Adding/refactoring unified config

fixes to the conf functionality

lint fix

add configuration data to readme

address pr comments

README.md

@@ -115,6 +115,20 @@ Get the code:
 
     Note: If you are using an [Azure Ubuntu HPC-AI](https://github.com/Azure/azhpc-images) VM image you can find the Moneo in this path: /opt/azurehpc/tools/Moneo
 
+### Configuration File ###
+
+The [moneo_config.json](./moneo_config.json) file can be used to specify certain deployment settings prior to moneo deployment.
+
+There are 4 groups of configurations:
+
+  1. exporter_conf - This applies to all deployments. See the following settings:
+     - gpu_sample_interval - Sample rate per minute for Nvidia GPU exporter. Choices are [1, 2, 30, 60, 120, 600]. with 60 samples per minute being default.
+     - gpu_profiling - Switches on additional profile metrics (Tensor, FP16, FP32, and FP64). Choices are true/false with false as default.
+     - Note: These settings may have an impact on performance. Default settings were chosen to minimize impact.
+  2. prom_config - This group of settings applies to the Headless deployment method. Refer to [Headless Deployment Guide](./docs/HeadlessDeployment.md) for usage.
+  3. geneva_config - Applies to Geneva deployement. Refer to [Geneva deployment](./docs/GenevaAgent.MD) for usage.
+  4. publisher_config - Applies to both Geneva and Azure Monitor agent deployment methods see [Geneva deployment](./docs/GenevaAgent.MD) or [Azure Monitor Agent deployment](./docs/AzureMonitorAgent.md) for usage.
+
 ### Prefered Moneo Deployment ###
 
 The prefered way to deploy Moneo is the headless method using Azure Managaed Grafana and Prometheus resources.
@@ -185,7 +199,7 @@ Note: For more options check the Moneo help menu
 
 1. For Managed Grafana (headless) deployment
     - Verify that the user managed identity is assigned to the VM resource.
-    - Verify the prerequisite configure file (`Moneo/src/worker/publisher/config/managed_prom_config.json`) is configured correctly on each worker node.
+    - Verify the prerequisite configure file (`Moneo/moneo_config.json`) is configured correctly on each worker node.
     - On the worker nodes verify functionality of prometheus agent remote write:
         - Check prometheus docker with `sudo docker logs prometheus | grep 'Done replaying WAL'`
         It will have the result like this:

dockerfile/moneo-exporter-nvidia.dockerfile

@@ -6,8 +6,6 @@ ARG BRANCH_OR_TAG=main
 
 ENV DCGM_VERSION=3.1.1
 ENV OFED_VERSION=23.07-0.5.1.2
-ENV PROFILING false
-ENV GPU_SAMPLE_RATE 2
 
 # Install dependencies
 RUN apt-get update -y                           \
@@ -43,11 +41,14 @@ RUN cd /tmp && \
 RUN git config --global advice.detachedHead false
 RUN git clone --branch ${BRANCH_OR_TAG} https://github.com/Azure/Moneo.git
 
+# Set up tmp space for Moneo
+RUN mkdir -p /tmp/moneo-worker
+
 # Install DCGM
 WORKDIR Moneo/src/worker
 RUN sudo bash install/nvidia.sh
 
 # Set EntryPoint
 COPY dockerfile/moneo-exporter-nvidia_entrypoint.sh .
 RUN chmod +x moneo-exporter-nvidia_entrypoint.sh
-CMD /bin/bash moneo-exporter-nvidia_entrypoint.sh ${PROFILING} ${GPU_SAMPLE_RATE}
+CMD /bin/bash moneo-exporter-nvidia_entrypoint.sh

dockerfile/moneo-exporter-nvidia_entrypoint.sh

@@ -1,22 +1,17 @@
 #!/bin/bash
 set -e
 
-enable_profiling=$1
-gpu_sample_rate=$2
-ethernet_dev_name=$3
+# Ethernet device naming, if not present it will use default eth0 name
+ethernet_dev_name=$1
 
