ceph/doc/release-notes.rst

===============
 Release Notes
===============

v0.56 "bobtail"
---------------

Bobtail is the second stable release of Ceph, named in honor of the
`Bobtail Squid`: http://en.wikipedia.org/wiki/Bobtail_squid.

Key features since v0.48 "argonaut"
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

* Object Storage Daemon (OSD): improved threading, small-io performance, and performance during recovery
* Object Storage Daemon (OSD): regular "deep" scrubbing of all stored data to detect latent disk errors
* RADOS Block Device (RBD): support for copy-on-write clones of images.
* RADOS Block Device (RBD): better client-side caching.
* RADOS Block Device (RBD): advisory image locking
* Rados Gateway (RGW): support for efficient usage logging/scraping (for billing purposes)
* Rados Gateway (RGW): expanded S3 and Swift API coverage (e.g., POST, multi-object delete)
* Rados Gateway (RGW): improved striping for large objects
* Rados Gateway (RGW): OpenStack Keystone integration
* RPM packages for Fedora, RHEL/CentOS, OpenSUSE, and SLES
* mkcephfs: support for automatically formatting and mounting XFS and ext4 (in addition to btrfs)

Upgrading
~~~~~~~~~

Please refer to the document Upgrading from Argonaut to Bobtail.

* Cephx authentication is now enabled by default (since v0.55).
  Upgrading a cluster without adjusting the Ceph configuration will
  likely prevent the system from starting up on its own.  We recommend
  first modifying the configuration to indicate that authentication is
  disabled, and only then upgrading to the latest version.::

     auth client required = none
     auth service required = none
     auth cluster required = none

* Ceph daemons can be upgraded one-by-one while the cluster is online
  and in service.

* The ``ceph-osd`` daemons must be upgraded and restarted *before* any
  ``radosgw`` daemons are restarted, as they depend on some new
  ceph-osd functionality.  (The ``ceph-mon``, ``ceph-osd``, and
  ``ceph-mds`` daemons can be upgraded and restarted in any order.)

* Once each individual daemon has been upgraded and restarted, it
  cannot be downgraded.

* The cluster of ``ceph-mon`` daemons will migrate to a new internal
  on-wire protocol once all daemons in the quorum have been upgraded.
  Upgrading only a majority of the nodes (e.g., two out of three) may
  expose the cluster to a situation where a single additional failure
  may compromise availability (because the non-upgraded daemon cannot
  participate in the new protocol).  We recommend not waiting for an
  extended period of time between ``ceph-mon`` upgrades.

* The ops log and usage log for radosgw are now off by default.  If
  you need these logs (e.g., for billing purposes), you must enable
  them explicitly.  For logging of all operations to objects in the
  ``.log`` pool (see ``radosgw-admin log ...``)::

    rgw enable ops log = true

  For usage logging of aggregated bandwidth usage (see ``radosgw-admin
  usage ...``)::

    rgw enable usage log = true

* You should not create or use "format 2" RBD images until after all
  ``ceph-osd`` daemons have been upgraded.  Note that "format 1" is
  still the default.  You can use the new ``ceph osd ls`` and
  ``ceph tell osd.N version`` commands to doublecheck your cluster.
  ``ceph osd ls`` will give a list of all OSD IDs that are part of the
  cluster, and you can use that to write a simple shell loop to display
  all the OSD version strings: ::

      for i in $(ceph osd ls); do
          ceph tell osd.${i} version
      done



Compatibility changes
~~~~~~~~~~~~~~~~~~~~~

* The 'ceph osd create [<uuid>]' command now rejects an argument that
  is not a UUID.  (Previously it would take take an optional integer
  OSD id.)  This correct syntax has been 'ceph osd create [<uuid>]'
  since v0.47, but the older calling convention was being silently
  ignored.

* The CRUSH map root nodes now have type ``root`` instead of type
  ``pool``.  This avoids confusion with RADOS pools, which are not
  directly related.  Any scripts or tools that use the ``ceph osd
  crush ...`` commands may need to be adjusted accordingly.

* The ``ceph osd pool create <poolname> <pgnum>`` command now requires
  the ``pgnum`` argument. Previously this was optional, and would
  default to 8, which was almost never a good number.

* Degraded mode (when there fewer than the desired number of replicas)
  is now more configurable on a per-pool basis, with the min_size
  parameter. By default, with min_size 0, this allows I/O to objects
  with N - floor(N/2) replicas, where N is the total number of
  expected copies. Argonaut behavior was equivalent to having min_size
  = 1, so I/O would always be possible if any completely up to date
  copy remained. min_size = 1 could result in lower overall
  availability in certain cases, such as flapping network partitions.

* 'rbd lock list' and 'rbd showmapped' no longer use tabs as
  separators in their output.

