]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | ================== |
2 | Placement Groups | |
3 | ================== | |
4 | ||
11fdf7f2 TL |
5 | .. _pg-autoscaler: |
6 | ||
7 | Autoscaling placement groups | |
8 | ============================ | |
9 | ||
10 | Placement groups (PGs) are an internal implementation detail of how | |
11 | Ceph distributes data. You can allow the cluster to either make | |
12 | recommendations or automatically tune PGs based on how the cluster is | |
13 | used by enabling *pg-autoscaling*. | |
14 | ||
15 | Each pool in the system has a ``pg_autoscale_mode`` property that can be set to ``off``, ``on``, or ``warn``. | |
16 | ||
17 | * ``off``: Disable autoscaling for this pool. It is up to the administrator to choose an appropriate PG number for each pool. Please refer to :ref:`choosing-number-of-placement-groups` for more information. | |
18 | * ``on``: Enable automated adjustments of the PG count for the given pool. | |
19 | * ``warn``: Raise health alerts when the PG count should be adjusted | |
20 | ||
21 | To set the autoscaling mode for existing pools,:: | |
22 | ||
23 | ceph osd pool set <pool-name> pg_autoscale_mode <mode> | |
24 | ||
25 | For example to enable autoscaling on pool ``foo``,:: | |
26 | ||
27 | ceph osd pool set foo pg_autoscale_mode on | |
28 | ||
29 | You can also configure the default ``pg_autoscale_mode`` that is | |
30 | applied to any pools that are created in the future with:: | |
31 | ||
32 | ceph config set global osd_pool_default_autoscale_mode <mode> | |
33 | ||
34 | Viewing PG scaling recommendations | |
35 | ---------------------------------- | |
36 | ||
37 | You can view each pool, its relative utilization, and any suggested changes to | |
38 | the PG count with this command:: | |
39 | ||
40 | ceph osd pool autoscale-status | |
41 | ||
42 | Output will be something like:: | |
43 | ||
44 | POOL SIZE TARGET SIZE RATE RAW CAPACITY RATIO TARGET RATIO PG_NUM NEW PG_NUM AUTOSCALE | |
45 | a 12900M 3.0 82431M 0.4695 8 128 warn | |
46 | c 0 3.0 82431M 0.0000 0.2000 1 64 warn | |
47 | b 0 953.6M 3.0 82431M 0.0347 8 warn | |
48 | ||
49 | **SIZE** is the amount of data stored in the pool. **TARGET SIZE**, if | |
50 | present, is the amount of data the administrator has specified that | |
51 | they expect to eventually be stored in this pool. The system uses | |
52 | the larger of the two values for its calculation. | |
53 | ||
54 | **RATE** is the multiplier for the pool that determines how much raw | |
55 | storage capacity is consumed. For example, a 3 replica pool will | |
56 | have a ratio of 3.0, while a k=4,m=2 erasure coded pool will have a | |
57 | ratio of 1.5. | |
58 | ||
59 | **RAW CAPACITY** is the total amount of raw storage capacity on the | |
60 | OSDs that are responsible for storing this pool's (and perhaps other | |
61 | pools') data. **RATIO** is the ratio of that total capacity that | |
62 | this pool is consuming (i.e., ratio = size * rate / raw capacity). | |
63 | ||
64 | **TARGET RATIO**, if present, is the ratio of storage that the | |
65 | administrator has specified that they expect this pool to consume. | |
66 | The system uses the larger of the actual ratio and the target ratio | |
67 | for its calculation. If both target size bytes and ratio are specified, the | |
68 | ratio takes precedence. | |
69 | ||
70 | **PG_NUM** is the current number of PGs for the pool (or the current | |
71 | number of PGs that the pool is working towards, if a ``pg_num`` | |
72 | change is in progress). **NEW PG_NUM**, if present, is what the | |
73 | system believes the pool's ``pg_num`` should be changed to. It is | |
74 | always a power of 2, and will only be present if the "ideal" value | |
75 | varies from the current value by more than a factor of 3. | |
76 | ||
77 | The final column, **AUTOSCALE**, is the pool ``pg_autoscale_mode``, | |
78 | and will be either ``on``, ``off``, or ``warn``. | |
79 | ||
80 | ||
81 | Automated scaling | |
82 | ----------------- | |
83 | ||
84 | Allowing the cluster to automatically scale PGs based on usage is the | |
85 | simplest approach. Ceph will look at the total available storage and | |
