fix publish backup flow

This commit is contained in:
kdletters
2026-05-28 02:05:41 +08:00
parent 23f6317c8b
commit de8b82c575
5 changed files with 193 additions and 28 deletions

View File

@@ -95,7 +95,7 @@ npm run dev:admin-web
开发态 `npm run dev` / `npm run dev:api-server` 默认打开 `GENARRATIVE_DEV_PASSWORD_ENTRY_AUTO_REGISTER_ENABLED=true`,密码入口可以直接注册未知手机号账号;生产默认仍关闭该开关。
生产 `Genarrative-Stdb-Module-Publish` 的备份默认使用 `DATABASE_BACKUP_MODE=async`:流水线在 publish 成功并退出维护模式后只触发服务器上的 `genarrative-database-backup.service`,避免低带宽 OSS 上传长时间占住部署窗口。需要强制在 publish 前等待备份并让失败阻断发布时,手动选择 `DATABASE_BACKUP_MODE=sync`;已有其他备份窗口且明确接受风险时才选择 `skip`
生产 `Genarrative-Stdb-Module-Publish` 的备份默认使用 `DATABASE_BACKUP_MODE=async`:流水线在 publish 前先生成本地冷备份,随后继续 publish并把同一份发布前备份交给后台 Node 进程上传 OSS,避免低带宽 OSS 上传长时间占住部署窗口。需要强制在 publish 前等待打包和上传并让失败阻断发布时,手动选择 `DATABASE_BACKUP_MODE=sync`;已有其他备份窗口且明确接受风险时才选择 `skip`
查看本地 Rust/SpacetimeDB 日志:

View File

@@ -201,7 +201,7 @@ UI 相关修改要重点验证:
npm run database:backup:oss -- --data-dir /stdb --stop-service spacetimedb.service
```
脚本会将数据目录打包成 `tar.gz`,上传到 `oss://<bucket>/<prefix>/<database>/<database>-<UTC时间>.tar.gz`。生产建议做冷备份:传入 `--stop-service spacetimedb.service`,脚本会在打包前停止服务、打包后恢复服务,再上传 OSS。由于 OSS 上传可能受服务器带宽限制,`Genarrative-Stdb-Module-Publish` 默认使用 `DATABASE_BACKUP_MODE=async`在 publish 成功并退出维护模式后通过 `systemctl start --no-block genarrative-database-backup.service` 触发服务器后台备份,不等待上传完成,也不因备份上传耗时阻塞发布;需要强一致发布闸门时改用 `DATABASE_BACKUP_MODE=sync`(等价脚本参数 `--backup-mode sync`),备份会在 publish 前同步执行,失败会阻断 publish确认已有其他备份窗口时才使用 `DATABASE_BACKUP_MODE=skip`(兼容脚本参数 `--skip-backup`)。若业务不能接受停机窗口,应先规划 SpacetimeDB 原生快照或主备策略,不要直接在写入中的数据目录上做热拷贝并当作强一致备份。
脚本会将数据目录打包成 `tar.gz`,上传到 `oss://<bucket>/<prefix>/<database>/<database>-<UTC时间>.tar.gz`。生产建议做冷备份:传入 `--stop-service spacetimedb.service`,脚本会在打包前停止服务、打包后恢复服务,再上传 OSS。由于 OSS 上传可能受服务器带宽限制,`Genarrative-Stdb-Module-Publish` 默认使用 `DATABASE_BACKUP_MODE=async`:先在 publish 前用 `--defer-upload` 生成本地冷备份和 `.manifest.json`,随后继续执行 publish发布脚本退出前会用后台 `node ... --upload-archive <tar.gz>` 上传同一份发布前备份,不等待上传完成。需要强一致发布闸门时改用 `DATABASE_BACKUP_MODE=sync`(等价脚本参数 `--backup-mode sync`),备份会在 publish 前同步打包并上传,失败会阻断 publish确认已有其他备份窗口时才使用 `DATABASE_BACKUP_MODE=skip`(兼容脚本参数 `--skip-backup`)。若业务不能接受停机窗口,应先规划 SpacetimeDB 原生快照或主备策略,不要直接在写入中的数据目录上做热拷贝并当作强一致备份。
生产环境变量模板在 `deploy/env/api-server.env.example`

View File

@@ -27,7 +27,7 @@ pipeline {
string(name: 'SPACETIME_ROOT_DIR', defaultValue: '/stdb', description: 'spacetime CLI root-dir需与自托管 spacetimedb.service 一致')
string(name: 'SPACETIME_RUN_AS_USER', defaultValue: 'spacetimedb', description: '执行 spacetime publish 的本机用户,默认使用自托管服务用户')
booleanParam(name: 'CLEAR_DATABASE', defaultValue: false, description: '是否清空数据库后发布')