* There is configurable limit on the number of PGs when creating a new
  pool, to prevent a user from accidentally specifying a ridiculous
  number for pg_num.  It can be adjusted via the 'mon max pool pg num'
  option on the monitor, and defaults to 65536 (the current max
  supported by the Linux kernel client).

* The osd capabilities associated with a rados user have changed
  syntax since 0.48 argonaut. The new format is mostly backwards
  compatible, but there are two backwards-incompatible changes:

  * specifying a list of pools in one grant, i.e.
    'allow r pool=foo,bar' is now done in separate grants, i.e.
    'allow r pool=foo, allow r pool=bar'.

  * restricting pool access by pool owner ('allow r uid=foo') is
    removed. This feature was not very useful and unused in practice.

  The new format is documented in the ceph-authtool man page.

* 'rbd cp' and 'rbd rename' use rbd as the default destination pool,
  regardless of what pool the source image is in. Previously they
  would default to the same pool as the source image.

* 'rbd export' no longer prints a message for each object written. It
  just reports percent complete like other long-lasting operations.

* 'ceph osd tree' now uses 4 decimal places for weight so output is
  nicer for humans

* Several monitor operations are now idempotent:

  * ceph osd pool create
  * ceph osd pool delete
  * ceph osd pool mksnap
  * ceph osd rm
  * ceph pg <pgid> revert

Notable changes
~~~~~~~~~~~~~~~

* auth: enable cephx by default
* auth: expanded authentication settings for greater flexibility
* auth: sign messages when using cephx
* build fixes for Fedora 18, CentOS/RHEL 6
* ceph: new 'osd ls' and 'osd tell <osd.N> version' commands
* ceph-debugpack: misc improvements
* ceph-disk-prepare: creates and labels GPT partitions
* ceph-disk-prepare: support for external journals, default mount/mkfs options, etc.
* ceph-fuse/libcephfs: many misc fixes, admin socket debugging
* ceph-fuse: fix handling for .. in root directory
* ceph-fuse: many fixes (including memory leaks, hangs)
* ceph-fuse: mount helper (mount.fuse.ceph) for use with /etc/fstab
* ceph.spec: misc packaging fixes
* common: thread pool sizes can now be adjusted at runtime
* config: $pid is now available as a metavariable
* crush: default root of tree type is now 'root' instead of 'pool' (to avoid confusiong wrt rados pools)
* crush: fixed retry behavior with chooseleaf via tunable
* crush: tunables documented; feature bit now present and enforced
* libcephfs: java wrapper
* librados: several bug fixes (rare races, locking errors)
* librados: some locking fixes
* librados: watch/notify fixes, misc memory leaks
* librbd: a few fixes to 'discard' support
* librbd: fine-grained striping feature
* librbd: fixed memory leaks
* librbd: fully functional and documented image cloning
* librbd: image (advisory) locking
* librbd: improved caching (of object non-existence)
* librbd: 'flatten' command to sever clone parent relationship
* librbd: 'protect'/'unprotect' commands to prevent clone parent from being deleted
* librbd: clip requests past end-of-image.
* librbd: fixes an issue with some windows guests running in qemu (remove floating point usage)
* log: fix in-memory buffering behavior (to only write log messages on crash)
* mds: fix ino release on abort session close, relative getattr path, mds shutdown, other misc items
* mds: misc fixes
* mkcephfs: fix for default keyring, osd data/journal locations
* mkcephfs: support for formatting xfs, ext4 (as well as btrfs)
* init: support for automatically mounting xfs and ext4 osd data directories
* mon, radosgw, ceph-fuse: fixed memory leaks
* mon: improved ENOSPC, fs error checking
* mon: less-destructive ceph-mon --mkfs behavior
* mon: misc fixes
* mon: more informative info about stuck PGs in 'health detail'
* mon: information about recovery and backfill in 'pg <pgid> query'
* mon: new 'osd crush create-or-move ...' command
* mon: new 'osd crush move ...' command lets you rearrange your CRUSH hierarchy
* mon: optionally dump 'osd tree' in json
* mon: configurable cap on maximum osd number (mon max osd)
* mon: many bug fixes (various races causing ceph-mon crashes)
* mon: new on-disk metadata to facilitate future mon changes (post-bobtail)
* mon: election bug fixes
* mon: throttle client messages (limit memory consumption)
* mon: throttle osd flapping based on osd history (limits osdmap ΄thrashing' on overloaded or unhappy clusters)
* mon: 'report' command for dumping detailed cluster status (e.g., for use when reporting bugs)
* mon: osdmap flags like noup, noin now cause a health warning
* msgr: improved failure handling code
* msgr: many bug fixes
* osd, mon: honor new 'nobackfill' and 'norecover' osdmap flags
* osd, mon: use feature bits to lock out clients lacking CRUSH tunables when they are in use
* osd: backfill reservation framework (to avoid flooding new osds with backfill data)
* osd: backfill target reservations (improve performance during recovery)
* osd: better tracking of recent slow operations
* osd: capability grammar improvements, bug fixes
* osd: client vs recovery io prioritization
* osd: crush performance improvements
* osd: default journal size to 5 GB
* osd: experimental support for PG "splitting" (pg_num adjustment for existing pools)
* osd: fix memory leak on certain error paths
* osd: fixed detection of EIO errors from fs on read
* osd: major refactor of PG peering and threading
* osd: many bug fixes
* osd: more/better dump info about in-progress operations
* osd: new caps structure (see compatibility notes)
* osd: new 'deep scrub' will compare object content across replicas (once per week by default)
* osd: new 'lock' rados class for generic object locking
* osd: optional 'min' pg size
* osd: recovery reservations
* osd: scrub efficiency improvement
* osd: several out of order reply bug fixes
* osd: several rare peering cases fixed
* osd: some performance improvements related to request queuing
* osd: use entire device if journal is a block device
* osd: use syncfs(2) when kernel supports it, even if glibc does not
* osd: various fixes for out-of-order op replies
* rados: ability to copy, rename pools
* rados: bench command now cleans up after itself
* rados: 'cppool' command to copy rados pools
* rados: 'rm' now accepts a list of objects to be removed
* radosgw: POST support
* radosgw: REST API for managing usage stats
* radosgw: fix bug in bucket stat updates
* radosgw: fix copy-object vs attributes
* radosgw: fix range header for large objects, ETag quoting, GMT dates, other compatibility fixes
* radosgw: improved garbage collection framework
* radosgw: many small fixes, cleanups
* radosgw: openstack keystone integration
* radosgw: stripe large (non-multipart) objects
* radosgw: support for multi-object deletes
* radosgw: support for swift manifest objects
* radosgw: vanity bucket dns names
* radosgw: various API compatibility fixes
* rbd: import from stdin, export to stdout
* rbd: new 'ls -l' option to view images with metadata
* rbd: use generic id and keyring options for 'rbd map'
* rbd: don't issue usage on errors
* udev: fix symlink creation for rbd images containing partitions
* upstart: job files for all daemon types (not enabled by default)
* wireshark: ceph protocol dissector patch updated


v0.54
-----

Upgrading
~~~~~~~~~

