[CBRD-27336] Support a remote REPLACE ... SELECT whose source is a local table by kwangsoochae · Pull Request #7985 · CUBRID/cubrid

kwangsoochae · 2026-09-20T11:42:57Z

http://jira.cubrid.org/browse/CBRD-27336

Purpose

원격 테이블 대상 REPLACE 는 리터럴 형태만 동작하고(문장이 그대로 원격에 전달된다), 소스가 로컬 테이블이면
파서가 거절했다.

REPLACE INTO remote_t@srv (id, name) SELECT id, name FROM local_t;
-- dblink: this combination of local and remote references is not supported (...)

INSERT ... SELECT 는 로컬에서 행을 만들어 원격에 보내는 값 sink 로 로컬 소스를 받는다. 이 sink 는 소스의
행마다 원격에 한 문장씩 보내고, REPLACE 도 행 단위로 처리된다 — 같은 키가 소스에 여러 행 있으면 소스 질의
순서상 마지막 행이 남는다. 따라서 새 실행 모드 없이 sink 가 준비하는 문장을
REPLACE INTO 로 바꾸면 같은 결과에 도달한다.

Implementation

커밋 단위로 리뷰할 수 있도록 세 단계로 나눴다. 네 번째 커밋은 Remarks 첫 항목의 선행 결함
수정이고, 그 뒤는 주석 정리와 develop merge 다.

1. 원격에 보낼 문장을 조립한다

DBLINK_DML_KIND 에 REPLACE 를 더하고 문장 조립 함수가 접두만 가르게 한다. 컬럼 목록·자리표·바인드 값은
INSERT 와 같으므로, 실행기에서 INSERT_PROC 를 읽는 자리들은 두 종류가 case 를 공유한다.

INSERT INTO t (c1, c2) VALUES (?, ?)      -- AS-IS
REPLACE INTO t (c1, c2) VALUES (?, ?)     -- TO-BE (REPLACE 일 때)

문장 머리의 /* DBLINK ... */ 주석은 게이트웨이가 문장 종류로 읽는 값이고, 그 어휘에 REPLACE 가 없다.
그래서 두 종류 모두 INSERT 를 싣는다.

2. 원격이 보고한 행 수를 그대로 누적한다

INSERT sink 는 문장마다 1 을 셌지만, REPLACE 는 교체가 일어난 행에 원격이 2 를 돌려준다. 원격이 돌려주는
수를 누적하도록 바꿨다(원격 DELETE sink 와 같은 방식).

3. 게이트를 연다

sink 자격 조건에서 REPLACE 제외를 걷어내고 문장 종류를 실행 계획에 싣는다. 직렬화 포맷은 바뀌지 않는다 —
그 필드는 이미 전송되고 있었다.

소스가 로컬과 타깃 서버의 원격 테이블을 함께 읽는 형태도 함께 열린다. sink 자격을 넓히면 그대로 닿는 경로라
따로 막지 않았다.

Remarks

같은 실행 경로의 develop 선행 결함을 함께 고친다. 원격 대상 INSERT / REPLACE 의 로컬 소스가 컴파일 시점에
거짓이면(WHERE 1 = 0, 그런 조건을 정의에 가진 뷰) 클라이언트가 죽었다. develop 에서는 로컬 소스 REPLACE 가
거절돼 INSERT 에만 드러나던 결함인데, 이 PR 이 REPLACE 를 열면 그대로 상속한다. XASL 이 없는 원격 대상은
원격 DELETE 와 같이 0 행으로 끝낸다(do_execute_insert).
ON DUPLICATE KEY UPDATE 는 범위 밖이며 지금과 같이 거절한다(sink 에 갱신 대입을 실을 자리가 없다). 별도 PR 로 뗀다.
벤더 원격(게이트웨이): Oracle 은 REPLACE 구문이 없어 원격이 거절한다 — 리터럴 형태와 같은 자리, 같은 이유다.

github-actions · 2026-09-20T11:43:07Z

❌ TC Merge Gate — Merge Blocked

TC changes for this PR are not on TC develop yet.

