The dbt Fusion auto-switch holds when it can't tell, and offers a way back2026-11-20
未设置 dbt.dbtIntegration 时,Power User 只有在确认所选 Python 解释器没装 dbt-core、或根本没有解释器时,才会把窗口切到 dbt Fusion;检查无法判定,或项目的 dbt profile 用了 Fusion 不支持的适配器(如 sqlserver、athena、oracle)时,窗口保持 dbt Core 模式并说明原因,仍可一键切换。自动切换后若 Fusion 解析失败,会提供「切回 dbt Core」或「保留 dbt Fusion」,每个工作区只提示一次。Notebook 打开时,SQL 单元格会直接显示存放结果的 Python 变量 cell_<id>['data']。讨论中 @ 提及显示名为多词的同事现在能正确标记。桌面版 VS Code 和 Cursor 的浏览器登录恢复可用。附带 @altimateai/dbt-integration 0.3.21:Fusion 与 corecommand 模式下查询预览的列带类型,数字可按值排序筛选,大数保持精确。
已知问题:Fusion 模式下经 DataPilot 集成「defer to production」每个项目警告一次且不生效,本地 state 路径可用;Better Jinja 仍安装时 dbt SQL/YAML 的 Toggle Line Comment 会插入 {# #},卸载即可恢复。
dbt Fusion without the Python extension, automatic Fusion mode, and docs on dbt Fusion2026-11-18
dbt Power User 0.70.0:Fusion 项目不再依赖 Python 扩展。未配置 dbt.dbtIntegration 时,dbt 2.x 项目会自动以 Fusion 模式打开;同时装了 dbt Core 与 Fusion 的用户只会收到一次提示,一键即可切换。若所选 Python 解释器已被删除或移动,插件会自动找到能导入 dbt 的解释器供当前窗口使用。对已保存模型执行「Execute SQL」预览时,会按真实节点运行,this 与 model.config 能正确解析。扩展改为自带 Jinja 语言支持,不再安装 Better Jinja,各平台安装包体积因此缩减约一半。Query Results 修复了新窗口首次预览不显示、提前关闭标签页后停止刷新等问题,并会显示每条查询的实测耗时。随附 @altimateai/dbt-integration 0.3.20:在 dbt Fusion 2.0.9 及以上版本中,文档生成与分享可用,且探测时不再执行已删除的解释器。
已知问题:Fusion 模式下经 DataPilot 做「defer to production」会每个项目警告一次并跳过 defer(指向本地 state 路径仍正常);若还装着 Better Jinja,dbt SQL/YAML 的行注释会输出 {# #},卸载即可恢复。
dbt Fusion without Python, notebooks start again, and parse errors on their own line2026-11-13
dbt Fusion 模式下不再依赖 Python 解释器:解析、编译、查询预览、血缘与 Validate SQL 均可直接运行,仅笔记本和健康检查仍会提示需要解释器(健康检查暂未适配 Fusion 模式)。Altimate 笔记本的内核改为从已安装扩展启动,此前加载 zeromq 失败的问题也一并修好;Fusion 的解析错误现在会按文件和行号出现在「问题」面板里,而不是笼统报错。
内置的 @altimateai/dbt-integration 升到 0.3.19:只有报告 dbt 2.x 的 dbt 才按 Fusion 运行,dbt Cloud CLI 不再被误判;target 读取方式与 dbt 一致,改为从 profiles.yml 解析;Fusion 下的列级血缘每次最多编译 50 个模型,并支持在 Fusion 2.0.9 中定位单一 source 的列;corecommand 模式在 dbt-core 1.6、1.7 上恢复可用。
已知限制:Fusion 模式下「defer 到生产」会警告一次并无延用(指向本地 state 路径则正常),docs 生成尚不支持,新窗口首次查询预览偶尔卡在加载态,重跑一次即可。
Lineage cards that name the fix, and recovery without a reload2026-11-11
When an error stops the lineage graph, the panel now shows a card that names the problem and offers the fix, and in dbt Core and corecommand modes a project that stops parsing keeps its last good graph under an "Out of date" chip until saving the fix brings it back, with no window reload. A new "Re-parse project" command re-parses on demand, a selected Python interpreter that no longer exists gets one prompt that names it, the Get Started setup check waits for the Python environment instead of failing on a first run, and column lineage that can't be computed ends with dbt's reason instead of spinning until Cancel. Includes @altimateai/dbt-integration 0.3.18: that recovery, the dbt version, target switching and defer to production in dbt Fusion mode, dbt's v2 parser on dbt-core 1.12 when DBTENGINEUSEV2PARSER=true is set, no extension host crash on Windows when a watched folder such as target/ is deleted by dbt clean, column lineage through a source in corecommand mode again, and interpreter switches without a logged NameError.
