This is a Jaeger gRPC storage plugin implementation for storing traces in ClickHouse.
This is a community-driven project, and we would love to hear your issues and feature requests. Pull requests are also greatly appreciated.
ClickHouse is an analytical column-oriented database management system. It is designed to analyze streams of events which are kind of resemblant to spans. It's open-source, optimized for performance, and actively developed.
Jaeger spans are stored in 2 tables. The first contains the whole span encoded either in JSON or Protobuf. The second stores key information about spans for searching. This table is indexed by span duration and tags. Also, info about operations is stored in the materialized view. There are not indexes for archived spans. Storing data in replicated local tables with distributed global tables is natively supported. Spans are bufferized. Span buffers are flushed to DB either by timer or after reaching max batch size. Timer interval and batch size can be set in config file.
Database schema generated by JetBrains DataGrip
Refer to the config.yaml for all supported configuration options.
docker run --rm -it -p9000:9000 --name some-clickhouse-server --ulimit nofile=262144:262144 clickhouse/clickhouse-server:22
GOOS=linux make build run
make run-hotrodOpen localhost:16686 and localhost:8080.
You need to specify connection options in config.yaml, then you can run
make build
SPAN_STORAGE_TYPE=grpc-plugin {Jaeger binary adress} --query.ui-config=jaeger-ui.json --grpc-storage-plugin.binary=./{name of built binary} --grpc-storage-plugin.configuration-file=config.yaml --grpc-storage-plugin.log-level=debugThis project is originally based on this clickhouse plugin implementation.
See also jaegertracing/jaeger/issues/1438 for historical discussion regarding the implementation of a ClickHouse plugin.