An open-source observability
data lake engine.
Fast, cost-effective and scalable observability data storage, built on open standards. Logs, metrics, traces and events land in Apache Iceberg tables on your own object storage — and stay queryable by any Apache Iceberg-compatible tool.
Serves Loki®- and Tempo®-compatible read APIs, plus Apache Arrow Flight SQL
Also readable by any Apache Iceberg-compatible engine
How it works
Compute and storage, separated.
Four core components, each scaling on its own. None of them owns state — that all lives in your bucket.
Ingest
OpenTelemetry Protocol accepts observability data with exactly-once delivery. Data flows through a write-ahead log on object storage for durability.
OpenTelemetry · WAL
Query
Apache DataFusion and Arrow run fast queries across logs, metrics, traces and events — live data in the WAL and historical data in Parquet files.
DataFusion · Arrow
Maintain
Apache Parquet optimizes data with statistics, bloom filters and TTL support. Background compaction keeps query performance high.
Parquet
Alert
A rule management engine generates alerts from your observability data. Define alert rules and get notified in real time.
Rule engine
Why IceGate
Pay object storage prices.
Not observability platform prices — and keep the data in a format you can read without us.
The problem
Traditional observability platforms lock you into vendor-specific formats and expensive infrastructure. Storage costs grow linearly with data volume, and you’re forced to choose between data retention and budget. Proprietary formats mean your data is trapped.
The solution
IceGate stores all data — WAL, catalog and data layer — in object storage using open formats. Query your observability data with any Apache Iceberg-compatible tool. Scale compute independently. Pay object storage prices, not observability platform prices.
Cost-effective observability
Store telemetry on object storage instead of expensive block storage or a proprietary vendor tier. You pay your cloud provider’s storage rate, with no platform markup layered on top of it.
Multi-tenant scaling
Scale components independently based on demand. Handle spikes in one tenant without affecting others.
Long-term retention
Keep observability data for years at object storage prices. Query historical data without rehydration.
Real-time monitoring
Live data access through the WAL combined with historical queries. Get alerts in real time while analyzing trends.
Capabilities
Built for modern observability.
What the engine does, and what it refuses to hold hostage.
Highly scalable
Independent scaling of compute resources based on workload demands. Scale ingest and query separately.
ACID transactions
Full transaction support without a dedicated OLTP database. Built on Apache Iceberg and a queue based on a WAL in object storage.
Exactly-once delivery
Reliable data ingestion with no data loss or duplication. Every event is processed exactly once.
Real-time queries
Access live data through the WAL while keeping historical query capability. No lag between ingest and query.
Open standards
Built on Apache Iceberg, Arrow and Parquet with OpenTelemetry protocol support. No proprietary formats.
Cost-effective
An object-storage-based architecture minimizes infrastructure costs. Pay only for what you use.
Fault-tolerant
Designed for resilience and high availability. Automatic recovery from failures.
No vendor lock-in
Standard formats let you query with any Apache Iceberg-compatible tool. Your data, your choice.
Built with IceGate
The same engine, in production.
Who builds IceGate
IceGate is built by TripleCloud and released under Apache 2.0. Run it yourself on your own object storage — the engine is the same one we run in production.
IceGate in production
IceGate powers the TripleCloud LLM observability platform. Our engineering blog documents how the engine works — S3-only ingestion with no Kafka and no coordination layer, and how telemetry costs are shaped by it.
Powered by open source
Get started
Run it on your own bucket.
Apache 2.0, no account required. Start with the docs, or come ask in Slack.