Storage Strategies Guide
Creating storage strategies for Red Hat Ceph Storage clusters
Abstract
Chapter 1. Overview
From the perspective of a Ceph client, interacting with the Ceph storage cluster is remarkably simple:
- Connect to the Cluster
- Create a Pool I/O Context
This remarkably simple interface is how a Ceph client selects one of the storage strategies you define. Storage strategies are invisible to the Ceph client in all but storage capacity and performance.
1.1. What are Storage Strategies?
A storage strategy is a method of storing data that serves a particular use case. For example, if you need to store volumes and images for a cloud platform like OpenStack, you might choose to store data on reasonably performant SAS drives with SSD-based journals. By contrast, if you need to store object data for an S3- or Swift-compliant gateway, you might choose to use something more economical, like SATA drives. Ceph can accommodate both scenarios in the same Ceph cluster, but you need a means of providing the SAS/SSD storage strategy to the cloud platform (e.g., Glance and Cinder in OpenStack), and a means of providing SATA storage for your object store.
Storage strategies include the storage media (hard drives, SSDs, etc.), the CRUSH maps that set up performance and failure domains for the storage media, the number of placement groups, and the pool interface. Ceph supports multiple storage strategies. Use cases, cost/benefit performance tradeoffs and data durability are the primary considerations that drive storage strategies.
- Use Cases: Ceph provides massive storage capacity, and it supports numerous use cases. For example, the Ceph Block Device client is a leading storage backend for cloud platforms like OpenStack—providing limitless storage for volumes and images with high performance features like copy-on-write cloning. By contrast, the Ceph Object Gateway client is a leading storage backend for cloud platforms that provides RESTful S3-compliant and Swift-compliant object storage for objects like audio, bitmap, video and other data.
- Cost/Benefit of Performance: Faster is better. Bigger is better. High durability is better. However, there is a price for each superlative quality, and a corresponding cost/benefit trade off. Consider the following use cases from a performance perspective: SSDs can provide very fast storage for relatively small amounts of data and journaling. Storing a database or object index may benefit from a pool of very fast SSDs, but prove too expensive for other data. SAS drives with SSD journaling provide fast performance at an economical price for volumes and images. SATA drives without SSD journaling provide cheap storage with lower overall performance. When you create a CRUSH hierarchy of OSDs, you need to consider the use case and an acceptable cost/performance trade off.
-
Durability: In large scale clusters, hardware failure is an expectation, not an exception. However, data loss and service interruption remain unacceptable. For this reason, data durability is very important. Ceph addresses data durability with multiple deep copies of an object or with erasure coding and multiple coding chunks. Multiple copies or multiple coding chunks present an additional cost/benefit tradeoff: it’s cheaper to store fewer copies or coding chunks, but it may lead to the inability to service write requests in a degraded state. Generally, one object with two additional copies (i.e.,
size = 3
) or two coding chunks may allow a cluster to service writes in a degraded state while the cluster recovers. The CRUSH algorithm aids this process by ensuring that Ceph stores additional copies or coding chunks in different locations within the cluster. This ensures that the failure of a single storage device or node doesn’t lead to a loss of all of the copies or coding chunks necessary to preclude data loss.
You can capture use cases, cost/benefit performance tradeoffs and data durability in a storage strategy and present it to a Ceph client as a storage pool.
Ceph’s object copies or coding chunks make RAID obsolete. Do not use RAID, because Ceph already handles data durability, a degraded RAID has a negative impact on performance, and recovering data using RAID is substantially slower than using deep copies or erasure coding chunks.
1.2. Configuring Storage Strategies
Configuring storage strategies is about assigning Ceph OSDs to a CRUSH hierarchy, defining the number of placement groups for a pool, and creating a pool. The general steps are:
- Define a Storage Strategy: Storage strategies require you to analyze your use case, cost/benefit performance tradeoffs and data durability. Then, you create OSDs suitable for that use case. For example, you can create SSD-backed OSDs for a high performance pool; SAS drive/SSD journal-backed OSDs for high-performance block device volumes and images; or, SATA-backed OSDs for low cost storage. Ideally, each OSD for a use case should have the same hardware configuration so that you have a consistent performance profile.
-
Define a CRUSH Hierarchy: Ceph rules select a node (usually the
root
) in a CRUSH hierarchy, and identify the appropriate OSDs for storing placement groups and the objects they contain. You must create a CRUSH hierarchy and a CRUSH rule for your storage strategy. CRUSH hierarchies get assigned directly to a pool by the CRUSH rule setting. - Calculate Placement Groups: Ceph shards a pool into placement groups. You need to set an appropriate number of placement groups for your pool, and remain within a healthy maximum number of placement groups in the event that you assign multiple pools to the same CRUSH rule.
