1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
|
=======================
Config Settings
=======================
See `Block Device`_ for additional details.
Generic IO Settings
===================
.. confval:: rbd_compression_hint
.. confval:: rbd_read_from_replica_policy
.. confval:: rbd_default_order
Cache Settings
=======================
.. sidebar:: Kernel Caching
The kernel driver for Ceph block devices can use the Linux page cache to
improve performance.
The user space implementation of the Ceph block device (i.e., ``librbd``) cannot
take advantage of the Linux page cache, so it includes its own in-memory
caching, called "RBD caching." RBD caching behaves just like well-behaved hard
disk caching. When the OS sends a barrier or a flush request, all dirty data is
written to the OSDs. This means that using write-back caching is just as safe as
using a well-behaved physical hard disk with a VM that properly sends flushes
(i.e. Linux kernel >= 2.6.32). The cache uses a Least Recently Used (LRU)
algorithm, and in write-back mode it can coalesce contiguous requests for
better throughput.
The librbd cache is enabled by default and supports three different cache
policies: write-around, write-back, and write-through. Writes return
immediately under both the write-around and write-back policies, unless there
are more than ``rbd_cache_max_dirty`` unwritten bytes to the storage cluster.
The write-around policy differs from the write-back policy in that it does
not attempt to service read requests from the cache, unlike the write-back
policy, and is therefore faster for high performance write workloads. Under the
write-through policy, writes return only when the data is on disk on all
replicas, but reads may come from the cache.
Prior to receiving a flush request, the cache behaves like a write-through cache
to ensure safe operation for older operating systems that do not send flushes to
ensure crash consistent behavior.
If the librbd cache is disabled, writes and
reads go directly to the storage cluster, and writes return only when the data
is on disk on all replicas.
.. note::
The cache is in memory on the client, and each RBD image has
its own. Since the cache is local to the client, there's no coherency
if there are others accessing the image. Running GFS or OCFS on top of
RBD will not work with caching enabled.
Option settings for RBD should be set in the ``[client]``
section of your configuration file or the central config store. These settings
include:
.. confval:: rbd_cache
.. confval:: rbd_cache_policy
.. confval:: rbd_cache_writethrough_until_flush
.. confval:: rbd_cache_size
.. confval:: rbd_cache_max_dirty
.. confval:: rbd_cache_target_dirty
.. confval:: rbd_cache_max_dirty_age
.. _Block Device: ../../rbd
Read-ahead Settings
=======================
librbd supports read-ahead/prefetching to optimize small, sequential reads.
This should normally be handled by the guest OS in the case of a VM,
but boot loaders may not issue efficient reads. Read-ahead is automatically
disabled if caching is disabled or if the policy is write-around.
.. confval:: rbd_readahead_trigger_requests
.. confval:: rbd_readahead_max_bytes
.. confval:: rbd_readahead_disable_after_bytes
Image Features
==============
RBD supports advanced features which can be specified via the command line when
creating images or the default features can be configured via
``rbd_default_features = <sum of feature numeric values>`` or
``rbd_default_features = <comma-delimited list of CLI values>``.
``Layering``
:Description: Layering enables cloning.
:Internal value: 1
:CLI value: layering
:Added in: v0.52 (Bobtail)
:KRBD support: since v3.10
:Default: yes
``Striping v2``
:Description: Striping spreads data across multiple objects. Striping helps with
parallelism for sequential read/write workloads.
:Internal value: 2
:CLI value: striping
:Added in: v0.55 (Bobtail)
:KRBD support: since v3.10 (default striping only, "fancy" striping added in v4.17)
:Default: yes
``Exclusive locking``
:Description: When enabled, it requires a client to acquire a lock on an object
before making a write. Exclusive lock should only be enabled when
a single client is accessing an image at any given time.
:Internal value: 4
:CLI value: exclusive-lock
:Added in: v0.92 (Hammer)
:KRBD support: since v4.9
:Default: yes
``Object map``
:Description: Object map support depends on exclusive lock support. Block
devices are thin provisioned, which means that they only store
data that actually has been written, ie. they are *sparse*. Object
map support helps track which objects actually exist (have data
stored on a device). Enabling object map support speeds up I/O
operations for cloning, importing and exporting a sparsely
populated image, and deleting.
:Internal value: 8
:CLI value: object-map
:Added in: v0.93 (Hammer)
:KRBD support: since v5.3
:Default: yes
``Fast-diff``
:Description: Fast-diff support depends on object map support and exclusive lock
support. It adds another property to the object map, which makes
it much faster to generate diffs between snapshots of an image.
