#tetcoin #telemetry #analytics #node #db #benchmark #schema

app tetcore-analytics

Tetcore Telemetry Analytics for Rust

Show the crate…

1 unstable release

0.1.4 Feb 9, 2021

#64 in #tetcoin

GPL-3.0 license

150KB
3K SLoC

Tetcore Analytics

* to connect to tetcore-analytics you must whitelist your IP address in deployment.template.yml

Comprises a websocket server accepting incoming telemetry from multiple Tetcore nodes. tetcore-analytics is designed to be resilient (to network errors), performant and horizontally scalable by deploying more servers.

Telemetry is stored in a PostgreSQL database. Management of the database schema is via diesel migrations.

Stored data is purged from the DB according to LOG_EXPIRY_H

For convenience there are also some JSON endpoints to make ad-hoc queries, although it is expected that the data is accessed directly from the database by a suitable dashboard (eg. Grafana).

Routes

Data ingestion

tetcore-analytics can work in one of two modes: with or without purging data after LOG_EXPIRY_H hours. The mode it operates under depends on which of the following two endpoints you send data to from your tetcore nodes.

  • /
    • incoming telemetry (with expiry as set by LOG_EXPIRY_H) (ws) - set with this option in tetcore cli: --telemetry-url 'ws://127.0.0.1:8080 5'
  • /audit
    • incoming telemetry with no expiry (ws) - set with this option in tetcore cli: --telemetry-url 'ws://127.0.0.1:8080/audit 5'

JSON endpoints

subtrate-analytics includes a few convenience endpoints to query for common data.

  • /stats/db
    • statistics about the postgres db, showing table and index sizes on disk
  • /nodes
    • list of logged nodes
  • /nodes/log_stats?peer_id=Qmd5K38Yti1NStacv7fjJwsXDCUZcf1ioKcAuFkq88RKtx
    • shows the quantity of each type of log message received
  • /nodes/logs?peer_id=Qmd5K38Yti1NStacv7fjJwsXDCUZcf1ioKcAuFkq88RKtx&limit=1&msg=tracing.profiling&target=pallet_babe&start_time=2020-03-25T13:17:09.008533
    • recent log messages. Required params: peer_id, Optional params: msg, target, start_time, end_time, limit.

      msg: String. Type of log message received, e.g. block.import. See ./telemetry_messages.json for the current list of message types.

      target: String. Origin of the message, e.g. NetworkInitialSync

      start_time: String. Include entries more recent than this; format: 2019-01-01T00:00:00. Default: NOW.

      end_time: String. Include entries less recent than this; format: 2019-01-01T00:00:00. Default: NOW.

      limit: Number. Don't include more results than this. Default: 100

  • /reputation/{peer_id}
    • reported reputation for peer_id from the POV of other nodes.
  • /reputation/logged
    • reported reputation for all peers from the POV of all logged (past/present) nodes
  • /reputation
    • reported reputation for all peers unfiltered (note that this can contain many entries that are not even part of the network)

reputation routes take the following optional parameters (with sensible defaults if not specified):

  • max_age_s in the format: 10
  • limit in the format: 100

Self-monitoring

Tetcore Analytics provides a /metrics endpoint for Prometheus to useful to monitor the analytics instance itself. Visit the endpoint in a browser to see what metrics are available.

Set up for development and deployment

  • Install Postgres
  • For development, create a .env file in the project root containing:
    • DATABASE_URL=postgres://username:password@localhost/tetcore-analytics
    • PORT=8080
    • any other settings from the list of environment variables below
  • Next, install Diesel cli
  • Run diesel database setup to initialise the postgres DB
  • You must diesel migration run after any changes to the database schema

Optionally specify the following environment variables:

  • HEARTBEAT_INTERVAL (default: 5)
  • CLIENT_TIMEOUT_S (default: 10)
  • PURGE_INTERVAL_S (default: 600)
  • LOG_EXPIRY_H (default: 280320)
  • MAX_PENDING_CONNECTIONS (default: 8192)
  • WS_MAX_PAYLOAD (default: 524_288)
  • NUM_THREADS (default: CPUs * 3)
  • DB_POOL_SIZE (default: NUM_THREADS)
  • DB_BATCH_SIZE (default: 1024) - batch size for insert
  • DB_SAVE_LATENCY_MS (default: 100) - max latency (ms) for insert
  • CACHE_UPDATE_TIMEOUT_S (default: 15) - seconds before timeout warning - aborts update after 4* timeout
  • CACHE_UPDATE_INTERVAL_MS (default: 1000) - time interval (ms) between updates
  • CACHE_EXPIRY_S (default: 3600) - expiry time (s) of log messages
  • ASSETS_PATH (default: ./static) - static files path

Include RUST_LOG in your .env file to make tetcore-analytics log to stdout. A good development setting is RUST_LOG = debug.

Tetcore log messages are batched together before they are sent off for storage in the postgres DB by the actor for INSERT. Batches include up to DB_BATCH_SIZE messages or DB_SAVE_LATENCY_MS, whichever is reached sooner.

Benchmarking

Tetcore-analytics has endpoints to define benchmarks and host systems that run the benchmarks. This is designed to be cross-referenced with telemetry data to provide insights into the node and system under test.

JSON endpoints:

  • /host_systems: the server machines we're benchmarking
    • GET to list all; POST to create new using the format (returns object with newly created id):
{
   "cpu_clock":2600,
   "cpu_qty":4,
   "description":"Any notes to go here",
   "disk_info":"NVME",
   "os":"freebsd",
   "ram_mb":8192
}
  • /benchmarks:
    • GET to list all, POST to create new using the format (returns object with newly created id):
{
   "benchmark_spec":{
      "tdb":"tbd"
   },
   "chain_spec":{
      "name":"Development",
      "etc": "more chain spec stuff"
   },
   "description":"notes",
   "host_system_id":2,
   "ts_end":"2019-10-28T14:05:27.618903",
   "ts_start":"1970-01-01T00:00:01"
}

Dependencies

~63MB
~1.5M SLoC