --bind 0.0.0.0 "
+        f"--appendonly no "
+        f'--save "" '
+        f"--maxmemory 30gb "
+        f"--maxmemory-policy allkeys-lru "
+        f"--io-threads 4 "
+        f"--io-threads-do-reads yes "
+        f"--stream-node-max-bytes 8192 "
+        f"--stream-node-max-entries 1000 "
+        f"--hz 100 "
+        f"--tcp-backlog 2048 "
+        f"--timeout 0 "
+        f"--lazyfree-lazy-eviction yes "
+        f"--lazyfree-lazy-expire yes "
+        f"--lazyfree-lazy-server-del yes "
+        f"--activedefrag yes "
+        f"--requirepass {redis_password}"
     )
     
     logging.info("Starting Redis container on %s:6379: %s", redis_host, redis_cmd)
@@ -1553,8 +1654,9 @@ def redis_setup_execute(self: ActionInstance):
 
     # bg-redis management container with --net=host (Port: 8082)
     cmd = (
-        f"docker run --net=host "
-         f"--cidfile ./{self.action_record_id}.cid "
+        f"docker run -d --net=host "
+        f"--restart unless-stopped "
+        f"--name bg-redis_{self.action_record_id} "
         f"-e REDIS_URL={shlex.quote(env_vars['REDIS_URL'])} "
         f"-e REDIS_PASSWORD={shlex.quote(env_vars['REDIS_PASSWORD'])} "
         f"-e MATRICE_ACCESS_KEY_ID={shlex.quote(self.matrice_access_key_id)} "
@@ -1581,7 +1683,8 @@ def deploy_aggregator_execute(
     if not action_details:
         return
     self.setup_action_requirements(action_details, work_fs)
-    cmd = f'{self.get_base_docker_cmd(work_fs)} python3 /usr/src/app/deploy_aggregator.py {self.action_record_id} "'
+    container_name = f"deploy_aggregator_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, container_name=container_name)} python3 /usr/src/app/deploy_aggregator.py {self.action_record_id} "'
     logging.info("cmd: %s", cmd)
     self.start(cmd, "deploy_aggregator")
 
@@ -1597,6 +1700,10 @@ def model_deploy_execute(self: ActionInstance):
         return
     action_id = action_details["_id"]
     model_family = action_details["actionDetails"]["modelFamily"]
+
+    # Get the service ID to track deployments
+    service_id = action_details.get("_idService")
+
     self.setup_action_requirements(
         action_details,
         work_fs,
@@ -1604,17 +1711,29 @@ def model_deploy_execute(self: ActionInstance):
         action_id=action_id,
     )
 
-    # Get GPU configuration based on requirements and availability
-    # This uses the best-fit algorithm to select the most appropriate GPU(s)
-    use_gpu = self.get_gpu_config(action_details)
-
-    # Override: If GPU is required, use all available GPUs
+    # Use all GPUs if GPU is required
     gpuRequired = action_details["actionDetails"].get("gpuRequired", False)
     if gpuRequired:
         use_gpu = "--runtime=nvidia --gpus all"
+    else:
+        use_gpu = ""
+
+    logging.info(
+        "Action %s: Model deployment GPU config: %s",
+        action_id,
+        use_gpu if use_gpu else "CPU-only"
+    )
+
+    # Get or create TRITON_PORTS (uses utility method)
+    triton_ports = ActionInstance.get_or_create_triton_ports(service_id, self.scaling)
 
-    extra_env_vars = {"INTERNAL_PORT": internal_port}
-    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, mount_docker_sock=True, action_id=action_id, extra_env_vars=extra_env_vars, extra_pkgs=["matrice_inference", "matrice_analytics"])} python3 deploy.py {self.action_record_id} {external_port}"'
+    extra_env_vars = {
+        "INTERNAL_PORT": internal_port,
+        "TRITON_PORTS": triton_ports
+    }
+
+    container_name = f"model_deploy_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, mount_docker_sock=True, action_id=action_id, extra_env_vars=extra_env_vars, extra_pkgs=["matrice_inference", "matrice_analytics"], container_name=container_name)} python3 deploy.py {self.action_record_id} {external_port}"'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "deploy_log")
 
@@ -1647,7 +1766,8 @@ def model_train_execute(self: ActionInstance):
         self.start(cmd, "train_log")
         return
     
