ClickHouse Output

Writes data to ClickHouse tables via the HTTP API using INSERT FORMAT JSONEachRow.

How it works

1. Receives Arrow RecordBatch streams.
2. Auto-creates tables with CREATE TABLE IF NOT EXISTS using Nullable MergeTree columns.
3. Converts each batch row to JSON and inserts via the ClickHouse HTTP bulk insert endpoint.
4. Table name defaults to the namespace (dots replaced with underscores) or can be overridden.

Configuration

data_sinks:
  sink:
    Clickhouse:
      url: "http://localhost:8123"
      database: default
      user: default
      password: secret
      table: my_table

Configuration variables

Variable	Default	Description
url		ClickHouse HTTP endpoint
database		Database name
user		Username
password		Password
table	(from namespace)	Target table name
format	json	Data format

Authentication

Use the configured ClickHouse user and password, or rely on the default local user for development. For security best practices, we strongly advise against storing the password in skippr.yml. Use environment variable interpolation instead: replace the password value with your own ${ENV_VAR} reference.

The relevant part of skippr.yml looks like this:

data_sinks:
  warehouse:
    Clickhouse:
      password: "${CLICKHOUSE_PASSWORD}"

Set the env var before running skipprd:

macOS / Linux

export CLICKHOUSE_PASSWORD="secret"

Windows PowerShell

$env:CLICKHOUSE_PASSWORD = "secret"

Windows Command Prompt

set CLICKHOUSE_PASSWORD=secret

Troubleshooting

Symptom	Fix
authentication failed	Verify the ClickHouse user, password, URL, and database name.
writes succeed but deduped state looks delayed	This destination relies on ReplacingMergeTree merges. Use FINAL for point-in-time correctness when querying fresh data.