86 | target number of PGs for the whole system, look at how much data is | |
87 | stored in each pool, and try to apportion the PGs accordingly. The | |
88 | system is relatively conservative with its approach, only making | |
89 | changes to a pool when the current number of PGs (``pg_num``) is more | |
90 | than 3 times off from what it thinks it should be. | |
91 | ||
92 | The target number of PGs per OSD is based on the | |
93 | ``mon_target_pg_per_osd`` configurable (default: 100), which can be | |
94 | adjusted with:: | |
95 | ||
96 | ceph config set global mon_target_pg_per_osd 100 | |
97 | ||
98 | The autoscaler analyzes pools and adjusts on a per-subtree basis. | |
99 | Because each pool may map to a different CRUSH rule, and each rule may | |
100 | distribute data across different devices, Ceph will consider | |
101 | utilization of each subtree of the hierarchy independently. For | |
102 | example, a pool that maps to OSDs of class `ssd` and a pool that maps | |
103 | to OSDs of class `hdd` will each have optimal PG counts that depend on | |
104 | the number of those respective device types. | |
105 | ||
106 | ||
107 | .. _specifying_pool_target_size: | |
108 | ||
109 | Specifying expected pool size | |
110 | ----------------------------- | |
111 | ||
112 | When a cluster or pool is first created, it will consume a small | |
113 | fraction of the total cluster capacity and will appear to the system | |
114 | as if it should only need a small number of placement groups. | |
115 | However, in most cases cluster administrators have a good idea which | |
116 | pools are expected to consume most of the system capacity over time. | |
117 | By providing this information to Ceph, a more appropriate number of | |
118 | PGs can be used from the beginning, preventing subsequent changes in | |
119 | ``pg_num`` and the overhead associated with moving data around when | |
120 | those adjustments are made. | |
121 | ||
122 | The *target size** of a pool can be specified in two ways: either in | |
123 | terms of the absolute size of the pool (i.e., bytes), or as a ratio of | |
124 | the total cluster capacity. | |
125 | ||
126 | For example,:: | |
127 | ||
128 | ceph osd pool set mypool target_size_bytes 100T | |
129 | ||
130 | will tell the system that `mypool` is expected to consume 100 TiB of | |
131 | space. Alternatively,:: | |
132 | ||
133 | ceph osd pool set mypool target_size_ratio .9 | |
134 | ||
135 | will tell the system that `mypool` is expected to consume 90% of the | |
136 | total cluster capacity. | |
137 | ||
138 | You can also set the target size of a pool at creation time with the optional ``--target-size-bytes <bytes>`` or ``--target-size-ratio <ratio>`` arguments to the ``ceph osd pool create`` command. | |
139 | ||
140 | Note that if impossible target size values are specified (for example, | |
141 | a capacity larger than the total cluster, or ratio(s) that sum to more | |
142 | than 1.0) then a health warning | |
143 | (``POOL_TARET_SIZE_RATIO_OVERCOMMITTED`` or | |
144 | ``POOL_TARGET_SIZE_BYTES_OVERCOMMITTED``) will be raised. | |
145 | ||
146 | Specifying bounds on a pool's PGs | |
147 | --------------------------------- | |
148 | ||
149 | It is also possible to specify a minimum number of PGs for a pool. | |
150 | This is useful for establishing a lower bound on the amount of | |
151 | parallelism client will see when doing IO, even when a pool is mostly | |
152 | empty. Setting the lower bound prevents Ceph from reducing (or | |
153 | recommending you reduce) the PG number below the configured number. | |
154 | ||
155 | You can set the minimum number of PGs for a pool with:: | |
156 | ||
157 | ceph osd pool set <pool-name> pg_num_min <num> | |
158 | ||
159 | You can also specify the minimum PG count at pool creation time with | |
160 | the optional ``--pg-num-min <num>`` argument to the ``ceph osd pool | |
161 | create`` command. | |
162 | ||
7c673cae FG |
163 | .. _preselection: |
164 | ||
165 | A preselection of pg_num | |
166 | ======================== | |
167 | ||
168 | When creating a new pool with:: | |
169 | ||
170 | ceph osd pool create {pool-name} pg_num | |
