Skip to content

GreptimeDB Enterprise

Metrics, logs, and traces in one database

Build your observability data platform on object storage. Query telemetry with SQL, connect it to your analytics stack, and make it available to engineers and AI agents.

Data path for
  • write path
  • data at rest
  • read path
SourcesReadersMetricsRemote Write · OTLPLogs and eventsFluent Bit · Loki · KafkaTracesOTLP · OTel CollectorGreptimeDB Enterpriseeach tier scales on its ownFrontendingest protocolsquery planningstatelessDatanodestorage engineindexesFlow rollupsflush Parquetread on queryYour object storageS3 · GCS · Azure Blob · OSSmetrics/9f3a.parquetlogs/41c7.parquettraces/e80b.parquetindex/41c7.puffinGrafanaPromQL · SQL · JaegerAgentsMCP Server · SQLSpark · Trino · DuckDBParquet · Iceberg RESTdirect readMetricsLogs, eventsTracesGreptimeDB EnterpriseFrontendprotocolsquery planningstatelessDatanodestorage engineindexesFlow rollupsflushreadYour object storageS3 · GCS · Azure Blob · OSSmetrics/9f3a.parquetlogs/41c7.parquettraces/e80b.parquetindex/41c7.puffinLakehouseSpark · TrinoGrafanaPromQL · SQLAgentsMCP · SQLdirect read
Trusted by
  • OpenAI
  • OceanBase
  • Xiaomi
  • JD.com
  • Poizon

Workloads

Two workloads, one copy of the data

Keep every high-cardinality series at full resolution, for as long as you need it.

  • Prometheus agents and Grafana dashboards keep working, PromQL included.
  • Raw samples live in your own bucket as Parquet.
  • Query the same data with SQL, Spark, or Trino.
Prometheus Remote WriteOpenTelemetryInfluxDB line protocolGrafana Alloy
Discuss a metrics workload

One table, two query languages

PromQL
sum by (namespace) (
  rate(container_cpu_usage_seconds_total[5m])
)
SQL
SELECT namespace, approx_distinct(pod) AS pods
FROM container_cpu_usage_seconds_total
WHERE greptime_timestamp > now() - INTERVAL '1' DAY
GROUP BY namespace;
Try GreptimeDB in the live demo

Scale and cost

Scale compute and storage separately

Data lives in object storage. You add nodes for throughput, not for disk.

  • Storage in your bucket

    Data and indexes sit in S3, GCS, Azure Blob, or OSS. Nodes keep only a cache.

  • Scale writes and queries independently

    Add ingest or query capacity on its own. Enterprise repartitions and rebalances automatically.

  • Fast on cheap storage

    Indexes and a local cache keep queries fast over object storage.

Compression

Up to 30×

Raw logs
In GreptimeDB
See the log benchmark

Get a cost estimate

Tell us your daily volume, retention, and where you deploy. An engineer sizes the cluster and the monthly cost.

Request an estimate

Analytics and agents

Open to your analytics stack and your agents

One copy of the data serves dashboards, SQL, lakehouse engines, and agents.

SQL and PromQL

SQL over MySQL, PostgreSQL, or HTTP. PromQL for metrics.

Open formats

Parquet in your bucket. Spark, Trino, and DuckDB read it through the Iceberg REST catalog, no copy.

Agent access

Agents query through the GreptimeDB MCP Server with read-only SQL and PromQL.

What stays in your stack

Your collectors, Grafana, Alertmanager, and lakehouse tables.

Customers

In production

Spark metrics, events, and logs

“…just because we found that it was more effective.”

OpenAI data platform engineers, on moving Spark observability to GreptimeDB

Per-task Spark metrics add millions of new series a day. OpenAI moved Spark metrics, events, and logs into GreptimeDB, read by Grafana, a Rust Spark History Server, and an internal agent.

  • Replaced what the team called one of OpenAI's most expensive Flink jobs
  • Handles per-task cardinality their existing systems couldn't

From two OpenAI talks at Apache Community Over Code EU 2026.

Read the talk recap

Multi-cloud logs, migrated from Loki

80+
GreptimeDB clusters
300 TB
Logs and SQL audit data
~1 GB/s
Sustained writes
60%+
Lower log storage cost after leaving Loki

One GreptimeDB cluster per public cloud, on that cloud's object storage. Searches that timed out on Loki now come back in about a second.

Read the case study

Evaluation

How an evaluation runs

One workload, clear pass criteria, real traffic.

  1. 1

    Pick one workload

    A metric source or log stream with a known pain.

  2. 2

    Define data and queries

    Volume, retention, and the queries that must keep working.

  3. 3

    Set pass criteria

    Throughput, query latency, storage, and monthly cost.

  4. 4

    Run side by side

    Mirror real traffic and compare with what you run today.

Talk to an engineer

Describe your workload

An engineer reads your request and replies within one business day.

You agree on one workload and its pass criteria.You run GreptimeDB Enterprise on your own data.

Prefer to try it first? Open the live demo or download GreptimeDB. Compare open source and Enterprise.