ceph.rst 7.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190
  1. .. SPDX-License-Identifier: GPL-2.0
  2. ============================
  3. Ceph Distributed File System
  4. ============================
  5. Ceph is a distributed network file system designed to provide good
  6. performance, reliability, and scalability.
  7. Basic features include:
  8. * POSIX semantics
  9. * Seamless scaling from 1 to many thousands of nodes
  10. * High availability and reliability. No single point of failure.
  11. * N-way replication of data across storage nodes
  12. * Fast recovery from node failures
  13. * Automatic rebalancing of data on node addition/removal
  14. * Easy deployment: most FS components are userspace daemons
  15. Also,
  16. * Flexible snapshots (on any directory)
  17. * Recursive accounting (nested files, directories, bytes)
  18. In contrast to cluster filesystems like GFS, OCFS2, and GPFS that rely
  19. on symmetric access by all clients to shared block devices, Ceph
  20. separates data and metadata management into independent server
  21. clusters, similar to Lustre. Unlike Lustre, however, metadata and
  22. storage nodes run entirely as user space daemons. File data is striped
  23. across storage nodes in large chunks to distribute workload and
  24. facilitate high throughputs. When storage nodes fail, data is
  25. re-replicated in a distributed fashion by the storage nodes themselves
  26. (with some minimal coordination from a cluster monitor), making the
  27. system extremely efficient and scalable.
  28. Metadata servers effectively form a large, consistent, distributed
  29. in-memory cache above the file namespace that is extremely scalable,
  30. dynamically redistributes metadata in response to workload changes,
  31. and can tolerate arbitrary (well, non-Byzantine) node failures. The
  32. metadata server takes a somewhat unconventional approach to metadata
  33. storage to significantly improve performance for common workloads. In
  34. particular, inodes with only a single link are embedded in
  35. directories, allowing entire directories of dentries and inodes to be
  36. loaded into its cache with a single I/O operation. The contents of
  37. extremely large directories can be fragmented and managed by
  38. independent metadata servers, allowing scalable concurrent access.
  39. The system offers automatic data rebalancing/migration when scaling
  40. from a small cluster of just a few nodes to many hundreds, without
  41. requiring an administrator carve the data set into static volumes or
  42. go through the tedious process of migrating data between servers.
  43. When the file system approaches full, new nodes can be easily added
  44. and things will "just work."
  45. Ceph includes flexible snapshot mechanism that allows a user to create
  46. a snapshot on any subdirectory (and its nested contents) in the
  47. system. Snapshot creation and deletion are as simple as 'mkdir
  48. .snap/foo' and 'rmdir .snap/foo'.
  49. Ceph also provides some recursive accounting on directories for nested
  50. files and bytes. That is, a 'getfattr -d foo' on any directory in the
  51. system will reveal the total number of nested regular files and
  52. subdirectories, and a summation of all nested file sizes. This makes
  53. the identification of large disk space consumers relatively quick, as
  54. no 'du' or similar recursive scan of the file system is required.
  55. Finally, Ceph also allows quotas to be set on any directory in the system.
  56. The quota can restrict the number of bytes or the number of files stored
  57. beneath that point in the directory hierarchy. Quotas can be set using
  58. extended attributes 'ceph.quota.max_files' and 'ceph.quota.max_bytes', eg::
  59. setfattr -n ceph.quota.max_bytes -v 100000000 /some/dir
  60. getfattr -n ceph.quota.max_bytes /some/dir
  61. A limitation of the current quotas implementation is that it relies on the
  62. cooperation of the client mounting the file system to stop writers when a
  63. limit is reached. A modified or adversarial client cannot be prevented
  64. from writing as much data as it needs.
  65. Mount Syntax
  66. ============
  67. The basic mount syntax is::
  68. # mount -t ceph monip[:port][,monip2[:port]...]:/[subdir] mnt
  69. You only need to specify a single monitor, as the client will get the
  70. full list when it connects. (However, if the monitor you specify
  71. happens to be down, the mount won't succeed.) The port can be left
  72. off if the monitor is using the default. So if the monitor is at
  73. 1.2.3.4::
  74. # mount -t ceph 1.2.3.4:/ /mnt/ceph
  75. is sufficient. If /sbin/mount.ceph is installed, a hostname can be
  76. used instead of an IP address.
  77. Mount Options
  78. =============
  79. ip=A.B.C.D[:N]
  80. Specify the IP and/or port the client should bind to locally.
  81. There is normally not much reason to do this. If the IP is not
  82. specified, the client's IP address is determined by looking at the
  83. address its connection to the monitor originates from.
  84. wsize=X
  85. Specify the maximum write size in bytes. Default: 64 MB.
  86. rsize=X
  87. Specify the maximum read size in bytes. Default: 64 MB.
  88. rasize=X
  89. Specify the maximum readahead size in bytes. Default: 8 MB.
  90. mount_timeout=X
  91. Specify the timeout value for mount (in seconds), in the case
  92. of a non-responsive Ceph file system. The default is 60
  93. seconds.
  94. caps_max=X
  95. Specify the maximum number of caps to hold. Unused caps are released
  96. when number of caps exceeds the limit. The default is 0 (no limit)
  97. rbytes
  98. When stat() is called on a directory, set st_size to 'rbytes',
  99. the summation of file sizes over all files nested beneath that
  100. directory. This is the default.
  101. norbytes
  102. When stat() is called on a directory, set st_size to the
  103. number of entries in that directory.
  104. nocrc
  105. Disable CRC32C calculation for data writes. If set, the storage node
  106. must rely on TCP's error correction to detect data corruption
  107. in the data payload.
  108. dcache
  109. Use the dcache contents to perform negative lookups and
  110. readdir when the client has the entire directory contents in
  111. its cache. (This does not change correctness; the client uses
  112. cached metadata only when a lease or capability ensures it is
  113. valid.)
  114. nodcache
  115. Do not use the dcache as above. This avoids a significant amount of
  116. complex code, sacrificing performance without affecting correctness,
  117. and is useful for tracking down bugs.
  118. noasyncreaddir
  119. Do not use the dcache as above for readdir.
  120. noquotadf
  121. Report overall filesystem usage in statfs instead of using the root
  122. directory quota.
  123. nocopyfrom
  124. Don't use the RADOS 'copy-from' operation to perform remote object
  125. copies. Currently, it's only used in copy_file_range, which will revert
  126. to the default VFS implementation if this option is used.
  127. recover_session=<no|clean>
  128. Set auto reconnect mode in the case where the client is blocklisted. The
  129. available modes are "no" and "clean". The default is "no".
  130. * no: never attempt to reconnect when client detects that it has been
  131. blocklisted. Operations will generally fail after being blocklisted.
  132. * clean: client reconnects to the ceph cluster automatically when it
  133. detects that it has been blocklisted. During reconnect, client drops
  134. dirty data/metadata, invalidates page caches and writable file handles.
  135. After reconnect, file locks become stale because the MDS loses track
  136. of them. If an inode contains any stale file locks, read/write on the
  137. inode is not allowed until applications release all stale file locks.
  138. More Information
  139. ================
  140. For more information on Ceph, see the home page at
  141. https://ceph.com/
  142. The Linux kernel client source tree is available at
  143. - https://github.com/ceph/ceph-client.git
  144. - git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client.git
  145. and the source for the full system is at
  146. https://github.com/ceph/ceph.git