171 | ||
11fdf7f2 | 172 | it is mandatory to choose the value of ``pg_num`` because it cannot (currently) be |
7c673cae FG |
173 | calculated automatically. Here are a few values commonly used: |
174 | ||
175 | - Less than 5 OSDs set ``pg_num`` to 128 | |
176 | ||
177 | - Between 5 and 10 OSDs set ``pg_num`` to 512 | |
178 | ||
179 | - Between 10 and 50 OSDs set ``pg_num`` to 1024 | |
180 | ||
181 | - If you have more than 50 OSDs, you need to understand the tradeoffs | |
182 | and how to calculate the ``pg_num`` value by yourself | |
183 | ||
11fdf7f2 | 184 | - For calculating ``pg_num`` value by yourself please take help of `pgcalc`_ tool |
7c673cae | 185 | |
11fdf7f2 | 186 | As the number of OSDs increases, choosing the right value for pg_num |
7c673cae FG |
187 | becomes more important because it has a significant influence on the |
188 | behavior of the cluster as well as the durability of the data when | |
189 | something goes wrong (i.e. the probability that a catastrophic event | |
190 | leads to data loss). | |
191 | ||
192 | How are Placement Groups used ? | |
193 | =============================== | |
194 | ||
195 | A placement group (PG) aggregates objects within a pool because | |
196 | tracking object placement and object metadata on a per-object basis is | |
197 | computationally expensive--i.e., a system with millions of objects | |
198 | cannot realistically track placement on a per-object basis. | |
199 | ||
200 | .. ditaa:: | |
201 | /-----\ /-----\ /-----\ /-----\ /-----\ | |
202 | | obj | | obj | | obj | | obj | | obj | | |
203 | \-----/ \-----/ \-----/ \-----/ \-----/ | |
204 | | | | | | | |
205 | +--------+--------+ +---+----+ | |
206 | | | | |
207 | v v | |
208 | +-----------------------+ +-----------------------+ | |
209 | | Placement Group #1 | | Placement Group #2 | | |
210 | | | | | | |
211 | +-----------------------+ +-----------------------+ | |
212 | | | | |
213 | +------------------------------+ | |
214 | | | |
215 | v | |
216 | +-----------------------+ | |
217 | | Pool | | |
218 | | | | |
219 | +-----------------------+ | |
220 | ||
221 | The Ceph client will calculate which placement group an object should | |
222 | be in. It does this by hashing the object ID and applying an operation | |
223 | based on the number of PGs in the defined pool and the ID of the pool. | |
224 | See `Mapping PGs to OSDs`_ for details. | |
225 | ||
226 | The object's contents within a placement group are stored in a set of | |
227 | OSDs. For instance, in a replicated pool of size two, each placement | |
228 | group will store objects on two OSDs, as shown below. | |
229 | ||
230 | .. ditaa:: | |
231 | ||
232 | +-----------------------+ +-----------------------+ | |
233 | | Placement Group #1 | | Placement Group #2 | | |
234 | | | | | | |
235 | +-----------------------+ +-----------------------+ | |
236 | | | | | | |
237 | v v v v | |
238 | /----------\ /----------\ /----------\ /----------\ | |
239 | | | | | | | | | | |
240 | | OSD #1 | | OSD #2 | | OSD #2 | | OSD #3 | | |
241 | | | | | | | | | | |
242 | \----------/ \----------/ \----------/ \----------/ | |
243 | ||
244 | ||
245 | Should OSD #2 fail, another will be assigned to Placement Group #1 and | |
246 | will be filled with copies of all objects in OSD #1. If the pool size | |
247 | is changed from two to three, an additional OSD will be assigned to | |
248 | the placement group and will receive copies of all objects in the | |
249 | placement group. | |
250 | ||
11fdf7f2 | 251 | Placement groups do not own the OSD; they share it with other |
7c673cae FG |
252 | placement groups from the same pool or even other pools. If OSD #2 |
253 | fails, the Placement Group #2 will also have to restore copies of | |
254 | objects, using OSD #3. | |
255 | ||
256 | When the number of placement groups increases, the new placement | |
257 | groups will be assigned OSDs. The result of the CRUSH function will | |
258 | also change and some objects from the former placement groups will be | |
259 | copied over to the new Placement Groups and removed from the old ones. | |
260 | ||
261 | Placement Groups Tradeoffs | |
