Skip to content

v4.7.0

Latest

Choose a tag to compare

@github-actions github-actions released this 18 Nov 01:35
· 9 commits to master since this release

v4.7.0

亮点

  • 主机部署发行包(ADR-0014):每个 tag 同时发布 addax-admin-<version>-host.tar.gz 与 .sha256。发行包内含预编译后端 jar、前端静态文件、deploy/install.sh、Nginx 与 systemd 示例、BUILD-INFO 与校验文件,适配没有 Maven/Node 工具链的主机;重复执行安装脚本即为升级,保留配置、驱动、作业与日志。
  • 导航按采集生命周期重组:采集配置 / 采集运行 / 数据治理 / 系统管理,取消单孩子下拉,结构更新动作移入采集表管理页,命令面板改为直接消费导航树。
  • 停采语义修正:手工执行与补数不再悄悄把停采表(X)改回自动调度。

新增

  • 主机部署发行包及其安装脚本、部署文档(docs/manual/install/02-host.md)。
  • 导航重组(PR #135)与命令面板继承导航树(PR #136)。
  • 采集源列表固定列宽,连接串为唯一弹性列;窄视口横向滚动、操作列吸右。

变更

  • 移除 etl_table.kind:字段从未参与调度,「只手工采集」由停采状态 X 表达;启动迁移删列并重建视图,执行快照版本升到 3。
  • spring.jpa.show-sql 默认关闭,本地开发用 SPRING_JPA_SHOW_SQL=true 开启。
  • PostgreSQL JDBC 驱动升级到 42.7.12(GHSA-j92g-9f8w-j867 / CVE-2026-54291),minimatch 固定到 11.0.1(GHSA-g3hv-75wg-m249)。
  • Docker 镜像只在 tag 推送时构建;删除根目录遗留 lockfile,frontend/yarn.lock 为唯一锁文件。

修复

  • 表格详情深链 ?tab=logs、?tab=job、?tab=fillback 在详情接口返回前被重置到 info。
  • 采集状态矩阵与过期单测与当前状态标签、通知 store 对齐。

破坏性变更

  • 启动迁移删除 etl_table.kind。旧版本节点仍映射该列,必须停止所有节点后统一升级,不支持滚动升级;回退只能恢复升级前的数据库备份。
  • 历史上被手工执行悄悄解除停采的表无法自动识别,升级后需要人工复查停采名单。

升级

  • 先备份数据库;升级期间停止所有节点。
  • Docker:拉取 4.7.0 镜像并重启;空库仍由应用自动初始化。
  • 主机:下载发行包与 .sha256,校验后解压并重复执行 deploy/install.sh;旧 jar 保留为 .before-<时间戳>。
  • 已有主机安装保留原 config/application.properties,新模板写到同目录 .new。

验证

  • 后端 mvn clean package(JDK 21,单测 2/2),前端 type-check、ESLint 与生产构建通过。
  • 前端单测 64/64;history-snapshot、collection-state-matrix Playwright 用例在独立后端上通过。
  • 主机发行包 E2E 24/24:打包、校验、安装、启动、Nginx 反代、重装升级保留配置。
  • 测试环境 wgzhao-pc 已部署,启动迁移与冒烟检查通过。

v4.7.0

Highlights

  • Host deployment bundle (ADR-0014): every tag now ships addax-admin-<version>-host.tar.gz plus its .sha256. The bundle contains the prebuilt backend jar, frontend assets, deploy/install.sh, Nginx and systemd examples, BUILD-INFO and checksums, so a host without a Maven/Node toolchain can deploy it directly. Re-running the installer is the upgrade path and preserves config, drivers, job files and logs.
  • Navigation regrouped around the collection lifecycle: collection configuration / collection runs / data governance / administration, no more single-child dropdowns, structure-refresh actions moved into the table management page, and the command palette now derives from the navigation tree.
  • Paused-table semantics fixed: manual runs and fillbacks no longer silently turn a paused (X) table back into automatic scheduling.

Added

  • Host deployment bundle, installer and deployment chapter (docs/manual/install/02-host.md).
  • Navigation regroup (PR #135) and command palette derived from the navigation tree (PR #136).
  • Fixed pixel column widths on the source list with the connection string as the only elastic column; narrow viewports scroll horizontally and pin the action column.

Changed

  • Removed etl_table.kind: it never drove scheduling, and "manual only" is expressed by the paused status X. The startup migration drops the column and recreates the view; the execution snapshot version is now 3.
  • spring.jpa.show-sql defaults to off; local development enables it with SPRING_JPA_SHOW_SQL=true.
  • PostgreSQL JDBC driver upgraded to 42.7.12 (GHSA-j92g-9f8w-j867 / CVE-2026-54291) and minimatch pinned to 11.0.1 (GHSA-g3hv-75wg-m249).
  • Docker images are built on tag pushes only; stale root lockfiles were removed and frontend/yarn.lock is the single lockfile.

Fixed

  • Table detail deep links for ?tab=logs, ?tab=job and ?tab=fillback were reset to info before the detail request returned.
  • The collection state-matrix spec and stale unit specs now match the current status labels and notifier store.

Breaking Changes

  • The startup migration drops etl_table.kind. Older nodes still map the column, so all nodes must be stopped and upgraded together; rolling upgrades are not supported. Rolling back requires restoring the pre-upgrade database backup.
  • Tables that were silently unpaused by an earlier manual run cannot be detected automatically; re-check the paused list after upgrading.

Upgrade

  • Back up the database first and stop every node during the upgrade.
  • Docker: pull the 4.7.0 image and restart; an empty database is still initialised by the application.
  • Host: download the bundle and its .sha256, verify, extract and re-run deploy/install.sh; the previous jar is kept as .before-<timestamp>.
  • Existing host installs keep their config/application.properties; the new template is written next to it as .new.

Validation

  • Backend mvn clean package (JDK 21, unit tests 2/2); frontend type-check, ESLint and production build pass.
  • Frontend unit tests 64/64; the history-snapshot and collection-state-matrix Playwright specs pass against a dedicated backend.
  • Host bundle E2E 24/24: packaging, checksums, install, boot, Nginx reverse proxy, and re-install upgrades that preserve config.
  • Test environment wgzhao-pc deployed with the startup migration and smoke checks passing.