choice(name: 'DATABASE_BACKUP_MODE', choices: ['async', 'sync', 'skip'], description: '数据库备份策略async 在 publish 成功后触发服务器 systemd 备份并继续sync 在 publish 前等待备份完成且失败阻断skip 跳过')
choice(name: 'DATABASE_BACKUP_MODE', choices: ['async', 'sync', 'skip'], description: '数据库备份策略async 在 publish 前生成本地冷备份、后台上传 OSSsync 在 publish 前等待上传完成且失败阻断skip 跳过')
}
stages {
@@ -143,7 +143,18 @@ pipeline {
if (!(backupMode in ['async', 'sync', 'skip'])) {
error("DATABASE_BACKUP_MODE 只能是 async、sync 或 skip: ${backupMode}")
}
def backupArg = "--backup-mode \"${backupMode}\""
def publishScriptPath = 'scripts/deploy/production-stdb-publish.sh'
def publishScriptText = readFile(file: publishScriptPath, encoding: 'UTF-8')
def publishScriptSupportsBackupMode = publishScriptText.contains('--backup-mode')
def backupArg = ''
if (publishScriptSupportsBackupMode) {
backupArg = "--backup-mode \"${backupMode}\""
} else if (backupMode == 'sync') {
error('当前工作区的 scripts/deploy/production-stdb-publish.sh 还不支持 --backup-mode无法保证 sync 语义。请先更新工作区脚本后再运行。')
} else {
echo "[Jenkins] 当前工作区脚本还不支持 --backup-modeasync/skip 自动降级为 --skip-backup避免参数不兼容导致发布失败。"
backupArg = '--skip-backup'
}
def rootArg = "--root-dir \"${params.SPACETIME_ROOT_DIR?.trim() ? params.SPACETIME_ROOT_DIR.trim() : '/stdb'}\""
def runAsArg = params.SPACETIME_RUN_AS_USER?.trim()
? "--run-as-user \"${params.SPACETIME_RUN_AS_USER.trim()}\""

View File

@@ -20,10 +20,12 @@ const UNSIGNED_PAYLOAD = 'UNSIGNED-PAYLOAD';
function usage() {
console.log(`用法:
npm run database:backup:oss -- [--data-dir <path>] [--work-dir <path>] [--bucket <bucket>] [--object-prefix <prefix>] [--keep-local]
node scripts/database-backup-to-oss.mjs [--stop-service spacetimedb.service]
node scripts/database-backup-to-oss.mjs [--stop-service spacetimedb.service] [--defer-upload]
node scripts/database-backup-to-oss.mjs --upload-archive <path>
说明:
将 SpacetimeDB 数据目录打包成 .tar.gz并上传到阿里云 OSS 指定 bucket。
--defer-upload 只生成本地冷备份和 manifest不上传后续用 --upload-archive 异步上传。
默认读取 .env / .env.local / .env.secrets.local生产服务可传 --env-file /etc/genarrative/api-server.env。
shell 环境变量优先级最高,不会被 env 文件覆盖。
@@ -100,6 +102,11 @@ function parseArgs(argv) {
stopService: '',
database: '',
dryRun: false,
deferUpload: false,
uploadArchive: '',
manifestFile: '',
objectKey: '',
resultFile: '',
};
for (let index = 0; index < argv.length; index += 1) {
@@ -134,6 +141,9 @@ function parseArgs(argv) {
case '--object-prefix':
options.objectPrefix = readValue();
break;
case '--object-key':
options.objectKey = readValue();
break;
case '--access-key-id':
options.accessKeyId = readValue();
break;
@@ -155,6 +165,19 @@ function parseArgs(argv) {
case '--dry-run':
options.dryRun = true;
break;
case '--defer-upload':
options.deferUpload = true;
options.keepLocal = true;
break;
case '--upload-archive':
