Nvidia SMI and PostgreSQL Integration

Input and output integration overview

The Nvidia SMI Plugin enables the retrieval of detailed statistics about NVIDIA GPUs attached to the host system, providing essential insights for performance monitoring.

The Telegraf PostgreSQL plugin allows you to efficiently write metrics to a PostgreSQL database while automatically managing the database schema.

Integration details

Nvidia SMI

The Nvidia SMI Plugin is designed to gather metrics regarding the performance and status of NVIDIA GPUs on the host machine. By leveraging the capabilities of the nvidia-smi command-line tool, this plugin pulls crucial information such as GPU memory utilization, temperature, fan speed, and various performance metrics. This data is essential for monitoring GPU health and performance in real-time, particularly in environments where GPU performance directly impacts computing tasks, such as machine learning, 3D rendering, and high-performance computing. The plugin provides flexibility by allowing users to specify the path to the nvidia-smi binary and configure polling timeouts, accommodating both Linux and Windows systems where the nvidia-smi tool is commonly located. With its ability to collect detailed statistics on each GPU, this plugin becomes a vital resource for any infrastructure relying on NVIDIA hardware, facilitating proactive management and performance tuning.

PostgreSQL

The PostgreSQL plugin enables users to write metrics to a PostgreSQL database or a compatible database, providing robust support for schema management by automatically updating missing columns. The plugin is designed to facilitate integration with monitoring solutions, allowing users to efficiently store and manage time series data. It offers configurable options for connection settings, concurrency, and error handling, and supports advanced features such as JSONB storage for tags and fields, foreign key tagging, templated schema modifications, and support for unsigned integer data types through the pguint extension.

Configuration

Nvidia SMI

[[inputs.nvidia_smi]]
  ## Optional: path to nvidia-smi binary, defaults "/usr/bin/nvidia-smi"
  ## We will first try to locate the nvidia-smi binary with the explicitly specified value (or default value),
  ## if it is not found, we will try to locate it on PATH(exec.LookPath), if it is still not found, an error will be returned
  # bin_path = "/usr/bin/nvidia-smi"

  ## Optional: timeout for GPU polling
  # timeout = "5s"

PostgreSQL

# Publishes metrics to a postgresql database
[[outputs.postgresql]]
  ## Specify connection address via the standard libpq connection string:
  ##   host=... user=... password=... sslmode=... dbname=...
  ## Or a URL:
  ##   postgres://[user[:password]]@localhost[/dbname]?sslmode=[disable|verify-ca|verify-full]
  ## See https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNSTRING
  ##
  ## All connection parameters are optional. Environment vars are also supported.
  ## e.g. PGPASSWORD, PGHOST, PGUSER, PGDATABASE
  ## All supported vars can be found here:
  ##  https://www.postgresql.org/docs/current/libpq-envars.html
  ##
  ## Non-standard parameters:
  ##   pool_max_conns (default: 1) - Maximum size of connection pool for parallel (per-batch per-table) inserts.
  ##   pool_min_conns (default: 0) - Minimum size of connection pool.
  ##   pool_max_conn_lifetime (default: 0s) - Maximum age of a connection before closing.
  ##   pool_max_conn_idle_time (default: 0s) - Maximum idle time of a connection before closing.
  ##   pool_health_check_period (default: 0s) - Duration between health checks on idle connections.
  # connection = ""

  ## Postgres schema to use.
  # schema = "public"

  ## Store tags as foreign keys in the metrics table. Default is false.
  # tags_as_foreign_keys = false

  ## Suffix to append to table name (measurement name) for the foreign tag table.
  # tag_table_suffix = "_tag"

  ## Deny inserting metrics if the foreign tag can't be inserted.
  # foreign_tag_constraint = false

  ## Store all tags as a JSONB object in a single 'tags' column.
  # tags_as_jsonb = false

  ## Store all fields as a JSONB object in a single 'fields' column.
  # fields_as_jsonb = false

  ## Name of the timestamp column
  ## NOTE: Some tools (e.g. Grafana) require the default name so be careful!
  # timestamp_column_name = "time"

  ## Type of the timestamp column
  ## Currently, "timestamp without time zone" and "timestamp with time zone"
  ## are supported
  # timestamp_column_type = "timestamp without time zone"

  ## Templated statements to execute when creating a new table.
  # create_templates = [
  #   '''CREATE TABLE {{ .table }} ({{ .columns }})''',
  # ]