-
Create a Pool: Finally, you must create a pool and determine whether it uses replicated or erasure-coded storage. You must set the number of placement groups for the pool, the rule for the pool and the durability (size or
K+M
coding chunks).
Remember, the pool is the Ceph client’s interface to the storage cluster, but the storage strategy is completely transparent to the Ceph client (except for capacity and performance).
Chapter 2. CRUSH Administration
The CRUSH (Controlled Replication Under Scalable Hashing) algorithm determines how to store and retrieve data by computing data storage locations.
Any sufficiently advanced technology is indistinguishable from magic. | ||
-- Arthur C. Clarke |
2.1. Introducing CRUSH
The CRUSH map for your storage cluster describes your device locations within CRUSH hierarchies and a rule for each hierarchy that determines how Ceph stores data.
The CRUSH map contains at least one hierarchy of nodes and leaves. The nodes of a hierarchy, called "buckets" in Ceph, are any aggregation of storage locations as defined by their type. For example, rows, racks, chassis, hosts, and devices. Each leaf of the hierarchy consists essentially of one of the storage devices in the list of storage devices. A leaf is always contained in one node or "bucket." A CRUSH map also has a list of rules that determine how CRUSH stores and retrieves data.
Storage devices are added to the CRUSH map when adding an OSD to the cluster.
The CRUSH algorithm distributes data objects among storage devices according to a per-device weight value, approximating a uniform probability distribution. CRUSH distributes objects and their replicas or erasure-coding chunks according to the hierarchical cluster map an administrator defines. The CRUSH map represents the available storage devices and the logical buckets that contain them for the rule, and by extension each pool that uses the rule.
To map placement groups to OSDs across failure domains or performance domains, a CRUSH map defines a hierarchical list of bucket types; that is, under types
in the generated CRUSH map. The purpose of creating a bucket hierarchy is to segregate the leaf nodes by their failure domains or performance domains or both. Failure domains include hosts, chassis, racks, power distribution units, pods, rows, rooms, and data centers. Performance domains include failure domains and OSDs of a particular configuration. For example, SSDs, SAS drives with SSD journals, SATA drives, and so on. In RHCS 3, devices have the notion of a class
, such as hdd
, ssd
and nvme
to more rapidly build CRUSH hierarchies with a class of devices.
With the exception of the leaf nodes representing OSDs, the rest of the hierarchy is arbitrary, and you can define it according to your own needs if the default types do not suit your requirements. We recommend adapting your CRUSH map bucket types to your organization’s hardware naming conventions and using instance names that reflect the physical hardware names. Your naming practice can make it easier to administer the cluster and troubleshoot problems when an OSD or other hardware malfunctions and the administrator needs remote or physical access to the host or other hardware.
In the following example, the bucket hierarchy has four leaf buckets (osd 1-4
), two node buckets (host 1-2
) and one rack node (rack 1
).
Since leaf nodes reflect storage devices declared under the devices
list at the beginning of the CRUSH map, there is no need to declare them as bucket instances. The second lowest bucket type in the hierarchy usually aggregates the devices; that is, it is usually the computer containing the storage media, and uses whatever term administrators prefer to describe it, such as "node", "computer", "server," "host", "machine", and so on. In high density environments, it is increasingly common to see multiple hosts/nodes per card and per chassis. Make sure to account for card and chassis failure too, for example, the need to pull a card or chassis if a node fails can result in bringing down numerous hosts/nodes and their OSDs.
When declaring a bucket instance, specify its type, give it a unique name as a string, assign it an optional unique ID expressed as a negative integer, specify a weight relative to the total capacity or capability of its items, specify the bucket algorithm such as straw
, and the hash that is usually 0
reflecting hash algorithm rjenkins1
. A bucket can have one or more items. The items can consist of node buckets or leaves. Items can have a weight that reflects the relative weight of the item.
2.1.1. Placing Data Dynamically
Ceph Clients and Ceph OSDs both use the CRUSH map and the CRUSH algorithm.
- Ceph Clients: By distributing CRUSH maps to Ceph clients, CRUSH empowers Ceph clients to communicate with OSDs directly. This means that Ceph clients avoid a centralized object look-up table that could act as a single point of failure, a performance bottleneck, a connection limitation at a centralized look-up server and a physical limit to the storage cluster’s scalability.
- Ceph OSDs: By distributing CRUSH maps to Ceph OSDs, Ceph empowers OSDs to handle replication, backfilling and recovery. This means that the Ceph OSDs handle storage of object replicas (or coding chunks) on behalf of the Ceph client. It also means that Ceph OSDs know enough about the cluster to re-balance the cluster (backfilling) and recover from failures dynamically.
