Canonical: https://molo17.com/blog/gluesync-clickhouse-target-agent-replacingmergetree/ Markdown mirror: https://molo17.com/blog/gluesync-clickhouse-target-agent-replacingmergetree/index.md Title: New ClickHouse target agent in Gluesync 2.3 · MOLO17 Description: Replicate to ClickHouse with the new Gluesync 2.3 target agent: versioned ReplacingMergeTree writes, tombstone deletes, FINAL reads, every data type mapped. [← All articles](/blog/) Article # New in Gluesync 2.3: the ClickHouse target agent  15 Oct 2026 [Gluesync](/blog/?category=gluesync) 4 min read Gluesync 2.3 adds a new ClickHouse target agent. Every change lands as a new version of its key in a ReplacingMergeTree(\_version, \_is\_deleted) table, deletes become tombstones, and Query Studio reads the current row with FINAL. Gluesync 2.3 adds a new **ClickHouse target agent**. Every change lands as a new version of its key in a `ReplacingMergeTree(_version, _is_deleted)` table that Gluesync generates, written in multi-row `INSERT … VALUES` batches over HTTP/S JDBC. Deletes become tombstones. Query Studio and the Validator read with `final = 1`, so you see the current row of each key. ClickHouse is a target in Gluesync. Any supported source feeds it through a pipeline in Core Hub: snapshot first, then the source agent's CDC stream. The agent runs against any ClickHouse deployment, on-prem or DBaaS on Amazon AWS, Google Cloud or Azure. Docs: [ClickHouse agent](https://docs.molo17.com/gluesync/latest/agents/clickhouse-intro.html) · [Target setup](https://docs.molo17.com/gluesync/latest/agents/clickhouse-target.html) · [Changelog](https://docs.molo17.com/gluesync/latest/agents/clickhouse-agent-changelog.html). Overview: [Replicate to ClickHouse with Gluesync](https://molo17.com/solutions/replicate-to-clickhouse/). ## What the agent does - Writes every change as a new version of its key in a generated `ReplacingMergeTree(_version, _is_deleted)` table, in multi-row `INSERT … VALUES` batches over HTTP/S JDBC, with deletes written as tombstones - Reads with `final = 1` in Query Studio and the Validator - Writes user-owned tables of another shape in append mode, with a warning - Maps every Gluesync data type ## Why versions and tombstones ClickHouse is built for appends. Row-by-row `UPDATE` and `DELETE` fight that design. So the agent writes the way ClickHouse likes to be written: - An insert or an update is a new row with a new `_version` and `_is_deleted = 0`. - A delete is a tombstone: the old key (the before-image when the source sends one), a new `_version`, and `_is_deleted = 1`. - `FINAL` shows the last version of each key and hides tombstones. `_version` is epoch nanoseconds from the agent, strictly increasing within the agent, and the connector writes it on every row. The generated column also declares `DEFAULT toUnixTimestamp64Nano(now64(9))` for any `INSERT` that leaves it out. Your CDC stream becomes a clean, queryable history in ClickHouse, with the current state one `FINAL` away. ## Tables Gluesync generates | Source table | Generated ClickHouse table | | --- | --- | | Has keys | `ReplacingMergeTree(_version, _is_deleted)`, ordered by those keys, plus `_version` (UInt64) and `_is_deleted` (UInt8) | | No keys | `MergeTree ORDER BY tuple()` | `_version` and `_is_deleted` are reserved names on a generated keyed table. Keyless tables stay on `MergeTree`, because `ReplacingMergeTree` would collapse them to one row. ## Tables you create Bring your own table and the agent adapts to its engine: - **`ReplacingMergeTree`, `ReplicatedReplacingMergeTree` or `SharedReplacingMergeTree`** whose last two arguments are `_version` and `_is_deleted`: written with versions and tombstones. - **Any other engine:** append mode. Updates are appended as new rows, deletes are lightweight `DELETE` mutations that match the old key. A keyed table in append mode raises a warning the first time it is written, since it keeps every version of a row. ## Write path - Batches are multi-row `INSERT … VALUES`. **Bulk insert batch size** (`maxWriteBatchSize`) defaults to 10000 per chunk. - The transaction is written in its own order. The connector deliberately skips grouping by operation, since that could delete a row inserted again later in the same transaction. That is also why bulk load mode is off for this agent. - **Delete mutation batch size** (`maxDeleteBatchSize`) defaults