262 | ========================== | |
263 | ||
264 | Data durability and even distribution among all OSDs call for more | |
265 | placement groups but their number should be reduced to the minimum to | |
266 | save CPU and memory. | |
267 | ||
268 | .. _data durability: | |
269 | ||
270 | Data durability | |
271 | --------------- | |
272 | ||
273 | After an OSD fails, the risk of data loss increases until the data it | |
274 | contained is fully recovered. Let's imagine a scenario that causes | |
275 | permanent data loss in a single placement group: | |
276 | ||
277 | - The OSD fails and all copies of the object it contains are lost. | |
278 | For all objects within the placement group the number of replica | |
11fdf7f2 | 279 | suddenly drops from three to two. |
7c673cae | 280 | |
11fdf7f2 | 281 | - Ceph starts recovery for this placement group by choosing a new OSD |
7c673cae FG |
282 | to re-create the third copy of all objects. |
283 | ||
284 | - Another OSD, within the same placement group, fails before the new | |
285 | OSD is fully populated with the third copy. Some objects will then | |
286 | only have one surviving copies. | |
287 | ||
288 | - Ceph picks yet another OSD and keeps copying objects to restore the | |
289 | desired number of copies. | |
290 | ||
291 | - A third OSD, within the same placement group, fails before recovery | |
292 | is complete. If this OSD contained the only remaining copy of an | |
293 | object, it is permanently lost. | |
294 | ||
295 | In a cluster containing 10 OSDs with 512 placement groups in a three | |
296 | replica pool, CRUSH will give each placement groups three OSDs. In the | |
297 | end, each OSDs will end up hosting (512 * 3) / 10 = ~150 Placement | |
298 | Groups. When the first OSD fails, the above scenario will therefore | |
299 | start recovery for all 150 placement groups at the same time. | |
300 | ||
301 | The 150 placement groups being recovered are likely to be | |
302 | homogeneously spread over the 9 remaining OSDs. Each remaining OSD is | |
303 | therefore likely to send copies of objects to all others and also | |
304 | receive some new objects to be stored because they became part of a | |
305 | new placement group. | |
306 | ||
307 | The amount of time it takes for this recovery to complete entirely | |
308 | depends on the architecture of the Ceph cluster. Let say each OSD is | |
309 | hosted by a 1TB SSD on a single machine and all of them are connected | |
310 | to a 10Gb/s switch and the recovery for a single OSD completes within | |
311 | M minutes. If there are two OSDs per machine using spinners with no | |
312 | SSD journal and a 1Gb/s switch, it will at least be an order of | |
313 | magnitude slower. | |
314 | ||
315 | In a cluster of this size, the number of placement groups has almost | |
316 | no influence on data durability. It could be 128 or 8192 and the | |
317 | recovery would not be slower or faster. | |
318 | ||
319 | However, growing the same Ceph cluster to 20 OSDs instead of 10 OSDs | |
320 | is likely to speed up recovery and therefore improve data durability | |
321 | significantly. Each OSD now participates in only ~75 placement groups | |
322 | instead of ~150 when there were only 10 OSDs and it will still require | |
323 | all 19 remaining OSDs to perform the same amount of object copies in | |
324 | order to recover. But where 10 OSDs had to copy approximately 100GB | |
325 | each, they now have to copy 50GB each instead. If the network was the | |
326 | bottleneck, recovery will happen twice as fast. In other words, | |
327 | recovery goes faster when the number of OSDs increases. | |
328 | ||
329 | If this cluster grows to 40 OSDs, each of them will only host ~35 | |
330 | placement groups. If an OSD dies, recovery will keep going faster | |
331 | unless it is blocked by another bottleneck. However, if this cluster | |
332 | grows to 200 OSDs, each of them will only host ~7 placement groups. If | |
333 | an OSD dies, recovery will happen between at most of ~21 (7 * 3) OSDs | |
334 | in these placement groups: recovery will take longer than when there | |