2.1.2. Establishing Failure Domains
Having multiple object replicas or M
erasure coding chunks helps prevent data loss, but it is not sufficient to address high availability. By reflecting the underlying physical organization of the Ceph Storage Cluster, CRUSH can model—and thereby address—potential sources of correlated device failures. By encoding the cluster’s topology into the cluster map, CRUSH placement policies can separate object replicas or erasure coding chunks across different failure domains while still maintaining the desired pseudo-random distribution. For example, to address the possibility of concurrent failures, it may be desirable to ensure that data replicas or erasure coding chunks are on devices using different shelves, racks, power supplies, controllers or physical locations. This helps to prevent data loss and allows the cluster to operate in a degraded state.
2.1.3. Establishing Performance Domains
Ceph can support multiple hierarchies to separate one type of hardware performance profile from another type of hardware performance profile. For example, CRUSH can create one hierarchy for hard disk drives and another hierarchy for SSDs. Performance domains—hierarchies that take the performance profile of the underlying hardware into consideration—are increasingly popular due to the need to support different performance characteristics. Operationally, these are just CRUSH maps with more than one root
type bucket. Use case examples include:
- VMs: Ceph hosts that serve as a back end to cloud platforms like OpenStack, CloudStack, ProxMox or OpenNebula tend to use the most stable and performant filesystem such as XFS on SAS drives with a partitioned high performance SSD for journaling, because XFS does not journal and write simultaneously. To maintain a consistent performance profile, such use cases should aggregate similar hardware in a CRUSH hierarchy.
- Object Storage: Ceph hosts that serve as an object storage back end for S3 and Swift interfaces may take advantage of less expensive storage media such as SATA drives that may not be suitable for VMs—reducing the cost per gigabyte for object storage, while separating more economical storage hosts from more performant ones intended for storing volumes and images on cloud platforms. HTTP tends to be the bottleneck in object storage systems.
- Cold Storage: Systems designed for cold storage—infrequently accessed data, or data retrieval with relaxed performance requirements—may take advantage of less expensive storage media and erasure coding. However, erasure coding may require a bit of additional RAM and CPU, and thus differ in RAM and CPU requirements from a host used for object storage or VMs.
-
SSD-backed Pools: SSDs are expensive, but they provide significant advantages over hard disk drives. SSDs have no seek time and they provide high total throughput. In addition to using SSDs for journaling, a cluster can support SSD-backed pools. Common use cases include high performance SSD pools. For example, it is possible to map the
.rgw.buckets.index
pool for the Ceph Object Gateway to SSDs instead of SATA drives.
In RHCS 3 and later releases, a CRUSH map supports the notion of a device class
. Ceph can discover aspects of a storage device and automatically assign a class such as hdd
, ssd
or nvme
. However, CRUSH is not limited to these defaults. For example, CRUSH hierarchies may also be used to separate different types of workloads. For example, an SSD might be used for a journal or write-ahead log, a bucket index or for raw object storage. CRUSH can support different device classes, such as ssd-bucket-index
or ssd-object-storage
so Ceph does not use the same storage media for different workloads—making performance more predictable and consistent.
2.1.3.1. Using Different Device Classes in RHCS 2 and Earlier
The notion of device classes does not exist in RHCS 2 and earlier. In order to use different classes of storage drives in RHCS 2 and earlier, such as SSDs and SATA drives, create pools that will point to the SSD and SATA drives. The data going to these pools must be segregated by creating separate root hierarchies for the SSD and SATA drives respectively. The CRUSH map must define one hierarchy for SSDs and one hierarchy for SATA drives. SSDs and SATA drives may reside on the same hosts, or in separate hosts.
Once the two CRUSH hierarchies exist, create separate rules for the SSD and SATA hierarchies and set the respective pools to use the appropriate ruleset.
- NOTE
-
Red Hat does not support the notion of
ruleset
in RHCS 3.0 and later releases. See RHCS 2 documentation for details on CRUSH rulesets.
To achieve this, modify the CRUSH map (decompile the CRUSH map to modify—see the Decompile a CRUSH Map section for further details). Create two different root or entry points from which the CRUSH algorithm will go through to store objects. There will be one root for SSD disks and another one for SATA disks. The following example has two SATA disks and two SSD disks on each host, and three hosts in total. In a way, CRUSH thinks there are two different platforms.