to 1000 and applies to append-mode tables. - Snapshots populate ClickHouse tables for initial loads, and the agent honors the Core Hub "TRUNCATE before snapshot" option. Turn it off per target with `disableTruncateOnSnapshot`. - Connection pooling is supported across pipelines. Optional switches: HTTP compression (on by default), ClickHouse async inserts (`asyncInsert`, off by default, with a `flushIntervalMs` of 1000 when enabled). Leave `useRowBinary` off: the write path is the textual `INSERT … VALUES` batch. ## Data types The agent maps every Gluesync type. A few defaults worth knowing: | Gluesync type | Default ClickHouse type | | --- | --- | | `BIG_DECIMAL` | `Decimal(P, S)`, up to 76 digits. Unsized is `Decimal(38, 9)` | | `LOCAL_DATE` | `Date32` | | `LOCAL_TIME` | `Time64(6)` | | `LOCAL_DATE_TIME` | `DateTime64(p)`, p from the source precision | | `OFFSET_DATE_TIME` | `DateTime64(6, 'UTC')` | | `ARRAY` | `Array(Nullable(T))` | | `MAP` | `Map(String, Nullable(String))` | Array and Map can't be nullable in ClickHouse, so a NULL array or map is stored as `[]` or `{}`. The full table, with alternatives per type, is in the [target setup docs](https://docs.molo17.com/gluesync/latest/agents/clickhouse-target.html). ## Connection setup In the Control Plane, add ClickHouse as the **target** of a pipeline: - **Hostname / IP Address:** your ClickHouse HTTP endpoint - **Port:** defaults to `8123` - **Database name**, **Username**, **Password** Credentials need read and write access to the target tables and database: the agent inserts rows, truncates tables, reads metadata and creates target tables when Gluesync generates them. `enableTls` switches the JDBC URL to https. `trustServerCertificate` accepts self-signed certificates. Prefer the API? The same connection goes through the Core Hub REST endpoint: ```bash curl --location --request PUT 'http://core-hub-ip-address/pipelines/{pipelineId}/agents/{agentId}/config/credentials' \ --header 'Content-Type: application/json' \ --header 'Authorization: ••••••' \ --data '{ "hostCredentials": { "connectionName": "myAgentNickName", "host": "host-address", "port": 8123, "databaseName": "db_name", "username": "", "password": "", "maxConnectionsCount": 25, "enableTls": false, "trustServerCertificate": false } }' ``` ## Where it fits Operational databases on one side, ClickHouse analytics on the other. Pick any Gluesync source, IBM i, Oracle, SQL Server, PostgreSQL, Couchbase and the rest of the catalog, and land its snapshot and CDC changes in ClickHouse as versioned rows ready for `FINAL` queries. The full list of sources and targets lives on the [Gluesync](https://molo17.com/gluesync/) page. ## Next steps 1. Read the [ClickHouse target setup](https://docs.molo17.com/gluesync/latest/agents/clickhouse-target.html). 2. Deploy **Gluesync 2.3** and add ClickHouse as the target of a pipeline in Core Hub. 3. Query with `FINAL` and watch each key resolve to its latest version. Want to see it on your data? [Request a free trial](https://molo17.com/get-gluesync/) or [book a demo](https://molo17.com/book-a-demo/). Questions on a concrete build: I am around. [← Older article Gluesync 2.3.0: AI Studio, Enterprise brain, and a Fields Editor you draw](/blog/gluesync-2-3-0-ai-studio-enterprise-brain-fields-editor/) ## Keep reading 1. [Gluesync Query Forge: federated SQL across agents, without a warehouse One JDBC URL into Core Hub. Schemas map to the agents you already run. Cross-agent joins without standing up a warehouse. 24 Sep 2026](/blog/query-forge-federated-sql-why-when-what-how/) 2. [Gluesync Gluesync Couchbase DCP source agent: native CDC without Eventing If you’ve ever wanted Couchbase CDC without standing up Eventing, Gluesync 2.2.11.2 adds that path: a source-only DCP agent. No Eventing service, nothing written to the source bucket, CE and EE including Capella. It sits next to the Eventing source, it doesn’t replace it. 17 Sep 2026](/blog/gluesync-couchbase-dcp-source-agent/) 3. [Gluesync Gluesync Core Hub MCP server: pipelines, SQL, and ops from your AI client Embedded Model Context Protocol server inside Core Hub: 55 RBAC-safe tools, SSE and streamable HTTP, dynamic Query Studio database tools. Available in Core Hub 2.2.10.x. 9 Sep 2026](/blog/gluesync-core-hub-mcp-server/) [Back to all articles](/blog/) [More in Gluesync →](/blog/?category=gluesync)