* The osd capabilities associated with a rados user have changed
  syntax since 0.48 argonaut. The new format is mostly backwards
  compatible, but there are two backwards-incompatible changes:

  * specifying a list of pools in one grant, i.e.
    'allow r pool=foo,bar' is now done in separate grants, i.e.
    'allow r pool=foo, allow r pool=bar'.

  * restricting pool access by pool owner ('allow r uid=foo') is
    removed. This feature was not very useful and unused in practice.

  The new format is documented in the ceph-authtool man page.

* Bug fixes to the new osd capability format parsing properly validate
  the allowed operations. If an existing rados user gets permissions
  errors after upgrading, its capabilities were probably
  misconfigured. See the ceph-authtool man page for details on osd
  capabilities.

* 'rbd lock list' and 'rbd showmapped' no longer use tabs as
  separators in their output.


v0.48.2 "argonaut"
------------------

Upgrading
~~~~~~~~~

* The default search path for keyring files now includes /etc/ceph/ceph.$name.keyring.  If such files are present on your cluster, be aware that by default they may now be used.

* There are several changes to the upstart init files.  These have not been previously documented or recommended.  Any existing users should review the changes before upgrading.

* The ceph-disk-prepare and ceph-disk-active scripts have been updated significantly.  These have not been previously documented or recommended.  Any existing users should review the changes before upgrading.

Notable changes
~~~~~~~~~~~~~~~

* mkcephfs: fix keyring generation for mds, osd when default paths are used
* radosgw: fix bug causing occasional corruption of per-bucket stats
* radosgw: workaround to avoid previously corrupted stats from going negative
* radosgw: fix bug in usage stats reporting on busy buckets
* radosgw: fix Content-Range: header for objects bigger than 2 GB.
* rbd: avoid leaving watch acting when command line tool errors out (avoids 30s delay on subsequent operations)
* rbd: friendlier use of --pool/--image options for import (old calling convention still works)
* librbd: fix rare snapshot creation race (could "lose" a snap when creation is concurrent)
* librbd: fix discard handling when spanning holes
* librbd: fix memory leak on discard when caching is enabled
* objecter: misc fixes for op reordering
* objecter: fix for rare startup-time deadlock waiting for osdmap
* ceph: fix usage
* mon: reduce log noise about "check_sub"
* ceph-disk-activate: misc fixes, improvements
* ceph-disk-prepare: partition and format osd disks automatically
* upstart: start everyone on a reboot
* upstart: always update the osd crush location on start if specified in the config
* config: add /etc/ceph/ceph.$name.keyring to default keyring search path
* ceph.spec: don't package crush headers