Known issues
- In dbt Fusion mode, defer to production through the DataPilot dbt integration warns once per project and runs without deferral. Defer to a local state path works.
- On dbt Fusion 2.0.9, looking up a single source's columns returns none. It now says so in an error instead of passing on Fusion's warning.
- Altimate notebooks still can't start their kernel in the installed extension; a fix is coming in the next release.
What's Changed
* feat: mark a lineage graph drawn from an out-of-date manifest by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2123
* fix: stop the capped dbt message in the out-of-date popover from flickering by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2130
* fix: name a missing Python interpreter and offer to pick another by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2116
* feat: re-parse a dbt project without reloading the window by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2121
* fix: end the column trace when column lineage fails, and say why by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2132
* fix: open the file from the out-of-date popover after a startup failure by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2133
* fix: open the project's file from the out-of-date popover after a rename dbt rejects by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2136
* feat: show a card with the fix when an error stops the lineage graph by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2134
* fix: make the Get Started setup check wait for the Python environment by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2137
* fix: start the notebook kernel and show SQL cell rows again by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2138
* bump: 0.68.0 (rolling release) by @ralphstodomingo in https://github.com/AltimateAI/vscode-dbt-power-user/pull/2131
Full Changelog: https://github.com/AltimateAI/vscode-dbt-power-user/compare/0.67.0...0.68.0
New onboarding flows2026-11-05
dbt Power User 0.67.0 上线,这次既有功能升级也有不少稳定性修补。
功能上,引导向导(onboarding wizard)全面改版,登录流程改为通过 SaaS 完成;Web 编辑器(如 Codespaces、VS Code Web)里也能走浏览器登录并关联工作区了。血缘图报错信息更清楚:会话遇到哪类 lineage 错误会明确提示一次,图渲染完成也会告知。
修复方面:激活流程中扩展上下文交接时机提前,避免首个 await 前丢上下文;source-model 代码镜头改用 for...of 遍历;包安装路径未知时跳过陈旧列检查,不再误报;dbt-loom 探测失败不再连带让项目构造函数抛错;查询结果载荷加上大小上限,查询历史也不再反复序列化,内存占用更稳。
Lineage across interpreter switches, Fusion 2.0 detection, compiled preview of the real model2026-10-29
本次更新引入 @altimateai/dbt-integration 0.3.16,主要修补一批日常使用中的稳定性问题:切换 Python 解释器后,大项目的血缘、跳转定义和悬浮提示不再失效;dbt Fusion 2.0 的发布包会被正确识别为 Fusion,其解析警告与错误现在会列在 Problems 面板里;报错「dbt configuration is invalid」也会指明出问题的文件和行号,不用再靠猜。
功能上,已保存模型的编译 SQL 预览改为真正编译该模型本身,model.name、this、model.config 的取值与 dbt compile 一致,之前预览结果对不上根节点的尴尬随之解决。文档生成失败时会给出原因,不再默默无反应;「新功能介绍」页改为等 feed 加载完成再打开,走代理的用户不会再撞上开场报错。
已知问题:项目初始化期间切换解释器,dbt 日志可能记一条错误,无弹窗提示,初始化会自行完成。
restore the dbt file icon for jinja-sql files2026-10-23
v0.66.2 只有一个修复:恢复 jinja-sql 文件的 dbt 图标显示。此前编辑器里这类文件的图标丢失,现在重新正常展示,方便快速识别。
update the Marketplace description and docs home page2026-10-21
0.66.1 是一次纯文档修订:全站统一采用「Power User for dbt」作为产品名称,并同步更新了 Marketplace 描述与文档首页。功能层面没有变化,升级与否不影响日常使用。
rebrand to Power User for dbt2026-10-21
0.66.0 版本主要做了两件事:扩展正式更名为「Power User for dbt」,图标里保留的 dbt logo 也一并移除,品牌呈现与官方标识彻底脱钩;同时把 aireceipts-cli 及其整条依赖链锁定到经过审核的发布版本,避免依赖自动升级带来不可控的变更。功能层面没有实质改动,属于品牌调整与依赖治理的一次例行更新。
dbt-integration 0.3.15: Python bridge shutdown fixes2026-10-14
0.65.1 是一个修复版本,核心是解决 Python bridge 关闭不干净的问题:现在关闭时会等待进程真正退出,确保项目文件和数据库锁在新 bridge 启动前已释放,避免替换过程中出现文件占用或锁冲突。命令强制清理也更严谨——只有被 dispose 终止的命令才会被强杀,杀不掉的会记录日志而不是静默忽略。集成包可以从单文件可执行构建中启动 bridge。
依赖升级至 @altimateai/dbt-integration 0.3.15。内部方面,错误遥测新增 errorClass 字段,可直接判断故障归属(dbt 项目、Python 环境还是数据仓库),排查报错不用再逐条读消息。