 # Start NVIDIA, Net and Node Exporter
 echo "Starting NVIDIA, Net and Node Exporter"
 
-if [ $enable_profiling = true ]; then
-    python3 exporters/nvidia_exporter.py -m -s $gpu_sample_rate &
-else
-    python3 exporters/nvidia_exporter.py -s $gpu_sample_rate &
-fi
+python3 exporters/nvidia_exporter.py  &
 
 python3 exporters/net_exporter.py --inifiband_sysfs=/hostsys/class/infiniband &
 
-if [-n $ethernet_dev_name]; then
+if [ -n "$ethernet_dev_name" ]; then
     python3 exporters/node_exporter.py -e $ethernet_dev_name &
 else
     python3 exporters/node_exporter.py &

docs/AzureMonitorAgent.md

@@ -8,7 +8,7 @@ Prequisites:
 1. An Azure Monitor Metrics (Application Insights) resource, please enable alerting on custom metric dimensions by refering this [document](https://learn.microsoft.com/en-us/azure/azure-monitor/app/pre-aggregated-metrics-log-metrics#custom-metrics-dimensions-and-pre-aggregation) to restore the metrics dimentions.(Lead to a extra cost)
 2. PSSH installed on manager nodes.
 3. Ensure passwordless ssh is installed in you environment.
-4. Config publisher config file in `Moneo/src/worker/publisher/config/publisher_config.json`.
+4. Config publisher config file in `Moneo/moneo_config.json`.
     Note: You can obtain your connection string from the Application Insights pages you created in the Azure portal.
     ```
     {

docs/HeadlessDeployment.md

@@ -36,13 +36,13 @@ Follow steps outlined in [Infrastructure deployment](../deploy_managed_infra/REA
     3. Skip to step 5.
     Note: This step can be performed in parallel using pssh. Reference step 4 for start and stop commands.
 
-3. Modify the managed prometheus config file in `Moneo/src/worker/publisher/config/managed_prom_config.json`.
+3. Modify the managed prometheus config file in `Moneo/moneo_config.json`.
     - Reference the user managed identity created during infrastructure deployment to get the "identity client id"
     - Reference the Managed Prometheus resource created during infrastructure deployment to get the "metrics ingestion endpoint"
     - The config file modifcations must be distributed to the Moneo directories on all workers.
 
     ```json
-    {
+    "prom_config": {
         "IDENTITY_CLIENT_ID": "<identity client id>",
         "INGESTION_ENDPOINT": "<metrics ingestion endpoint>"
     } 

docs/ManagedPrometheusAgent.md

@@ -18,11 +18,11 @@ This guide will provide step-by-step instructions on how to  to publish your exp
     - Click add at the bottom of the open blade.
 3. PSSH installed on manager nodes.
 4. Ensure passwordless ssh is installed in you environment.
-5. Config managed prometheus config file in `Moneo/src/worker/publisher/config/managed_prom_config.json`.
+5. Config managed prometheus config file in `Moneo/moneo_config.json`.
     Note: You can obtain your IDENTITY_CLIENT_ID in your indentity resource page and your metrics ingestion endpoint from the AWM pages you created in the Azure portal.
 
     ``` json
-        {
+        "prom_config": {
             "IDENTITY_CLIENT_ID": "<identity client id>",
             "INGESTION_ENDPOINT": "<metrics ingestion endpoint>"
         }

linux_service/README.md

@@ -45,17 +45,18 @@ Configuration/Installation is only required once. After that is complete the Lin
 
 1. Configuration and installation of the Linux service is done with the following command:
    ```parallel-ssh -i -t 0 -h hostfile "sudo /opt/azurehpc/tools/Moneo/linux_service/configure_service.sh"```
-     - Note: If using Azure monitor or Geneva add an extra argument "./start_moneo_services.sh azure_monitor"  or "./configure_service.sh geneva" respectively.
+     - Note: If using the Azure AI/HPC VM market place image, this step is already completed for managed prometheus deployment
+     - Note: If using Azure monitor or Geneva add an extra argument "./configure_service.sh azure_monitor"  or "./configure_service.sh geneva" respectively.
      - Note: Geneva authentication is user managed identity "umi" by default, you can choose to change to "cert" method by modifiying [the start script](./configure_service.sh) "PUBLISHER_AUTH" variable.
 