335 | were 40 OSDs, meaning the number of placement groups should be | |
336 | increased. | |
337 | ||
338 | No matter how short the recovery time is, there is a chance for a | |
339 | second OSD to fail while it is in progress. In the 10 OSDs cluster | |
340 | described above, if any of them fail, then ~17 placement groups | |
341 | (i.e. ~150 / 9 placement groups being recovered) will only have one | |
342 | surviving copy. And if any of the 8 remaining OSD fail, the last | |
343 | objects of two placement groups are likely to be lost (i.e. ~17 / 8 | |
344 | placement groups with only one remaining copy being recovered). | |
345 | ||
346 | When the size of the cluster grows to 20 OSDs, the number of Placement | |
347 | Groups damaged by the loss of three OSDs drops. The second OSD lost | |
348 | will degrade ~4 (i.e. ~75 / 19 placement groups being recovered) | |
349 | instead of ~17 and the third OSD lost will only lose data if it is one | |
350 | of the four OSDs containing the surviving copy. In other words, if the | |
351 | probability of losing one OSD is 0.0001% during the recovery time | |
11fdf7f2 | 352 | frame, it goes from 17 * 10 * 0.0001% in the cluster with 10 OSDs to 4 * 20 * |
7c673cae FG |
353 | 0.0001% in the cluster with 20 OSDs. |
354 | ||
355 | In a nutshell, more OSDs mean faster recovery and a lower risk of | |
356 | cascading failures leading to the permanent loss of a Placement | |
357 | Group. Having 512 or 4096 Placement Groups is roughly equivalent in a | |
358 | cluster with less than 50 OSDs as far as data durability is concerned. | |
359 | ||
360 | Note: It may take a long time for a new OSD added to the cluster to be | |
361 | populated with placement groups that were assigned to it. However | |
362 | there is no degradation of any object and it has no impact on the | |
363 | durability of the data contained in the Cluster. | |
364 | ||
365 | .. _object distribution: | |
366 | ||
367 | Object distribution within a pool | |
368 | --------------------------------- | |
369 | ||
370 | Ideally objects are evenly distributed in each placement group. Since | |
371 | CRUSH computes the placement group for each object, but does not | |
372 | actually know how much data is stored in each OSD within this | |
373 | placement group, the ratio between the number of placement groups and | |
374 | the number of OSDs may influence the distribution of the data | |
375 | significantly. | |
376 | ||
11fdf7f2 | 377 | For instance, if there was a single placement group for ten OSDs in a |
7c673cae FG |
378 | three replica pool, only three OSD would be used because CRUSH would |
379 | have no other choice. When more placement groups are available, | |
380 | objects are more likely to be evenly spread among them. CRUSH also | |
381 | makes every effort to evenly spread OSDs among all existing Placement | |
382 | Groups. | |
383 | ||
384 | As long as there are one or two orders of magnitude more Placement | |
eafe8130 TL |
385 | Groups than OSDs, the distribution should be even. For instance, 256 |
386 | placement groups for 3 OSDs, 512 or 1024 placement groups for 10 OSDs | |
387 | etc. | |
7c673cae FG |
388 | |
389 | Uneven data distribution can be caused by factors other than the ratio | |
390 | between OSDs and placement groups. Since CRUSH does not take into | |
391 | account the size of the objects, a few very large objects may create | |
392 | an imbalance. Let say one million 4K objects totaling 4GB are evenly | |
eafe8130 | 393 | spread among 1024 placement groups on 10 OSDs. They will use 4GB / 10 |
7c673cae FG |
394 | = 400MB on each OSD. If one 400MB object is added to the pool, the |
395 | three OSDs supporting the placement group in which the object has been | |
396 | placed will be filled with 400MB + 400MB = 800MB while the seven | |
397 | others will remain occupied with only 400MB. | |
398 | ||
399 | .. _resource usage: | |
400 | ||
401 | Memory, CPU and network usage | |
402 | ----------------------------- | |
403 | ||
404 | For each placement group, OSDs and MONs need memory, network and CPU | |
405 | at all times and even more during recovery. Sharing this overhead by | |