options.uploadArchive = readValue();
break;
case '--manifest-file':
options.manifestFile = readValue();
break;
case '--result-file':
options.resultFile = readValue();
break;
default:
throw new Error(`未知参数: ${arg}`);
}
@@ -260,6 +283,17 @@ function createArchive({dataDir, workDir, fileName}) {
return archivePath;
}
function writeManifest({manifestPath, payload}) {
writeFileSync(manifestPath, `${JSON.stringify(payload, null, 2)}\n`, 'utf8');
}
function readManifest(manifestPath) {
if (!existsSync(manifestPath)) {
throw new Error(`备份清单不存在: ${manifestPath}`);
}
return JSON.parse(readFileSync(manifestPath, 'utf8'));
}
function hmac(key, content, encoding) {
return createHmac('sha256', key).update(content).digest(encoding);
}
@@ -372,6 +406,59 @@ async function uploadArchive({archivePath, bucket, endpoint, objectKey, accessKe
};
}
async function uploadExistingArchive({args, env, bucket, endpoint, accessKeyId, accessKeySecret, objectPrefix}) {
const archivePath = resolvePath(args.uploadArchive);
if (!existsSync(archivePath)) {
throw new Error(`待上传备份文件不存在: ${archivePath}`);
}
const manifestPath = resolvePath(args.manifestFile || `${archivePath}.manifest.json`);
const manifest = existsSync(manifestPath) ? readManifest(manifestPath) : {};
const dataDir = firstNonEmpty(manifest.dataDir, env.GENARRATIVE_DATABASE_BACKUP_DATA_DIR, DEFAULT_PRODUCTION_DATA_DIR);
const database = firstNonEmpty(args.database, manifest.database, env.GENARRATIVE_SPACETIME_DATABASE, basename(dataDir));
const objectKey = firstNonEmpty(args.objectKey, manifest.objectKey, buildBackupNames({database, dataDir, objectPrefix}).objectKey);
console.log(`[database-backup] 上传已有备份: ${archivePath}`);
console.log(`[database-backup] 目标对象: oss://${bucket}/${objectKey}`);
if (args.dryRun) {
console.log('[database-backup] dry-run仅校验上传配置。');
return;
}
const result = await uploadArchive({archivePath, bucket, endpoint, objectKey, accessKeyId, accessKeySecret});
console.log(`[database-backup] 上传完成: ${JSON.stringify(result)}`);
const uploadedAt = new Date().toISOString();
writeManifest({
manifestPath,
payload: {
...manifest,
database,
bucket: result.bucket,
objectKey: result.objectKey,
contentLength: result.contentLength,
etag: result.etag,
uploadedAt,
uploadStatus: 'uploaded',
},
});
if (args.resultFile) {
writeFileSync(resolvePath(args.resultFile), `${JSON.stringify({archivePath, manifestPath, ...result, uploadedAt}, null, 2)}\n`, 'utf8');
}
const keepLocal = args.keepLocal || String(env.GENARRATIVE_DATABASE_BACKUP_KEEP_LOCAL ?? '').trim().toLowerCase() === 'true';
if (!keepLocal) {
rmSync(archivePath, {force: true});
rmSync(manifestPath, {force: true});
console.log('[database-backup] 已删除本地临时备份文件;如需保留请设置 --keep-local。');
} else {