❌ cubrid-testcases: TC PR tc/pr-7985 is open (draft) — merge it, or close it if it carries no change
❌ cubrid-testcases-private-ex: TC PR tc/pr-7985 is open (draft) — merge it, or close it if it carries no change
Steps to unblock:

Merge the TC PRs listed above. Close one only when it carries no change.
Re-run this check: Actions tab → TC Merge Gate → Re-run failed jobs

kwangsoochae · 2026-09-20T11:43:56Z

@greptileai review

github-actions · 2026-09-20T11:44:18Z

🧪 TC Test Environment Ready

CircleCI Testing:

CircleCI will automatically test using the branches below.

TC Repositories & Branches:

cubrid-testcases Draft PR (Branch: tc/pr-7985)
cubrid-testcases-private-ex Draft PR (Branch: tc/pr-7985)

Next Steps:

Wait for CircleCI tests to complete
If CircleCI tests failed, please check the test results and fix the issues.
When ready to merge this PR, please merge the TC PR first, then merge this PR.

kwangsoochae · 2026-09-20T11:44:36Z

/run all

greptile-apps · 2026-09-20T11:46:37Z

이전 리뷰 이후 변경에서 병합을 막을 새 결함이 확인되지 않아 현재 상태는 병합해도 안전해 보입니다.

_{Reviews (2) · Last reviewed commit: "Merge remote-tracking branch 'upstream/d..."}

kwangsoochae · 2026-09-20T11:52:52Z

/run all

kwangsoochae · 2026-09-22T00:23:38Z

/run shell

The sink builds one statement kind, a plain INSERT. A REPLACE that reached it would insert instead of replacing, so the parser gate refuses the shape and the statement falls through to the mixed local/remote rejection. Give DBLINK_DML_KIND a REPLACE member and let the builder switch the prefix. Nothing else about the statement changes -- REPLACE takes an INSERT's shape, so the column list, the placeholders and the values the sink binds are the same, and the two kinds share a case wherever the executor reads INSERT_PROC. The builder's own diagnostics name the statement it was asked for rather than always saying INSERT. The leading /* DBLINK ... */ comment is not a label. A CUBRID gateway reads the word after DBLINK and maps it to a statement type, refusing the statement when the word is not one it knows, and that vocabulary has no REPLACE. Both kinds carry INSERT there and the statement itself says which one it is. The gate stays shut: do_replace never reaches the plan yet, so the executor still picks INSERT. The commit that opens it comes last, so no revision in between sends the wrong statement to a remote.

The INSERT sink counted one row per statement it sent, because a positional INSERT of one row was expected to affect exactly one. A REPLACE breaks that: the remote counts the row it removed as well as the one it inserted, so a row that replaces answers 2. Read affected_rows and add it, the way the DELETE sink already does. The text-push path reports the remote's number as it stands, so the two paths now answer alike for the same statement. This also changes what a plain INSERT sink reports if a remote ever answers with something other than 1. The old comment named that case and left it out of scope; it was not reproduced on the remotes the INSERT sink was measured against, so nothing observed changes here. The callee's header no longer offers the INSERT sink as the example of a caller that ignores the count, and the comment naming which statements reach cci_execute is dropped rather than extended -- what runs there is whatever the prepare built.

The gate admitted only a plain INSERT ... SELECT, so a REPLACE whose source was local fell through to the mixed local/remote rejection. The sink can send the statement now, and the plan carries which one it is, so the shape has somewhere to go. Drop do_replace from the gate's exclusions and set it on the remote sink's plan. ODKU stays out: the sink has nowhere to carry the update assignments yet. This also admits a source that mixes a local table with a remote one on the target's own server. case PT_INSERT counts sub_sel_server_cnt whatever the sink kind, so the same-server conversion those sources need is already wired -- holding the shape back would mean adding a guard to refuse what otherwise works. The comment that deferred both clauses quoted a rejection message CBRD-26921 has since replaced, and is rewritten to say what the gate now admits.

kwangsoochae · 2026-09-22T11:01:19Z

/run all

do_prepare_insert () marks a source query with a compile-time false WHERE by leaving xasl_id NULL: there is nothing to execute. do_execute_insert () skipped that check for a remote target, so a remote INSERT or REPLACE with a local source that reduces to false -- WHERE 1 = 0, or a view defined with such a condition -- asked the server to run a NULL XASL and the client died. Return with no rows when a remote target has no XASL, as do_execute_delete () already does for false where or a statement that did not prepare correctly. The do_insert () fallback stays with local targets, as CBRD-24867 left it.

