Ship OTel data to ClickHouse

create_schema: true runs CREATE DATABASE and CREATE TABLE on every start, which is how a fresh ClickStack comes up with its tables; once they exist, set it to false so a TTL or engine change is a deliberate migration rather than whatever the next collector start applies.

The json switch (HYPERDX_OTEL_EXPORTER_CLICKHOUSE_JSON_ENABLE) makes the exporter create tables with the JSON column type instead of Map, and the exporter never alters tables that already exist, so decide before the first start against a database, not after.

The routing/logs match is IsMatch(attributes["rr-web.event"], ".*"), so any log record that carries an rr-web.event attribute, whatever its value, is treated as a session-replay event and skips the parsing transform.

Severity inference only runs on records that arrive with severity_number 0 and an empty severity_text, and the body scan only reads the first 256 characters, so a record whose only level marker sits deep in a long line is filed as info.

The JSON body parse merges with the upsert strategy, so a key inside the body overwrites an attribute the SDK already set with the same name, and flatten turns nested objects into dotted keys.

There is no file_storage extension and the clickhouse exporter runs on the default in-memory sending queue, exactly as upstream ships it; a ClickHouse outage longer than the retry window (max_elapsed_time: 300s) drops data unless you add a persistent queue or put the Fail over to S3 flow in front of it.

The otlp/hyperdx receiver allows every origin and header over HTTP because HyperDX's browser SDK posts straight to it; if only servers send to this collector, tighten cors before exposing :4318.

Metrics support in the clickhouse exporter is still alpha (traces and logs are beta), so treat the otel_metrics_* table schemas as subject to change between collector upgrades.