Ceph社区跟踪(2020-08-10 ~ 2020-08-23)




本文作者: 徐 桑 迪

Youtube Channel

https://www.youtube.com/c/Cephstorage/videos

  • Ceph Orchestrator Meeting 08-10
    • Rook 1.4发布
    • 更新了删除OSD的设计文档
  • Ceph DocuBetter Meeting 08-12
    • 优先编写Installation guide和development guild
    • 后续考虑录制短视频(~5mins)演示如何安装Ceph
  • Ceph Crimson/SeaStor OSD 08-12
    • 编写新的OSD实现crimson-osd,讨论功能设计和开发进度
    • 关键词:NVMe,SeaStore,Multi-cores,Cache,LBA tree
  • Ceph Performance Meeting 08-13
    • 开发中:根据buffer原大小动态调整append buffer的大小
    • Need QA:(ma jianpeng)RocksDB Env中避免一次性下刷过多的数据
    • Need Review:d3m caching
    • 开发中:(ma jianpeng)优化BlueFS中BufferList重建流程,正在考虑将其应用到通用BufferList中
    • 研究中:有人提出了一个CRUSH算法的优化版本,用来降低集群扩容时的数据迁移影响
    • Paper中仅测试了HDDs,可能是因为在这个硬件上效果更好
    • Sam提出他更关注减少数据迁移,从而可以提升NVMe设备的使用寿命
    • https://www.usenix.org/system/files/fast20-wang_li.pdf
  • Ceph Orchestrator Meeting 08-17
    • 考虑重构Cephadm工具
    • 从原有的Ceph-ansible中拷贝部分代码过来,应该是想后续完全替代掉ceph-ansible
    • 编写使用Cephadm工具部署Ceph集群的指导文档
  • Ceph Crimson/SeaStor OSD 08-19
    • OSD EIO处理
    • 使用crimson关键字跟踪相关BUG
  • Ceph Performance Meeting 08-20
    • 讨论之前提出的CRUSH扩展算法论文;其对新PG使用了新的rule,可能不是很好
      • 要么最终仍然需要迁移数据以达到集群平衡,这样带来了额外的PG合并开销
      • 要么得一直保留PG映射的特殊信息,而且这会随着集群的扩容不断增大
    • 正在积极开发QoS功能,初步验证效果显著
    • PG priming?(可能是高版本特性,暂不了解)
    • 持续关注研究领域的Ceph/CRUSH相关论文,定期一起阅读讨论

邮件列表

IRC

  • https://tracker.ceph.com/projects/ceph/wiki/Code_Walkthroughs
    • 2020-08-25 @ 7am PDT: KRBD I/O Flow – Ilya Dryomov
    • 2020-10-27 @ 7am PDT: Librbd I/O Flow – Jason Dillaman
  • Proposal: Encrypted Namespaces
  • CEPH dmclock per-client QoS control
    • not implemented, still need a lot of work
  • https://lists.ceph.io/hyperkitty/list/dev@ceph.io/thread/IMYRL55PURREJXFLVX3FKHJ4QQX57JA7/
    • 33% possible performance regression between 15.2.4 and HEAD?
    • checking bluestore disk allocator work
  • https://github.com/ceph/go-ceph/releases/tag/v0.5.0
    • go-ceph v0.5.0 released

社区博客

https://ceph.io/community/blog/

  • https://ceph.io/releases/v14-2-11-nautilus-released/
    • abort scrub/deep-scrub by setting certain osd flag
    • implement Hybrid allocator for bluestore
    • do not raise “client failing to respond to cap release” when client working set is reasonable

代码合入

主要通过新版本的Release Notes看合入的commits:https://docs.ceph.com/docs/master/releases/general/#release-timeline

V14.2.11 NAUTILUS
标签 Commits数 值得关注的修复/特性
bluestore 2
build/ops 1
ceph-volume 1
cephfs 27
core:mon 1
core:mgr 2
core:osd 6
mgr modules 10 nautilus: New msgr2 crc and secure modes (msgr2.1)
msgr 1
rbd 3
rgw 15
tools 1

其它

  • https://github.com/ceph/cbt
    • Ceph benchmarking tool(CBT)
  • https://docs.ceph.com/docs/master/dev/crimson/
  • https://docs.ceph.com/docs/master/dev/seastore/