Table-model support for the Flink connectors: concrete design and three open questions

Bắt đầu bằng cách build và chạy cả hai Flink connector hiện có, sau đó đọc IoTDBSinkFunction xung quanh phần parsing path được trích dẫn và các lệnh gọi Session. So sánh các cây Spark connector song song và kiểm tra các path source, CDC và lookup trước khi quyết định liệu một table connector riêng và một triển khai sink-first có phù hợp hay không. Công việc được xem là hoàn tất khi ba câu hỏi về thiết kế đã có câu trả lời và hình thức triển khai đã được thống nhất.

Concrete design: table-model support for the Flink connectors

Follow-up to [DISCUSS] Table-model support for the Flink connectors on
dev@iotdb.apache.org (2026-08-08). That thread received no replies, so what
follows is the design I proposed there written out concretely. The three
questions I asked on the list are still open, and I have kept them open here
rather than treating silence as agreement on any of them.

The problem

flink-sql-iotdb-connector's schema mapping is the tree model, not a
configuration of it. In IoTDBSinkFunction a Flink column name is parsed as an
IoTDB path and split into a device and a measurement:

:132-136  PathUtils.splitPathToDetachedNodes(fieldName);
          measurement = nodes[nodes.length - 1];
          device      = join(copyOfRange(nodes, 0, nodes.length - 1), '.');
:108-113  session.insertAlignedRecord(...) / session.insertRecord(...)
:86       new Session.Builder().nodeUrls(..).username(..).password(..).build()

In table mode there is no path to split. A column is a TAG, FIELD or ATTRIBUTE
under database.table, and which of the three it is carries meaning a name
cannot express. The connector's option list agrees that this is not a
configuration gap: there is no database and no dialect option, and aligned
and cdc.pattern are tree concepts.

Proposed shape

A new module flink-iotdb-table-connector, leaving flink-sql-iotdb-connector
untouched, mirroring how this repository already split Spark:
spark-iotdb-connector and spark-iotdb-table-connector are parallel trees with
their own parent poms, and the table one has its own spark-iotdb-table-common
rather than sharing the tree one's.

The objection to a separate module is duplicated CDC, lookup and bounded-scan
machinery. That objection applied equally to the Spark split and the project
accepted it there, so the cost is one this repository has already weighed for
this exact problem.

Three questions that are still open

The DISCUSS thread drew no replies. Lazy consensus covers "nobody objected to
the direction"; it does not answer these, and one of them rests on reading I
explicitly flagged as incomplete.

1. Is a separate flink-iotdb-table-connector the right shape here? The
   Spark precedent is the argument for it, but Spark's split may have had
   reasons that do not carry over.

2. Is the sink the right place to start? My reading was that the source's
   tree couplings are the dialect-less Session and the TIME clause — a
   smaller and different problem from a mapping with no table-mode analogue —
   which would put the design work in the sink. I have not read the CDC or
   lookup paths. If the source has couplings I have not found, this ordering
   is wrong and I would rather know before writing code than after.

3. Is anyone already working on this? I searched issues and pull requests in
   both iotdb-extras and iotdb and found nothing on Flink and the table
   model, but a search is not the same as asking.

What I plan to do next

Subject to the above: build and run both existing Flink connectors first. My
DISCUSS post was explicit that everything in it came from reading source and
that I had run neither connector. That is a reasonable basis for proposing a
shape; it is not a reasonable basis for implementing one, so running them is the
first step rather than a later one.

I am happy to take the implementation if the direction holds, and equally happy
to hand the design to whoever is better placed to do it.