It is also much faster to calculate the actual data usage of a
snapshot or volume (``rbd du``).
:Internal value: 16
:CLI value: fast-diff
:Added in: v9.0.1 (Infernalis)
:KRBD support: since v5.3
:Default: yes
``Deep-flatten``
:Description: Deep-flatten enables ``rbd flatten`` to work on all snapshots of
an image, in addition to the image itself. Without it, snapshots
of an image will still rely on the parent, so the parent cannot be
deleted until the snapshots are first deleted. Deep-flatten makes
a parent independent of its clones, even if they have snapshots,
at the expense of using additional OSD device space.
:Internal value: 32
:CLI value: deep-flatten
:Added in: v9.0.2 (Infernalis)
:KRBD support: since v5.1
:Default: yes
``Journaling``
:Description: Journaling support depends on exclusive lock support. Journaling
records all modifications to an image in the order they occur. RBD
mirroring can utilize the journal to replicate a crash-consistent
image to a remote cluster. It is best to let ``rbd-mirror``
manage this feature only as needed, as enabling it long term may
result in substantial additional OSD space consumption.
:Internal value: 64
:CLI value: journaling
:Added in: v10.0.1 (Jewel)
:KRBD support: no
:Default: no
``Data pool``
:Description: On erasure-coded pools, the image data block objects need to be stored on a separate pool from the image metadata.
:Internal value: 128
:Added in: v11.1.0 (Kraken)
:KRBD support: since v4.11
:Default: no
``Operations``
:Description: Used to restrict older clients from performing certain maintenance operations against an image (e.g. clone, snap create).
:Internal value: 256
:Added in: v13.0.2 (Mimic)
:KRBD support: since v4.16
``Migrating``
:Description: Used to restrict older clients from opening an image when it is in migration state.
:Internal value: 512
:Added in: v14.0.1 (Nautilus)
:KRBD support: no
``Non-primary``
:Description: Used to restrict changes to non-primary images using snapshot-based mirroring.
:Internal value: 1024
:Added in: v15.2.0 (Octopus)
:KRBD support: no
QoS Settings
============
librbd supports limiting per-image IO in several ways. These all apply
to a given image within a given process - the same image used in
multiple places, e.g. two separate VMs, would have independent limits.
* **IOPS:** number of I/Os per second (any type of I/O)
* **read IOPS:** number of read I/Os per second
* **write IOPS:** number of write I/Os per second
* **bps:** bytes per second (any type of I/O)
* **read bps:** bytes per second read
* **write bps:** bytes per second written
Each of these limits operates independently of each other. They are
all off by default. Every type of limit throttles I/O using a token
bucket algorithm, with the ability to configure the limit (average
speed over time) and potential for a higher rate (a burst) for a short
period of time (burst_seconds). When any of these limits is reached,
and there is no burst capacity left, librbd reduces the rate of that
type of I/O to the limit.
For example, if a read bps limit of 100MB was configured, but writes
were not limited, writes could proceed as quickly as possible, while
reads would be throttled to 100MB/s on average. If a read bps burst of
150MB was set, and read burst seconds was set to five seconds, reads
could proceed at 150MB/s for up to five seconds before dropping back
to the 100MB/s limit.
The following options configure these throttles:
.. confval:: rbd_qos_iops_limit
.. confval:: rbd_qos_iops_burst
.. confval:: rbd_qos_iops_burst_seconds
.. confval:: rbd_qos_read_iops_limit
.. confval:: rbd_qos_read_iops_burst
.. confval:: rbd_qos_read_iops_burst_seconds
.. confval:: rbd_qos_write_iops_limit
.. confval:: rbd_qos_write_iops_burst
.. confval:: rbd_qos_write_iops_burst_seconds
.. confval:: rbd_qos_bps_limit
.. confval:: rbd_qos_bps_burst
.. confval:: rbd_qos_bps_burst_seconds
.. confval:: rbd_qos_read_bps_limit
.. confval:: rbd_qos_read_bps_burst
.. confval:: rbd_qos_read_bps_burst_seconds
.. confval:: rbd_qos_write_bps_limit
.. confval:: rbd_qos_write_bps_burst
.. confval:: rbd_qos_write_bps_burst_seconds
.. confval:: rbd_qos_schedule_tick_min
.. confval:: rbd_qos_exclude_ops
|