For more detailed information, see :download:`the complete changelog <changelog/v0.48.2argonaut.txt>`.

v0.48.1 "argonaut"
------------------

Upgrading
~~~~~~~~~

* The radosgw usage trim function was effectively broken in v0.48.  Earlier it would remove more usage data than what was requested.  This is fixed in v0.48.1, but the fix is incompatible.  The v0.48 radosgw-admin tool cannot be used to initiate the trimming; please use the v0.48.1 version.

* v0.48.1 now explicitly indicates support for the CRUSH_TUNABLES feature.  No other version of Ceph requires this, yet, but future versions will when the tunables are adjusted from their historical defaults.

* There are no other compatibility changes between v0.48.1 and v0.48.

Notable changes
~~~~~~~~~~~~~~~

* mkcephfs: use default 'keyring', 'osd data', 'osd journal' paths when not specified in conf
* msgr: various fixes to socket error handling
* osd: reduce scrub overhead
* osd: misc peering fixes (past_interval sharing, pgs stuck in 'peering' states)
* osd: fail on EIO in read path (do not silently ignore read errors from failing disks)
* osd: avoid internal heartbeat errors by breaking some large transactions into pieces
* osd: fix osdmap catch-up during startup (catch up and then add daemon to osdmap)
* osd: fix spurious 'misdirected op' messages
* osd: report scrub status via 'pg ... query'
* rbd: fix race when watch registrations are resent
* rbd: fix rbd image id assignment scheme (new image data objects have slightly different names)
* rbd: fix perf stats for cache hit rate
* rbd tool: fix off-by-one in key name (crash when empty key specified)
* rbd: more robust udev rules
* rados tool: copy object, pool commands
* radosgw: fix in usage stats trimming
* radosgw: misc API compatibility fixes (date strings, ETag quoting, swift headers, etc.)
* ceph-fuse: fix locking in read/write paths
* mon: fix rare race corrupting on-disk data
* config: fix admin socket 'config set' command
* log: fix in-memory log event gathering
* debian: remove crush headers, include librados-config
* rpm: add ceph-disk-{activate, prepare}

For more detailed information, see :download:`the complete changelog <changelog/v0.48.1argonaut.txt>`.

v0.48 "argonaut"
----------------

Upgrading
~~~~~~~~~

* This release includes a disk format upgrade.  Each ceph-osd daemon, upon startup, will migrate its locally stored data to the new format.  This process can take a while (for large object counts, even hours), especially on non-btrfs file systems.  

* To keep the cluster available while the upgrade is in progress, we recommend you upgrade a storage node or rack at a time, and wait for the cluster to recover each time.  To prevent the cluster from moving data around in response to the OSD daemons being down for minutes or hours, you may want to::

    ceph osd set noout

  This will prevent the cluster from marking down OSDs as "out" and re-replicating the data elsewhere. If you do this, be sure to clear the flag when the upgrade is complete::

    ceph osd unset noout

* There is a encoding format change internal to the monitor cluster. The monitor daemons are careful to switch to the new format only when all members of the quorum support it.  However, that means that a partial quorum with new code may move to the new format, and a recovering monitor running old code will be unable to join (it will crash).  If this occurs, simply upgrading the remaining monitor will resolve the problem.

* The ceph tool's -s and -w commands from previous versions are incompatible with this version. Upgrade your client tools at the same time you upgrade the monitors if you rely on those commands.

* It is not possible to downgrade from v0.48 to a previous version.

Notable changes
~~~~~~~~~~~~~~~

* osd: stability improvements
* osd: capability model simplification
* osd: simpler/safer --mkfs (no longer removes all files; safe to re-run on active osd)
* osd: potentially buggy FIEMAP behavior disabled by default
* rbd: caching improvements
* rbd: improved instrumentation
* rbd: bug fixes
* radosgw: new, scalable usage logging infrastructure
* radosgw: per-user bucket limits
* mon: streamlined process for setting up authentication keys
* mon: stability improvements
* mon: log message throttling
* doc: improved documentation (ceph, rbd, radosgw, chef, etc.)
* config: new default locations for daemon keyrings
* config: arbitrary variable substitutions
* improved 'admin socket' daemon admin interface (ceph --admin-daemon ...)
* chef: support for multiple monitor clusters
* upstart: basic support for monitors, mds, radosgw; osd support still a work in progress.

The new default keyring locations mean that when enabling authentication (``auth supported = cephx``), keyring locations do not need to be specified if the keyring file is located inside the daemon's data directory (``/var/lib/ceph/$type/ceph-$id`` by default).

There is also a lot of librbd code in this release that is laying the groundwork for the upcoming layering functionality, but is not actually used. Likewise, the upstart support is still incomplete and not recommended; we will backport that functionality later if it turns out to be non-disruptive.