console.log(`[database-backup] 已保留本地备份: ${archivePath}`);
console.log(`[database-backup] 已保留备份清单: ${manifestPath}`);
}
}
async function main() {
const args = parseArgs(process.argv.slice(2));
const env = loadEffectiveEnv(args.envFiles);
@@ -400,6 +487,11 @@ async function main() {
}
}
if (args.uploadArchive) {
await uploadExistingArchive({args, env, bucket, endpoint, accessKeyId, accessKeySecret, objectPrefix});
return;
}
const {fileName, objectKey} = buildBackupNames({database, dataDir, objectPrefix});
console.log(`[database-backup] 数据目录: ${dataDir}`);
console.log(`[database-backup] 本地临时目录: ${workDir}`);
@@ -419,22 +511,47 @@ async function main() {
startServiceIfNeeded(args.stopService || firstNonEmpty(env.GENARRATIVE_DATABASE_BACKUP_STOP_SERVICE), serviceStopped);
}
const manifestPath = `${archivePath}.manifest.json`;
writeManifest({
manifestPath,
payload: {
createdAt: new Date().toISOString(),
database,
dataDir,
bucket,
objectKey,
archivePath,
uploadStatus: args.deferUpload ? 'deferred' : 'pending',
},
});
if (args.deferUpload) {
console.log(`[database-backup] 已生成本地冷备份,延后上传: ${archivePath}`);
console.log(`[database-backup] 已写入备份清单: ${manifestPath}`);
if (args.resultFile) {
writeFileSync(resolvePath(args.resultFile), `${JSON.stringify({archivePath, manifestPath, bucket, objectKey}, null, 2)}\n`, 'utf8');
}
return;
}
const result = await uploadArchive({archivePath, bucket, endpoint, objectKey, accessKeyId, accessKeySecret});
console.log(`[database-backup] 上传完成: ${JSON.stringify(result)}`);
const manifestPath = `${archivePath}.manifest.json`;
writeFileSync(
writeManifest({
manifestPath,
`${JSON.stringify({
payload: {
createdAt: new Date().toISOString(),
database,
dataDir,
bucket: result.bucket,
objectKey: result.objectKey,
archivePath,
contentLength: result.contentLength,
etag: result.etag,
}, null, 2)}\n`,
'utf8',
);
uploadedAt: new Date().toISOString(),
uploadStatus: 'uploaded',
},
});
if (!keepLocal) {
rmSync(archivePath, {force: true});

View File

@@ -12,7 +12,7 @@ usage() {
默认使用 http://127.0.0.1:3101避免与部署机本机 Git/Web 服务的 3000 端口冲突。
默认使用 /stdb 作为 spacetime CLI root-dir并以 spacetimedb 用户发布,避免 root CLI 身份污染自托管实例。
发布时固定追加 --no-config只使用显式参数避免工作区或用户目录里的 spacetime 配置干扰目标。
默认在 publish 成功后异步触发 genarrative-database-backup.service,避免低带宽 OSS 上传阻塞部署。
async 模式会在 publish 前先做本地冷备份,再在 publish 完成后后台上传 OSS,避免低带宽上传阻塞部署。
如需强制等待备份完成并在失败时阻断 publish传入 --backup-mode sync。
失败时保留维护模式。
EOF
@@ -48,6 +48,11 @@ CLEAR_DATABASE=0
BACKUP_MODE="${GENARRATIVE_STDB_PUBLISH_BACKUP_MODE:-async}"
DEPLOY_COMPLETED=0
PUBLISH_TMP_DIR=""
ASYNC_BACKUP_STATUS_FILE=""
ASYNC_BACKUP_SCRIPT=""
ASYNC_BACKUP_ARCHIVE=""
ASYNC_BACKUP_MANIFEST=""
ASYNC_BACKUP_LOG=""
while [[ $# -gt 0 ]]; do
case "$1" in
@@ -137,6 +142,9 @@ fi
on_exit() {
local exit_code=$?