  ## Templated statements to execute when adding columns to a table.
  ## Set to an empty list to disable. Points containing tags for which there is no column will be skipped. Points
  ## containing fields for which there is no column will have the field omitted.
  # add_column_templates = [
  #   '''ALTER TABLE {{ .table }} ADD COLUMN IF NOT EXISTS {{ .columns|join ", ADD COLUMN IF NOT EXISTS " }}''',
  # ]

  ## Templated statements to execute when creating a new tag table.
  # tag_table_create_templates = [
  #   '''CREATE TABLE {{ .table }} ({{ .columns }}, PRIMARY KEY (tag_id))''',
  # ]

  ## Templated statements to execute when adding columns to a tag table.
  ## Set to an empty list to disable. Points containing tags for which there is no column will be skipped.
  # tag_table_add_column_templates = [
  #   '''ALTER TABLE {{ .table }} ADD COLUMN IF NOT EXISTS {{ .columns|join ", ADD COLUMN IF NOT EXISTS " }}''',
  # ]

  ## The postgres data type to use for storing unsigned 64-bit integer values (Postgres does not have a native
  ## unsigned 64-bit integer type).
  ## The value can be one of:
  ##   numeric - Uses the PostgreSQL "numeric" data type.
  ##   uint8 - Requires pguint extension (https://github.com/petere/pguint)
  # uint64_type = "numeric"

  ## When using pool_max_conns>1, and a temporary error occurs, the query is retried with an incremental backoff. This
  ## controls the maximum backoff duration.
  # retry_max_backoff = "15s"

  ## Approximate number of tag IDs to store in in-memory cache (when using tags_as_foreign_keys).
  ## This is an optimization to skip inserting known tag IDs.
  ## Each entry consumes approximately 34 bytes of memory.
  # tag_cache_size = 100000

  ## Enable & set the log level for the Postgres driver.
  # log_level = "warn" # trace, debug, info, warn, error, none

Input and output integration examples

Nvidia SMI

1. Real-Time GPU Monitoring for ML Training: Continuously monitor the GPU utilization and memory usage during machine learning model training. This enables data scientists to ensure that their GPUs are not being overutilized or underutilized, optimizing resource allocation and reviewing performance bottlenecks in real-time.

2. Automated Alerts for Overheating GPUs: Implement a system using the Nvidia SMI plugin to track GPU temperatures and set alerts for instances where temperatures exceed safe thresholds. This proactive monitoring can prevent hardware damage and improve system reliability by alerting administrators to potential cooling issues before they result in failure.

3. Performance Baselines for GPU Resources: Establish baseline performance metrics for your GPU resources. By regularly collecting data and analyzing trends in GPU usage, organizations can identify anomalies and optimize their workloads accordingly, leading to enhanced operational efficiency.

4. Dockerized GPU Usage Insights: In a containerized environment, use the plugin to monitor GPU performance from within a Docker container. This allows developers to track GPU performance of their applications in production, facilitating troubleshooting and performance optimization within isolated environments.

PostgreSQL

1. Real-Time Analytics with Complex Queries: Leverage the PostgreSQL plugin to store metrics from various sources in a PostgreSQL database, enabling real-time analytics using complex queries. This setup can help data scientists and analysts uncover patterns and trends, as they manipulate relational data across multiple tables while utilizing PostgreSQL’s robust query optimization features. Specifically, users can create sophisticated reports with JOIN operations across different metric tables, revealing insights that would typically remain hidden in embedded systems.

2. Integrating with TimescaleDB for Time-Series Data: Utilize the PostgreSQL plugin within a TimescaleDB instance to efficiently handle and analyze time-series data. By implementing hypertables, users can achieve greater performance and partitioning of topics over the time dimension. This integration allows users to run analytical queries over large amounts of time-series data while retaining the full power of PostgreSQL’s SQL queries, ensuring reliability and efficiency in metrics analysis.

3. Data Versioning and Historical Analysis: Implement a strategy using the PostgreSQL plugin to maintain different versions of metrics over time. Users can set up an immutable data table structure where older versions of tables are retained, enabling easy historical analysis. This approach not only provides insights into data evolution but also aids compliance with data retention policies, ensuring that the historical integrity of the datasets remains intact.

4. Dynamic Schema Management for Evolving Metrics: Use the plugin’s templating capabilities to create a dynamically changing schema that responds to metric variations. This use case allows organizations to adapt their data structure as metrics evolve, adding necessary fields and ensuring adherence to data integrity policies. By leveraging templated SQL commands, users can extend their database without manual intervention, facilitating agile data management practices.

Feedback

Thank you for being part of our community! If you have any general feedback or found any bugs on these pages, we welcome and encourage your input. Please submit your feedback in the InfluxDB community Slack.