 2. For Azure Monitor or Managed Prometheus methods if you have not yet modified the configuration files reference the following:
    - For Azure Managed Prometheus:
-     - modify [managed_prom_config.json](../src/worker/publisher/config) and copy the file to the compute nodes.
-     - ```parallel-scp -h hostfile /opt/azurehpc/tools/Moneo/src/worker/publisher/config/managed_prom_config.json /opt/azurehpc/tools/Moneo/src/worker/publisher/config```
+     - modify [moneo_config.json](../moneo_config.json) and copy the file to the compute nodes.
+     - ```parallel-scp -h hostfile /opt/azurehpc/tools/Moneo/moneo_config.json /opt/azurehpc/tools/Moneo```
      - Lastly check that that the managed user identity used to set up Managed Prometheus (Azure role assignments) is assigned to your VMSS.
    - For Azure Monitor:
      - modify the connection string of "azure_monitor_agent_config" section and copy the file to the compute nodes.
-     - ```parallel-scp -h hostfile /opt/azurehpc/tools/Moneo/src/worker/publisher/config/publisher_config.json /opt/azurehpc/tools/Moneo/src/worker/publisher/config```
+     - ```parallel-scp -h hostfile /opt/azurehpc/tools/Moneo/moneo_config.json /opt/azurehpc/tools/Moneo```
 
 ### Launch Services ###
 
@@ -84,10 +85,10 @@ Stopping services is the same command for all methods.
 
 Assuming configuration files have been updated and user managed ID applied if necessary (Managed Prometheus) reference these commands for the work flow:
 
-- Configuration/Install:
+- Configuration/Install (not needed for market place image, using managed Prometheus):
    ```parallel-ssh -i -t 0 -h hostfile "sudo /opt/azurehpc/tools/Moneo/linux_service/configure_service.sh"```
 - Extra Configure step for AZ Monitor and/or Managed Prometheus
-   ```parallel-scp -h hostfile /opt/azurehpc/tools/Moneo/src/worker/publisher/config/<Respective config file> /opt/azurehpc/tools/Moneo/src/worker/publisher/config```
+   ```parallel-scp -h hostfile /opt/azurehpc/tools/Moneo/moneo_config.json /opt/azurehpc/tools/Moneo```
 - Start
    ```parallel-ssh -i -t 0 -h hostfile "sudo /opt/azurehpc/tools/Moneo/linux_service/start_moneo_services.sh"```
    Note:

linux_service/start_moneo_services.sh

@@ -61,6 +61,9 @@ function proc_check(){
     echo "All Services Running"
     exit 0
 }
+# stop nvidia exporter in the event there was a config change
+sudo systemctl stop moneo@nvidia_exporter.service 2> /dev/null
+sleep 2 # wait a bit for the exporter to stop
 