…k_replace_local_source

kwangsoochae · 2026-09-23T03:55:55Z

/run all

The comment over the remote INSERT gate ended by calling the mixed local/remote rejection "the same behavior develop gives for it". Once this lands, develop is this code, so the phrase stops saying anything. The comment in do_execute_insert () named CBRD-24867 as the origin of the local-only do_insert () fallback. That ticket gives no reason for it, so the number tells a reader no more than git blame does.

…k_replace_local_source

kwangsoochae · 2026-09-23T04:23:26Z

/run all

kwangsoochae · 2026-09-23T05:17:16Z

/run rerun 35818139527

kwangsoochae self-assigned this Sep 20, 2026

cubrid-tc-sync-bot Bot mentioned this pull request Sep 20, 2026

[CBRD-27336] Draft: TC changes for PR CUBRID/cubrid#7985 CUBRID/cubrid-testcases#3545

Draft

kwangsoochae force-pushed the CBRD-27336_dblink_replace_local_source branch from a4eb40f to 6a0cd50 Compare September 22, 2026 00:20

kwangsoochae force-pushed the CBRD-27336_dblink_replace_local_source branch from 6a0cd50 to fd49b61 Compare September 22, 2026 11:00

kwangsoochae added 3 commits September 22, 2026 20:00

kwangsoochae force-pushed the CBRD-27336_dblink_replace_local_source branch from fd49b61 to 996e1a0 Compare September 22, 2026 11:00

kwangsoochae added 2 commits September 23, 2026 12:54

Merge remote-tracking branch 'upstream/develop' into CBRD-27336_dblin…

f32d2eb

…k_replace_local_source

kwangsoochae added 2 commits September 23, 2026 13:21

Merge remote-tracking branch 'upstream/develop' into CBRD-27336_dblin…

73d6293

…k_replace_local_source

kwangsoochae requested review from airnet73, beyondykk9, childyouth, ctshim, hyunikn, jongmin-won and kangmin5505 September 23, 2026 05:40

kwangsoochae marked this pull request as ready for review September 23, 2026 05:40

beyondykk9 removed request for airnet73, childyouth, ctshim, hyunikn, jongmin-won and kangmin5505 September 27, 2026 22:49

beyondykk9 approved these changes Sep 28, 2026

View reviewed changes

Navigation Menu

[CBRD-27336] Support a remote REPLACE ... SELECT whose source is a local table - #7985

[CBRD-27336] Support a remote REPLACE ... SELECT whose source is a local table#7985
kwangsoochae wants to merge 7 commits into
CUBRID:developfrom
kwangsoochae:CBRD-27336_dblink_replace_local_source

kwangsoochae commented Sep 20, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Sep 20, 2026 •

edited

Loading

Uh oh!

kwangsoochae commented Sep 20, 2026

Uh oh!

github-actions Bot commented Sep 20, 2026

Uh oh!

kwangsoochae commented Sep 20, 2026

Uh oh!

greptile-apps Bot commented Sep 20, 2026 •

edited

Loading

Uh oh!

kwangsoochae commented Sep 20, 2026

Uh oh!

kwangsoochae commented Sep 22, 2026

Uh oh!

kwangsoochae commented Sep 22, 2026

Uh oh!

kwangsoochae commented Sep 23, 2026

Uh oh!

kwangsoochae commented Sep 23, 2026

Uh oh!

kwangsoochae commented Sep 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

kwangsoochae commented Sep 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Implementation

Remarks

Uh oh!

github-actions Bot commented Sep 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

❌ TC Merge Gate — Merge Blocked

Uh oh!

kwangsoochae commented Sep 20, 2026

Uh oh!

github-actions Bot commented Sep 20, 2026

🧪 TC Test Environment Ready

Uh oh!

kwangsoochae commented Sep 20, 2026

Uh oh!

greptile-apps Bot commented Sep 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kwangsoochae commented Sep 20, 2026

Uh oh!

kwangsoochae commented Sep 22, 2026

Uh oh!

kwangsoochae commented Sep 22, 2026

Uh oh!

kwangsoochae commented Sep 23, 2026

Uh oh!

kwangsoochae commented Sep 23, 2026

Uh oh!

kwangsoochae commented Sep 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

kwangsoochae commented Sep 20, 2026 •

edited

Loading

github-actions Bot commented Sep 20, 2026 •

edited

Loading

greptile-apps Bot commented Sep 20, 2026 •

edited

Loading