406 | clustering objects within a placement group is one of the main reasons | |
407 | they exist. | |
408 | ||
409 | Minimizing the number of placement groups saves significant amounts of | |
410 | resources. | |
411 | ||
11fdf7f2 TL |
412 | .. _choosing-number-of-placement-groups: |
413 | ||
7c673cae FG |
414 | Choosing the number of Placement Groups |
415 | ======================================= | |
416 | ||
11fdf7f2 TL |
417 | .. note: It is rarely necessary to do this math by hand. Instead, use the ``ceph osd pool autoscale-status`` command in combination with the ``target_size_bytes`` or ``target_size_ratio`` pool properties. See :ref:`pg-autoscaler` for more information. |
418 | ||
7c673cae FG |
419 | If you have more than 50 OSDs, we recommend approximately 50-100 |
420 | placement groups per OSD to balance out resource usage, data | |
11fdf7f2 | 421 | durability and distribution. If you have less than 50 OSDs, choosing |
7c673cae FG |
422 | among the `preselection`_ above is best. For a single pool of objects, |
423 | you can use the following formula to get a baseline:: | |
424 | ||
425 | (OSDs * 100) | |
426 | Total PGs = ------------ | |
427 | pool size | |
428 | ||
429 | Where **pool size** is either the number of replicas for replicated | |
430 | pools or the K+M sum for erasure coded pools (as returned by **ceph | |
431 | osd erasure-code-profile get**). | |
432 | ||
433 | You should then check if the result makes sense with the way you | |
434 | designed your Ceph cluster to maximize `data durability`_, | |
435 | `object distribution`_ and minimize `resource usage`_. | |
436 | ||
eafe8130 TL |
437 | The result should always be **rounded up to the nearest power of two**. |
438 | ||
439 | Only a power of two will evenly balance the number of objects among | |
440 | placement groups. Other values will result in an uneven distribution of | |
441 | data across your OSDs. Their use should be limited to incrementally | |
442 | stepping from one power of two to another. | |
7c673cae FG |
443 | |
444 | As an example, for a cluster with 200 OSDs and a pool size of 3 | |
445 | replicas, you would estimate your number of PGs as follows:: | |
446 | ||
447 | (200 * 100) | |
448 | ----------- = 6667. Nearest power of 2: 8192 | |
449 | 3 | |
450 | ||
451 | When using multiple data pools for storing objects, you need to ensure | |
452 | that you balance the number of placement groups per pool with the | |
453 | number of placement groups per OSD so that you arrive at a reasonable | |
454 | total number of placement groups that provides reasonably low variance | |
455 | per OSD without taxing system resources or making the peering process | |
456 | too slow. | |
457 | ||
458 | For instance a cluster of 10 pools each with 512 placement groups on | |
459 | ten OSDs is a total of 5,120 placement groups spread over ten OSDs, | |
460 | that is 512 placement groups per OSD. That does not use too many | |
461 | resources. However, if 1,000 pools were created with 512 placement | |
462 | groups each, the OSDs will handle ~50,000 placement groups each and it | |
463 | would require significantly more resources and time for peering. | |
464 | ||
224ce89b WB |
465 | You may find the `PGCalc`_ tool helpful. |
466 | ||
467 | ||
7c673cae FG |
468 | .. _setting the number of placement groups: |
469 | ||
470 | Set the Number of Placement Groups | |
471 | ================================== | |
472 | ||
473 | To set the number of placement groups in a pool, you must specify the | |
474 | number of placement groups at the time you create the pool. | |
11fdf7f2 | 475 | See `Create a Pool`_ for details. Even after a pool is created you can also change the number of placement groups with:: |
7c673cae FG |
476 | |
477 | ceph osd pool set {pool-name} pg_num {pg_num} | |
478 | ||
11fdf7f2 | 479 | After you increase the number of placement groups, you must also |
7c673cae FG |
480 | increase the number of placement groups for placement (``pgp_num``) |
481 | before your cluster will rebalance. The ``pgp_num`` will be the number of | |
482 | placement groups that will be considered for placement by the CRUSH | |
