Ceph社区跟踪(2020-10-01 ~ 2020-10-15)




本文作者: 吴 宏 松 https://zhuanlan.zhihu.com/c_1267088333848641536

youtube channel

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

  • Ceph Tech Talk: Karan Singh – Scale Testing Ceph with 10Billion+ Objects
    介绍了10Billion+级对象下对象存储的测试

Ceph Performance Meeting 2020-10-01

  • 讨论新pr:
    • https://github.com/ceph/ceph/pull/37274 (ceph-volume: retrieve device data concurrently)
    • https://github.com/ceph/ceph/pull/37496 (osd: optimize PG removal (part1&2 of 2))
    • 由于rocksdb的wal以及compaction的影响,所以考虑用myrock替代rocksdb(目前对myrock的理解应该是有限的,准备接下来好好研究下myrock)

Ceph Docubetter Meeting 2020-10-14

  • 新pr:
    https://github.com/ceph/ceph/pull/37451/(make cephadm faster and more scalable )
  • 改变了guthub的习惯,以后提doc相关的pr signed off by可以不再需要了
  • ceph文档的链接出问题了,不过他已经修复好了
  • 提到cephfs文档的问题,想parick donney检查并更新一下

Ceph Orchestrator Meeting 2020-10-05

  • 当前rook module(https://docs.ceph.com/en/latest/mgr/rook/)默认还没有enable,原因是还有一些bug没有解决。
  • redhat的人说他们那边只有3个人工作在编排这块,所以没有足够的人力同时进行cephadm以及rook的工作,他们会把工作重心几乎都放在cephadm
  • 目前在进行nfs ganesha在rook上工作的测试
  • 一些编排的指令不能工作(比如不能list host和进程),已经修复了这些bug

邮件列表

https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/

  • https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/ ceph大规模(10Billion+ Objects)下的测试情况
  • https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/thread/7IMIWCKIHXNULEBHVUIXQQGYUDJAO2SF/ osd_pglog导致的osd内存增大,所以考虑要不要增加一个配置项osd_pg_log_memory_limit,使得当pg_log对应的内存空间较大时,可以主动减少
  • https://www.mail-archive.com/ceph-users@ceph.io/msg06745.html 发现集群有slow request,进一步测试发现当把读和写分开放到两个pool则不会出现这种问题(我觉得可能是对象读写锁的问题)
  • https://www.spinics.net/lists/ceph-users/msg62766.html 有人询问cephfs多文件系统是否成熟, Patrick Donnelly说期望在P版本落实这个事情
  • https://lists.ceph.io/hyperkitty/list/ceph-users@ceph.io/thread/PRSXG7THZXRO3UE45S6NC6Y5PU2JMLZF/ go-ceph v0.6.0发布

社区博客

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

  • https://ceph.io/releases/v14-2-13-nautilus-released/ v14.2.13 Nautilus发布了

master近期合入代码

  • bug修复相关:
    • https://github.com/ceph/ceph/pull/37594 (librbd: fix race on watcher unregister)
    • https://github.com/ceph/ceph/pull/37581 (librbd: avoid failing IO with -ESHUTDOWN when disabling exclusive-lock)
    • https://github.com/ceph/ceph/pull/37580(os/bluestore: fix memory accounting in TwoQBufferCacheShard)
    • https://tracker.ceph.com/issues/47751 的修复代码,Normal级别bug,影响M/N/O (os/bluestore: fix segfault on out-of-bound offset provided)
    • https://tracker.ceph.com/issues/47734 的修复bug,Urgent级别,影响N/O (client: hang after statfs)
    • https://tracker.ceph.com/issues/47605 的修复bug,nornam级别,影响N/O (mds: purge_queue’s _calculate_ops is inaccurate)
    • https://tracker.ceph.com/issues/46024 的修复bug,影响N,master(larger osd_scrub_max_preemptions values cause Floating point exception)
    • https://tracker.ceph.com/issues/47608的修复bug,normal级别,影响N/O (mds: OpenFileTable::prefetch_inodes during rejoin can cause out-of-memory)
  • 其他:
    • https://github.com/ceph/ceph/pull/37504 (octopus: erasure-code: enable isa-l EC for aarch64 platform)