-    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, action_id=action_id, model_key=model_key)} python3 train.py {self.action_record_id} "'
+    container_name = f"model_train_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, action_id=action_id, model_key=model_key, container_name=container_name)} python3 train.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "train_log")
 
@@ -1670,7 +1790,7 @@ def model_eval_execute(self: ActionInstance):
     )
     if action_details["actionDetails"].get("containerId"):
         logging.info(
-            "Using existing container ID for training: %s",
+            "Using existing container ID for evaluation: %s",
             action_details["actionDetails"]["containerId"],
         )
         self.docker_container = action_details["actionDetails"]["containerId"]
@@ -1678,7 +1798,8 @@ def model_eval_execute(self: ActionInstance):
         self.start(cmd, "eval_log")
         return
     
-    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, action_id=action_id)} python3 eval.py {self.action_record_id} "'
+    container_name = f"model_eval_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, action_id=action_id, container_name=container_name)} python3 eval.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "eval_log")
 
@@ -1704,7 +1825,7 @@ def model_export_execute(self: ActionInstance):
     )
     if action_details["actionDetails"].get("containerId"):
         logging.info(
-            "Using existing container ID for training: %s",
+            "Using existing container ID for export: %s",
             action_details["actionDetails"]["containerId"],
         )
         self.docker_container = action_details["actionDetails"]["containerId"]
@@ -1712,7 +1833,8 @@ def model_export_execute(self: ActionInstance):
         self.start(cmd, "export_log")
         return
     
-    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, action_id=action_id)} python3 export.py {self.action_record_id} "'
+    container_name = f"model_export_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(work_fs, use_gpu, action_id=action_id, container_name=container_name)} python3 export.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "export_log")
 
@@ -1728,7 +1850,8 @@ def image_build_execute(self: ActionInstance):
     action_id = action_details["_id"]
     internal_api_key = self.get_internal_api_key(action_id)
     extra_env_vars = {"MATRICE_INTERNAL_API_KEY": internal_api_key}
-    cmd = f'{self.get_base_docker_cmd(mount_docker_sock=True, extra_env_vars=extra_env_vars)} python3 main.py {model_family_id} {action_id}"'
+    container_name = f"image_build_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(mount_docker_sock=True, extra_env_vars=extra_env_vars, container_name=container_name)} python3 main.py {model_family_id} {action_id}"'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "image_build_log")
 
@@ -1740,7 +1863,8 @@ def resource_clone_execute(self: ActionInstance):
     if not action_details:
         return
     self.setup_action_requirements(action_details)
-    cmd = f'{self.get_base_docker_cmd()} python3 main.py {self.action_record_id} "'
+    container_name = f"resource_clone_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(container_name=container_name)} python3 main.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "resource_clone")
 
@@ -1758,7 +1882,7 @@ def streaming_gateway_execute(self: ActionInstance):
         )
     if action_details["actionDetails"].get("containerId"):
         logging.info(
-            "Using existing container ID for training: %s",
+            "Using existing container ID for streaming gateway: %s",
             action_details["actionDetails"]["containerId"],
         )
         self.docker_container = action_details["actionDetails"]["containerId"]
@@ -1766,7 +1890,8 @@ def streaming_gateway_execute(self: ActionInstance):
         self.start(cmd, "streaming_gateway")
         return
     
-    cmd = f'{self.get_base_docker_cmd(extra_pkgs=["matrice_streaming"])} python3 /usr/src/app/streaming_gateway.py {self.action_record_id} "'
+    container_name = f"streaming_gateway_{self.action_record_id}"
+    cmd = f'{self.get_base_docker_cmd(extra_pkgs=["matrice_streaming"], container_name=container_name)} python3 /usr/src/app/streaming_gateway.py {self.action_record_id} "'
     logging.info("cmd is: %s", cmd)
     self.start(cmd, "streaming_gateway")
 
@@ -1862,7 +1987,7 @@ def kafka_setup_execute(self: ActionInstance):
 
     if action_details["actionDetails"].get("containerId"):
         logging.info(
-            "Using existing container ID for training: %s",
+            "Using existing container ID for kafka: %s",
             action_details["actionDetails"]["containerId"],
         )
         self.docker_container = action_details["actionDetails"]["containerId"]
@@ -1870,10 +1995,12 @@ def kafka_setup_execute(self: ActionInstance):
         self.start(cmd, "kafka_setup")
         return
 