if [[ "${BACKUP_MODE}" == "async" && -n "${ASYNC_BACKUP_STATUS_FILE}" && -f "${ASYNC_BACKUP_STATUS_FILE}" ]]; then
start_async_backup_upload || true
fi
if [[ -n "${PUBLISH_TMP_DIR}" && -d "${PUBLISH_TMP_DIR}" ]]; then
rm -rf "${PUBLISH_TMP_DIR}"
fi
@@ -148,25 +156,58 @@ on_exit() {
trap on_exit EXIT
trigger_async_backup() {
# Jenkins 发布路径不能被低带宽 OSS 上传长时间占住;默认只把已安装的 systemd
# oneshot 备份任务排队启动。必须放在 publish 成功后,避免冷备份停止 SpacetimeDB
# 与 spacetime publish 同时争用 spacetimedb.service。
if command -v systemctl >/dev/null 2>&1 && systemctl list-unit-files genarrative-database-backup.service --no-legend | grep -q '^genarrative-database-backup\.service'; then
echo "[production-stdb-publish] 异步触发数据库 OSS 备份,不等待上传完成"
if ! systemctl start --no-block genarrative-database-backup.service; then
echo "[production-stdb-publish] 警告:异步触发数据库备份失败;继续发布,请检查 genarrative-database-backup.service 日志" >&2
prepare_async_backup() {
ASYNC_BACKUP_SCRIPT="${SCRIPT_DIR}/../database-backup-to-oss.mjs"
if [[ ! -f "${ASYNC_BACKUP_SCRIPT}" ]]; then
ASYNC_BACKUP_SCRIPT="${SOURCE_DIR}/scripts/database-backup-to-oss.mjs"
fi
else
echo "[production-stdb-publish] 警告:未找到 genarrative-database-backup.service跳过异步备份触发" >&2
if [[ ! -f "${ASYNC_BACKUP_SCRIPT}" ]]; then
echo "[production-stdb-publish] 缺少数据库备份脚本: ${ASYNC_BACKUP_SCRIPT}" >&2
exit 1
fi
ASYNC_BACKUP_STATUS_FILE="$(mktemp /tmp/genarrative-stdb-backup-status.XXXXXX.json)"
echo "[production-stdb-publish] publish 前生成本地冷备份,随后会异步上传 OSS"
node "${ASYNC_BACKUP_SCRIPT}" \
--env-file /etc/genarrative/api-server.env \
--data-dir "${SPACETIME_ROOT_DIR}" \
--database "${DATABASE}" \
--stop-service spacetimedb.service \
--defer-upload \
--result-file "${ASYNC_BACKUP_STATUS_FILE}"
}
start_async_backup_upload() {
if [[ -z "${ASYNC_BACKUP_STATUS_FILE}" || ! -f "${ASYNC_BACKUP_STATUS_FILE}" ]]; then
echo "[production-stdb-publish] 警告:未找到可上传的本地备份状态文件,跳过异步上传" >&2
return 0
fi
ASYNC_BACKUP_ARCHIVE="$(node -e 'const fs=require("node:fs"); const p=process.argv[1]; const o=JSON.parse(fs.readFileSync(p,"utf8")); process.stdout.write(o.archivePath || "");' "${ASYNC_BACKUP_STATUS_FILE}")"
ASYNC_BACKUP_MANIFEST="$(node -e 'const fs=require("node:fs"); const p=process.argv[1]; const o=JSON.parse(fs.readFileSync(p,"utf8")); process.stdout.write(o.manifestPath || "");' "${ASYNC_BACKUP_STATUS_FILE}")"
if [[ -z "${ASYNC_BACKUP_ARCHIVE}" || -z "${ASYNC_BACKUP_MANIFEST}" ]]; then
echo "[production-stdb-publish] 警告:备份状态文件缺少 archivePath 或 manifestPath跳过异步上传" >&2
return 0
fi
mkdir -p "$(dirname "${ASYNC_BACKUP_ARCHIVE}")"
ASYNC_BACKUP_LOG="$(dirname "${ASYNC_BACKUP_ARCHIVE}")/${DATABASE}-upload.log"
echo "[production-stdb-publish] 后台上传本地备份到 OSS: ${ASYNC_BACKUP_ARCHIVE}"
nohup node "${ASYNC_BACKUP_SCRIPT}" \
--env-file /etc/genarrative/api-server.env \
--upload-archive "${ASYNC_BACKUP_ARCHIVE}" \
--manifest-file "${ASYNC_BACKUP_MANIFEST}" \
>"${ASYNC_BACKUP_LOG}" 2>&1 &
echo "[production-stdb-publish] OSS 后台上传日志: ${ASYNC_BACKUP_LOG}"
rm -f "${ASYNC_BACKUP_STATUS_FILE}"
ASYNC_BACKUP_STATUS_FILE=""
}
"${SCRIPT_DIR}/maintenance-on.sh" "spacetime module publish ${DATABASE}"
case "${BACKUP_MODE}" in
async)
echo "[production-stdb-publish] 将在 publish 成功后异步触发数据库 OSS 备份"
prepare_async_backup
;;
sync)
BACKUP_SCRIPT="${SCRIPT_DIR}/../database-backup-to-oss.mjs"
@@ -253,8 +294,4 @@ fi
"${SCRIPT_DIR}/maintenance-off.sh"
DEPLOY_COMPLETED=1
if [[ "${BACKUP_MODE}" == "async" ]]; then
trigger_async_backup
fi
echo "[production-stdb-publish] 完成"