📝 9 新 Issues
✅ 2 关闭
✨ 8 新 PRs
🎉 5 合并
🔥 核心要点
- 🎉 PR #3059 修复主键表 lake tiering 未启动时 pendingRecords 指标返回实际行数
- 🎉 PR #3057 修复 Spark 读取 log/upsert 路径的列投影问题
- 🎉 PR #2803 Netty 客户端优先使用堆内存解码消息，减少直接内存使用
- ✨ PR #3082 添加 fluss-microbench 模块：YAML 驱动的性能测试框架
- 📚 PR #3087 添加贡献者成为 Committer 的文档说明
💬 Issue/PR 动态
Issue 讨论
Issue #3073 add document for how contributors can be granted more rights
- @matrixsparse: 申请负责此文档编写，将参考 Paimon 的 committer 指南
- @luoyuxia: 已分配该 issue 给 matrixsparse
Issue #2007 Add missing authentication for RPC calls
- @vaibhavk1992: 计划分两阶段实现 RPC 授权：Phase 1 外部 RPC（高优先级），Phase 2 内部 RPC
Issue #3022 [Blog] realtime profiles blog post
- @polyzos: 关闭此 PR，改为使用 platinumhamburg 的版本
- @platinumhamburg: 感谢，将使用自己的版本继续
- @polyzos: 将此内容移植到博客，方便用户查找
PR Review
PR #3067 [filesystem] Support default AWS credential chain for S3 delegation
- @fresh-borzoni: 已处理 Copilot 评论，添加 IRSA/实例配置文档和配置示例
- @Copilot: 建议验证 accessKey 和 secretKey 同时设置或同时为空，避免配置错误
- @morazow: 建议添加只设置单个密钥的测试，并提取 buildHadoopConfiguration 方法
- @michaelkoepf: 建议使用 checkArgument 替代 checkNotNull 验证配置参数
PR #2803 [client] Netty client prefer heap memory to cumulate in decoder
- @fresh-borzoni: 指出 OOM 时 allocateForFrame 抛异常会导致 in 未释放，建议参考 expandCummulation 模式
- @loserwang1024: 已采纳建议修改，并解释服务端 ClientChannelInitializer 使用 netty.client.* 配置
PR #3026 [client] Replace Netty PooledByteBufAllocator with ChunkedAllocationManager
- @fresh-borzoni: 发现 ChunkedAllocationManager 存在竞态条件可能导致双重回收和内存损坏
- @loserwang1024: 确认只在所有数据读取完毕后才关闭 ChunkedAllocationManager，感谢建议
PR #3076 [lake/tiering] add tiering table pending time and freshness metrics
- @Copilot: 指出 pendingTime 使用 0L 作为哨兵值可能在时钟为 0 时出问题，建议使用 null 或负值
PR #3042 [spark] Support batch union read for lake-enabled primary key tables
- @Yohahaha: 建议 nice，等待 review
- @Copilot: 指出 lakeSource 未应用 projection 可能导致列顺序错误，以及日志偏移处理逻辑问题
PR #2997 [filesystem][bug] Fix client.fs.* config prefix not stripped
- @fresh-borzoni: 已处理所有审查意见，并修复了 Azure 中 flussConfig 应为 hadoopConfig 的 bug
- @morazow: 建议更新 Javadoc，扩展测试覆盖 OBS 和 OSS 的配置
PR #3071 [server] Make leader-only rebalance tasks execute sequentially
- @LiebingYu: 指出顺序迁移后单个 bucket 失败可能阻塞后续迁移，需用户手动处理
PR #3059 [metrics] Fix pendingRecords metric to return actual row count
- @luoyuxia: 询问 v0.9 之前创建的旧表如何处理
PR #2786 [client] cancel pending log download when closing remote log downloader
- @loserwang1024: close() 需要阻塞等待所有下载完成，否则会留下孤儿文件导致磁盘堆积
- @fresh-borzoni: 指出 cancel(true) 不等待，deleteDirectoryQuietly 会与下载任务竞争
PR #3077 [blog] taobao realtime decisions
- @wuchong: 指出作者应为 Howie Wang，需使用 GitHub 头像
📝 新建 Issue/PR
Issues
- #3085 Remote log consumption is stuck because the semaphore permit cannot be acquired @zuston
- #3084 Flink log table consumer threw an out-of-range offset exception @zuston
- #3083 [test] Unstable test RemoteLogITCase.testRemoteLogTTLWithDynamicLakeToggle @app/github-actions
- #3080 Add fluss-microbench module: YAML-driven performance testing framework @platinumhamburg
- #3075 [lake/tiering] Add pendingTime and freshness metrics for lake tiering tables @beryllw
- #3074 add a page for reporting security issue @luoyuxia
- #3073 add document for how contributors can be granted more rights such as commit access or decision power is clearly documented @luoyuxia
- #3072 Csp_check is warning @luoyuxia
- #3070 [server] Leader-only rebalance tasks should wait for tablet server acknowledgment before proceeding to the next migration @swuferhong
Pull Requests
- #3088 [Blog] Column Pruning @polyzos
- #3087 [ISSUE-3073] Add documentation for how contributors can become committers @matrixsparse
- #3082 Add fluss-microbench module with YAML-driven perf framework and JMH benchmarks @platinumhamburg
- #3078 [server] Add authorization to databaseExists and tableExists RPC call… @vaibhavk1992
- #3077 [blog] taobao realtime decisions @polyzos
- #3076 [lake/tiering] add tiering table pending time and freshness metrics @beryllw
- #3071 [server] Make leader-only rebalance tasks execute sequentially @swuferhong
- #3069 [client] Add validation for client.scanner.log.max-poll-records and client.connect-timeout @Prajwal-banakar
✅ 关闭 Issue/PR
已关闭 Issues
- #3029 [spark] Column projection broken on log/upsert read path for type-dependent accessors @fresh-borzoni
- #2773 Netty client prefers to use heap memory to decode message to reduce the direct memory use. @loserwang1024
已合并 PRs
- #3086 [website] Hide left sidebar for blog page and improve the fonts @wuchong
- #3079 [Blog] Realtime UV Dedup @polyzos
- #3059 [metrics] Fix pendingRecords metric to return actual row count for primary key tables when lake tiering hasn't started @beryllw
- #3057 [spark] Fix column projection on log/upsert read path @fresh-borzoni
- #2803 [client] Netty client prefer heap memory to cumulate in decoder. @loserwang1024