483 | algorithm. Increasing ``pg_num`` splits the placement groups but data | |
484 | will not be migrated to the newer placement groups until placement | |
485 | groups for placement, ie. ``pgp_num`` is increased. The ``pgp_num`` | |
486 | should be equal to the ``pg_num``. To increase the number of | |
487 | placement groups for placement, execute the following:: | |
488 | ||
489 | ceph osd pool set {pool-name} pgp_num {pgp_num} | |
490 | ||
11fdf7f2 TL |
491 | When decreasing the number of PGs, ``pgp_num`` is adjusted |
492 | automatically for you. | |
7c673cae FG |
493 | |
494 | Get the Number of Placement Groups | |
495 | ================================== | |
496 | ||
497 | To get the number of placement groups in a pool, execute the following:: | |
498 | ||
499 | ceph osd pool get {pool-name} pg_num | |
500 | ||
501 | ||
502 | Get a Cluster's PG Statistics | |
503 | ============================= | |
504 | ||
505 | To get the statistics for the placement groups in your cluster, execute the following:: | |
506 | ||
507 | ceph pg dump [--format {format}] | |
508 | ||
509 | Valid formats are ``plain`` (default) and ``json``. | |
510 | ||
511 | ||
512 | Get Statistics for Stuck PGs | |
513 | ============================ | |
514 | ||
515 | To get the statistics for all placement groups stuck in a specified state, | |
516 | execute the following:: | |
517 | ||
518 | ceph pg dump_stuck inactive|unclean|stale|undersized|degraded [--format <format>] [-t|--threshold <seconds>] | |
519 | ||
520 | **Inactive** Placement groups cannot process reads or writes because they are waiting for an OSD | |
521 | with the most up-to-date data to come up and in. | |
522 | ||
523 | **Unclean** Placement groups contain objects that are not replicated the desired number | |
524 | of times. They should be recovering. | |
525 | ||
526 | **Stale** Placement groups are in an unknown state - the OSDs that host them have not | |
527 | reported to the monitor cluster in a while (configured by ``mon_osd_report_timeout``). | |
528 | ||
529 | Valid formats are ``plain`` (default) and ``json``. The threshold defines the minimum number | |
530 | of seconds the placement group is stuck before including it in the returned statistics | |
531 | (default 300 seconds). | |
532 | ||
533 | ||
534 | Get a PG Map | |
535 | ============ | |
536 | ||
537 | To get the placement group map for a particular placement group, execute the following:: | |
538 | ||
539 | ceph pg map {pg-id} | |
540 | ||
541 | For example:: | |
542 | ||
543 | ceph pg map 1.6c | |
544 | ||
545 | Ceph will return the placement group map, the placement group, and the OSD status:: | |
546 | ||
547 | osdmap e13 pg 1.6c (1.6c) -> up [1,0] acting [1,0] | |
548 | ||
549 | ||
550 | Get a PGs Statistics | |
551 | ==================== | |
552 | ||
553 | To retrieve statistics for a particular placement group, execute the following:: | |
554 | ||
555 | ceph pg {pg-id} query | |
556 | ||
557 | ||
558 | Scrub a Placement Group | |
559 | ======================= | |
560 | ||
561 | To scrub a placement group, execute the following:: | |
562 | ||
563 | ceph pg scrub {pg-id} | |
564 | ||
565 | Ceph checks the primary and any replica nodes, generates a catalog of all objects | |
566 | in the placement group and compares them to ensure that no objects are missing | |
567 | or mismatched, and their contents are consistent. Assuming the replicas all | |
568 | match, a final semantic sweep ensures that all of the snapshot-related object | |
569 | metadata is consistent. Errors are reported via logs. | |
570 | ||
11fdf7f2 TL |
571 | To scrub all placement groups from a specific pool, execute the following:: |
572 | ||
573 | ceph osd pool scrub {pool-name} | |
574 | ||
c07f9fc5 FG |
575 | Prioritize backfill/recovery of a Placement Group(s) |
576 | ==================================================== | |
577 | ||
578 | You may run into a situation where a bunch of placement groups will require | |
579 | recovery and/or backfill, and some particular groups hold data more important | |
580 | than others (for example, those PGs may hold data for images used by running | |