 $MONEO_PATH/linux_service/moneo_prestart.sh $MONEO_PATH 2> /dev/null
 

moneo.py

@@ -123,8 +123,8 @@ def deploy_worker(self, hosts_file, max_threads=16):    # noqa: C901
         logging.info('Copying files to workers')
         out = pscp(copy_path, destination_dir, hosts_file, user=self.args.user)
         logging.info(out)
+        # Copy config file
         copy_path = './moneo_config.json'
-        destination_dir = '/tmp/moneo-worker'
         out = pscp(copy_path, destination_dir, hosts_file, user=self.args.user)
         logging.info(out)
         print('--------------------------')
@@ -150,12 +150,6 @@ def deploy_worker(self, hosts_file, max_threads=16):    # noqa: C901
         print('-Starting metric exporters on workers-')
         logging.info('Starting metric exporters on workers')
         cmd = '/tmp/moneo-worker/start.sh'
-        if self.args.profiler_metrics:
-            print('-Profiling enabled-')
-            logging.info('Profiling enabled')
-            cmd = cmd + ' true'
-        else:
-            cmd = cmd + ' false'
         if self.args.launch_publisher:
             agent = self.args.launch_publisher
             if agent == 'geneva' and not self.args.publisher_auth:
@@ -183,8 +177,8 @@ def deploy_worker(self, hosts_file, max_threads=16):    # noqa: C901
         else:
             cmd = cmd + ' false'
             cmd = cmd + " \"\""
-        #  gpu sample rate + ethernet device
-        cmd = cmd + " " + str(args.gpu_sample_rate) + " " + args.ethernet_device
+        # ethernet device
+        cmd = cmd + " " + args.ethernet_device
         if self.args.custom_metrics_file_path:
             print('-Custom exporter enabled-')
             logging.info('Custom exporter enabled')
@@ -201,6 +195,10 @@ def deploy_work_docker(self, hosts_file, max_threads=16):
         logging.info('Deploying docker container to workers')
         out = pscp(copy_path, destination_dir, hosts_file, user=self.args.user)
         logging.info(out)
+        # Copy config file
+        copy_path = './moneo_config.json'
+        out = pscp(copy_path, destination_dir, hosts_file, user=self.args.user)
+        logging.info(out)
         out = pssh(cmd='/tmp/moneo-worker/deploy_docker.sh',
                    hosts_file=hosts_file, max_threads=max_threads, user=self.args.user)
         logging.info(out)
@@ -356,13 +354,6 @@ def parallel_ssh_check():
         default='full',
         nargs="?",
         help='Type of deployment/shutdown. Choices: {manager,workers,full}. Default: full.')
-    parser.add_argument(
-        '-p',
-        '--profiler_metrics',
-        action='store_true',
-        default=False,
-        help='Enable profile metrics (Tensor Core,FP16,FP32,FP64 activity).'
-             'Addition of profile metrics encurs additional overhead on computer nodes.')
     parser.add_argument(
         '-r',
         '--container',
@@ -408,11 +399,6 @@ def parallel_ssh_check():
         '--custom_metrics_file_path',
         type=str,
         help='The path of the custom metrics file.')
-    parser.add_argument(
-        '--gpu_sample_rate',
-        type=int,
-        choices=[1, 2, 30, 60, 120, 600],
-        help='Number of samples per minute for GPU monitoring. Valid options are 1,2,3,10', default=60)
     parser.add_argument(
         '--ethernet_device',
         type=str,

moneo_config.json

@@ -1,5 +1,8 @@
-
 {
+    "exporter_config":{
+        "gpu_sample_interval": "60",
+        "gpu_profiling": "false"
+    },
     "prom_config":{
         "IDENTITY_CLIENT_ID": "<identity client id>",
         "INGESTION_ENDPOINT": "<metrics ingestion endpoint>"

src/worker/deploy_docker.sh

@@ -2,15 +2,14 @@
 
 IMAGE=azmoneo/moneo-exporter:nvidia
 CONT_NAME=moneo-exporter-nvidia
-PROFILING=$1
 
 if [ -e "/dev/nvidiactl" ]; then
     docker pull $IMAGE
 
     docker rm --force $CONT_NAME && \
     docker run --name=$CONT_NAME --net=host --restart=unless-stopped \
-        -e PROFILING=$PROFILING --rm --runtime=nvidia \
-        --cap-add SYS_ADMIN -v /sys:/hostsys/ -itd $IMAGE
+        --rm --runtime=nvidia \
+        --cap-add SYS_ADMIN -v /sys:/hostsys/ -v /tmp/moneo-worker/moneo_config.json:/tmp/moneo-worker/moneo_config.json -itd $IMAGE
 else
 
 echo 'No Nvidia devices found Docker deployment canceled'

src/worker/exporters/nvidia_exporter.py

@@ -3,7 +3,7 @@
 import time
 import signal
 import logging
-
+import json
 import prometheus_client
 
 sys.path.append('/usr/local/dcgm/bindings/python3')
@@ -276,15 +276,40 @@ def Loop(self):
             pass
 