+    container_name = f"kafka_{self.action_record_id}"
 
     # Kafka container with --net=host (Ports: 9092, 9093)
     cmd = (
-        f"docker run --net=host "
+        f"docker run -d --net=host "
+        f"--name {container_name} "
         f"{env_args} "
         f"--shm-size=30G --pull=always "
         f'aiforeveryone/matrice-kafka:latest /bin/bash -c "'
@@ -1906,6 +2033,8 @@ def inference_tracker_setup_execute(self: ActionInstance):
     
     self.setup_action_requirements(action_details)
 
+    container_name = f"inference_tracker_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
             "Using existing container ID for inference tracker: %s",
@@ -1919,14 +2048,13 @@ def inference_tracker_setup_execute(self: ActionInstance):
     # This is the existing Docker run command
     worker_cmd = (
         f"docker run -d --pull=always --net=host "
-         f"--cidfile ./{self.action_record_id}.cid "
-        f"--name inference-tracker-worker "
+        f"--name {container_name} "
         f"-v matrice_myvol:/matrice_data "
         f'-e ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
         f'-e ACTION_ID="{self.action_record_id}" '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image}"
     )
     
@@ -1948,9 +2076,11 @@ def video_storage_setup_execute(self: ActionInstance):
     
     self.setup_action_requirements(action_details)
 
+    container_name = f"video_storage_{self.action_record_id}"
+
     if action_details["actionDetails"].get("containerId"):
         logging.info(
-            "Using existing container ID for inference tracker: %s",
+            "Using existing container ID for video storage: %s",
             action_details["actionDetails"]["containerId"],
         )
         self.docker_container = action_details["actionDetails"]["containerId"]
@@ -1961,14 +2091,13 @@ def video_storage_setup_execute(self: ActionInstance):
     # This is the existing Docker run command
     worker_cmd = (
         f"docker run -d --pull=always --net=host "
-         f"--cidfile ./{self.action_record_id}.cid "
-        f"--name media_server "
+        f"--name {container_name} "
         f"-v matrice_myvol:/matrice_data "
         f'-e ENV="{os.environ.get("ENV", "prod")}" '
         f'-e MATRICE_SECRET_ACCESS_KEY="{self.matrice_secret_access_key}" '
         f'-e MATRICE_ACCESS_KEY_ID="{self.matrice_access_key_id}" '
         f'-e ACTION_ID="{self.action_record_id}" '
-        f' --restart=unless-stopped '
+        f'--restart=unless-stopped '
         f"{image}"
     )
     

{matrice_compute-0.1.36 → matrice_compute-0.1.38}/src/matrice_compute/resources_tracker.py

@@ -916,14 +916,27 @@ class ResourcesTracker:
         gpu_count = 0
         
         for gpu in gpu_data['gpus']:
-            gpu_memory_free += gpu['memory_total'] - gpu['memory_used']
+            # Be defensive: nvidia-smi can occasionally report N/A/0 for total while used is numeric,
+            # which would otherwise produce negative "free" memory.
+            total_mb = gpu.get('memory_total', 0) or 0
+            used_mb = gpu.get('memory_used', 0) or 0
+            free_mb = total_mb - used_mb
+            if free_mb < 0:
+                logging.debug(
+                    "Negative GPU free memory computed (gpu_idx=%s total_mb=%s used_mb=%s); clamping to 0",
+                    gpu.get('idx'),
+                    total_mb,
+                    used_mb,
+                )
+                free_mb = 0
+            gpu_memory_free += free_mb
             gpu_utilization += gpu['utilization']
             gpu_count += 1
         
         if gpu_count > 0:
             gpu_utilization /= gpu_count
-        
-        return gpu_memory_free, gpu_utilization
+
+        return max(0, gpu_memory_free), gpu_utilization
 
     @log_errors(default_return=(0, 0.0), raise_exception=False, log_error=False)
     def _get_gpu_resources_direct(self) -> Tuple[int, float]:
@@ -1218,7 +1231,7 @@ class MachineResourcesTracker:
             availableCPU=available_cpu,
             availableMemory=available_memory,
             availableGPU=100 - gpu_utilization,
-            availableGPUMemory=gpu_memory_free,
+            availableGPUMemory=max(0, gpu_memory_free),
         )
         if err is not None:
             logging.error(