581 | machines and other PGs may be used by inactive machines/less relevant data). | |
582 | In that case, you may want to prioritize recovery of those groups so | |
583 | performance and/or availability of data stored on those groups is restored | |
11fdf7f2 | 584 | earlier. To do this (mark particular placement group(s) as prioritized during |
c07f9fc5 FG |
585 | backfill or recovery), execute the following:: |
586 | ||
587 | ceph pg force-recovery {pg-id} [{pg-id #2}] [{pg-id #3} ...] | |
588 | ceph pg force-backfill {pg-id} [{pg-id #2}] [{pg-id #3} ...] | |
589 | ||
590 | This will cause Ceph to perform recovery or backfill on specified placement | |
591 | groups first, before other placement groups. This does not interrupt currently | |
592 | ongoing backfills or recovery, but causes specified PGs to be processed | |
593 | as soon as possible. If you change your mind or prioritize wrong groups, | |
594 | use:: | |
595 | ||
596 | ceph pg cancel-force-recovery {pg-id} [{pg-id #2}] [{pg-id #3} ...] | |
597 | ceph pg cancel-force-backfill {pg-id} [{pg-id #2}] [{pg-id #3} ...] | |
598 | ||
599 | This will remove "force" flag from those PGs and they will be processed | |
600 | in default order. Again, this doesn't affect currently processed placement | |
601 | group, only those that are still queued. | |
602 | ||
603 | The "force" flag is cleared automatically after recovery or backfill of group | |
604 | is done. | |
7c673cae | 605 | |
11fdf7f2 TL |
606 | Similarly, you may use the following commands to force Ceph to perform recovery |
607 | or backfill on all placement groups from a specified pool first:: | |
608 | ||
609 | ceph osd pool force-recovery {pool-name} | |
610 | ceph osd pool force-backfill {pool-name} | |
611 | ||
612 | or:: | |
613 | ||
614 | ceph osd pool cancel-force-recovery {pool-name} | |
615 | ceph osd pool cancel-force-backfill {pool-name} | |
616 | ||
617 | to restore to the default recovery or backfill priority if you change your mind. | |
618 | ||
619 | Note that these commands could possibly break the ordering of Ceph's internal | |
620 | priority computations, so use them with caution! | |
621 | Especially, if you have multiple pools that are currently sharing the same | |
622 | underlying OSDs, and some particular pools hold data more important than others, | |
623 | we recommend you use the following command to re-arrange all pools's | |
624 | recovery/backfill priority in a better order:: | |
625 | ||
626 | ceph osd pool set {pool-name} recovery_priority {value} | |
627 | ||
628 | For example, if you have 10 pools you could make the most important one priority 10, | |
629 | next 9, etc. Or you could leave most pools alone and have say 3 important pools | |
630 | all priority 1 or priorities 3, 2, 1 respectively. | |
631 | ||
7c673cae FG |
632 | Revert Lost |
633 | =========== | |
634 | ||
635 | If the cluster has lost one or more objects, and you have decided to | |
636 | abandon the search for the lost data, you must mark the unfound objects | |
637 | as ``lost``. | |
638 | ||
639 | If all possible locations have been queried and objects are still | |
640 | lost, you may have to give up on the lost objects. This is | |
641 | possible given unusual combinations of failures that allow the cluster | |
642 | to learn about writes that were performed before the writes themselves | |
643 | are recovered. | |
644 | ||
645 | Currently the only supported option is "revert", which will either roll back to | |
646 | a previous version of the object or (if it was a new object) forget about it | |
647 | entirely. To mark the "unfound" objects as "lost", execute the following:: | |
648 | ||
649 | ceph pg {pg-id} mark_unfound_lost revert|delete | |
650 | ||
651 | .. important:: Use this feature with caution, because it may confuse | |
652 | applications that expect the object(s) to exist. | |
653 | ||
654 | ||
655 | .. toctree:: | |
656 | :hidden: | |
657 | ||
658 | pg-states | |
659 | pg-concepts | |
660 | ||
661 | ||
662 | .. _Create a Pool: ../pools#createpool | |
663 | .. _Mapping PGs to OSDs: ../../../architecture#mapping-pgs-to-osds | |
664 | .. _pgcalc: http://ceph.com/pgcalc/ |