Skip to content
TensorDB

Time Series

TensorDB’s time series facet provides efficient storage and querying of timestamped metric data.

Storing Metrics

use tensordb::Value;


// Store a metric data point
db.put_temporal(
    "metrics/cpu/host-1",
    Value::Real(72.5),
    1718409600,  // valid_from: timestamp
    1718409660,  // valid_to: +60s (1-minute bucket)
)?;

Bucketed Aggregation

Query metrics with time buckets:

-- Average CPU per 5-minute bucket
SELECT
    (valid_from / 300) * 300 as bucket,
    AVG(CAST(value AS REAL)) as avg_cpu
FROM __keys__
WHERE key LIKE 'metrics/cpu/%'
GROUP BY bucket
ORDER BY bucket;

Use Cases

  • Infrastructure monitoring: CPU, memory, disk usage over time
  • Application metrics: Request latency, error rates, throughput
  • IoT data: Sensor readings with temporal bucketing
  • Financial data: Price history with business-time semantics

Bitemporal Advantage

Unlike traditional time-series databases, TensorDB’s bitemporal model lets you:

  • Backfill data: Write historical metrics without disrupting the timeline
  • Correct errors: Write a correction for past data while preserving the original
  • Audit changes: See what metrics were known at any point in system time