 
+def get_custom_config():
+    try:
+        with open('/tmp/moneo-worker/moneo_config.json') as f:
+            mon_config = json.load(f)
+
+        sample_per_min = int(mon_config['exporter_config']['gpu_sample_interval'])
+        sample_intervals = [1, 2, 30, 60, 120, 600]
+
+        if sample_per_min not in sample_intervals:
+            mon_config['exporter_config']['gpu_sample_interval'] = 60
+        else:
+            mon_config['exporter_config']['gpu_sample_interval'] = sample_per_min
+
+        if (mon_config['exporter_config']['gpu_profiling']).lower() == "true":
+            mon_config['exporter_config']['gpu_profiling'] = True
+        else:
+            mon_config['exporter_config']['gpu_profiling'] = False
+        return mon_config
+    except Exception:
+        mon_config = {'exporter_config': {'gpu_sample_interval': 60, 'gpu_profiling': False}}
+        return mon_config
+
+
 def init_config():
     global dcgm_config
+    mon_config = get_custom_config()
     dcgm_config = {
         'exit': False,
         'ignoreList': [],
         'dcgmHostName': None,
         'prometheusPort': None,
-        'prometheusPublishInterval': None,
+        'prometheusPublishInterval': mon_config['exporter_config']['gpu_sample_interval'],
         'publishFieldIds': None,
+        'profilerMetrics': mon_config['exporter_config']['gpu_profiling'],
         'last_value': {}
     }
 
@@ -304,22 +329,9 @@ def parse_dcgm_cli():
         publish_port=8000,
         log_level='INFO',
     )
-    parser.add_argument(
-        '-m',
-        '--profiler_metrics',
-        action='store_true',
-        help='Enable profile metrics (Tensor Core,FP16,FP32,FP64 activity).'
-             'Addition of profile metrics encurs additional overhead on computer nodes.')
-    parser.add_argument(
-        '-s',
-        '--sample_per_min',
-        type=int,
-        default=60,
-        choices=[1, 2, 30, 60, 120, 600],
-        help='Samples per minute. Default 60')
     args = dcgm_client_cli_parser.run_parser(parser)
     # add profiling metrics if flag enabled
-    if (args.profiler_metrics):
+    if (dcgm_config['profilerMetrics']):
         args.field_ids.extend(DCGM_PROF_FIELDS)
     field_ids = dcgm_client_cli_parser.get_field_ids(args)
     numeric_log_level = dcgm_client_cli_parser.get_log_level(args)
@@ -334,11 +346,9 @@ def parse_dcgm_cli():
     else:
         dcgm_config['dcgmHostName'] = args.hostname
     dcgm_config['prometheusPort'] = args.publish_port
-    dcgm_config['prometheusPublishInterval'] = int(args.sample_per_min)
     dcgm_config['publishFieldIds'] = field_ids
     dcgm_config['sendUuid'] = True
     dcgm_config['jobId'] = None
-    dcgm_config['profilerMetrics'] = args.profiler_metrics
     logging.basicConfig(
         level=numeric_log_level,
         filemode=filemode,

src/worker/publisher/metrics_publisher.py

@@ -94,7 +94,7 @@ def get_publisher_metrics_config():
     Returns:
         config(dict): The geneva metrics configuration
     """
-    with open('/tmp/moneo-worker/publisher/config/moneor_config.json') as f:
+    with open('/tmp/moneo-worker/moneo_config.json') as f:
         config = json.load(f)
     return config