CRUSH map example:
## # OSD SATA DECLARATION ## host ceph-osd2-sata { id -2 # do not change unnecessarily # weight 2.000 alg straw hash 0 # rjenkins1 item osd.0 weight 1.000 item osd.3 weight 1.000 } host ceph-osd1-sata { id -3 # do not change unnecessarily # weight 2.000 alg straw hash 0 # rjenkins1 item osd.2 weight 1.000 item osd.5 weight 1.000 } host ceph-osd0-sata { id -4 # do not change unnecessarily # weight 2.000 alg straw hash 0 # rjenkins1 item osd.1 weight 1.000 item osd.4 weight 1.000 } ## # OSD SSD DECLARATION ## host ceph-osd2-ssd { id -22 # do not change unnecessarily # weight 2.000 alg straw hash 0 # rjenkins1 item osd.6 weight 1.000 item osd.9 weight 1.000 } host ceph-osd1-ssd { id -23 # do not change unnecessarily # weight 2.000 alg straw hash 0 # rjenkins1 item osd.8 weight 1.000 item osd.11 weight 1.000 } host ceph-osd0-ssd { id -24 # do not change unnecessarily # weight 2.000 alg straw hash 0 # rjenkins1 item osd.7 weight 1.000 item osd.10 weight 1.000 }
Create two roots containing the OSDs:
## # SATA ROOT DECLARATION ## root sata { id -1 # do not change unnecessarily # weight 6.000 alg straw hash 0 # rjenkins1 item ceph-osd2-sata weight 2.000 item ceph-osd1-sata weight 2.000 item ceph-osd0-sata weight 2.000 } ## # SATA ROOT DECLARATION ## root ssd { id -21 # do not change unnecessarily # weight 6.000 alg straw hash 0 # rjenkins1 item ceph-osd2-ssd weight 2.000 item ceph-osd1-ssd weight 2.000 item ceph-osd0-ssd weight 2.000 }
Create two new rules for the SSDs:
## # SSD RULE DECLARATION ## # rules rule ssd { ruleset 0 type replicated min_size 1 max_size 10 step take ssd step chooseleaf firstn 0 type host step emit } ## # SATA RULE DECLARATION ## rule sata { ruleset 1 type replicated min_size 1 max_size 10 step take sata step chooseleaf firstn 0 type host step emit }
Compile and inject the new map:
$ crushtool -c lamap.txt -o lamap.coloc $ sudo ceph osd setcrushmap -i lamap.coloc
See the result:
$ sudo ceph osd tree # id weight type name up/down reweight -21 12 root ssd -22 2 host ceph-osd2-ssd 6 1 osd.6 up 1 9 1 osd.9 up 1 -23 2 host ceph-osd1-ssd 8 1 osd.8 up 1 11 1 osd.11 up 1 -24 2 host ceph-osd0-ssd 7 1 osd.7 up 1 10 1 osd.10 up 1 -1 12 root sata -2 2 host ceph-osd2-sata 0 1 osd.0 up 1 3 1 osd.3 up 1 -3 2 host ceph-osd1-sata 2 1 osd.2 up 1 5 1 osd.5 up 1 -4 2 host ceph-osd0-sata 1 1 osd.1 up 1 4 1 osd.4 up 1
Now work on the pools.
Create the pools:
pool
ssd
:root@ceph-mon0:~# ceph osd pool create ssd 128 128
pool
sata
:root@ceph-mon0:~# ceph osd pool create sata 128 128
Assign rules to the pools:
Set pool 8 crush_ruleset to 0:
root@ceph-mon0:~# ceph osd pool set ssd crush_ruleset 0
Set pool 9 crush_ruleset to 1:
root@ceph-mon0:~# ceph osd pool set sata crush_ruleset 1
Result from
ceph osd dump
:pool 8 'ssd' replicated size 2 min_size 1 crush_ruleset 0 object_hash rjenkins pg_num 128 pgp_num 128 last_change 116 flags hashpspool stripe_width 0 pool 9 'sata' replicated size 2 min_size 1 crush_ruleset 1 object_hash rjenkins pg_num 128 pgp_num 128 last_change 117 flags hashpspool stripe_width 0
To prevent the moved OSDs from moving back to their default CRUSH locations upon restart of the service, disable updating the CRUSH map on start of the daemon using osd crush update on start = false
.
2.1.3.2. Using Different Device Classes in RHCS 3 and Later
To create performance domains in RHCS 3, use device classes and a single CRUSH hierarchy. Simply add OSDs to the CRUSH hierarchy in the same manner as RHCS 2 using the CLI. Then do the following:
Add a class to each device. For example:
# ceph osd crush set-device-class <class> <osdId> [<osdId>] # ceph osd crush set-device-class hdd osd.0 osd.1 osd.4 osd.5 # ceph osd crush set-device-class ssd osd.2 osd.3 osd.6 osd.7
Then, create rules to use the devices.
# ceph osd crush rule create-replicated <rule-name> <root> <failure-domain-type> <class> # ceph osd crush rule create-replicated cold default host hdd # ceph osd crush rule create-replicated hot default host ssd
Finally, set pools to use the rules.
ceph osd pool set <poolname> crush_rule <rule-name> ceph osd pool set cold_tier crush_rule cold ceph osd pool set hot_tier crush_rule hot
With RHCS 3 and later, there is no need to manually edit the CRUSH map!
2.2. CRUSH Hierarchies
The CRUSH map is a directed acyclic graph, so it can accommodate multiple hierarchies (for example, performance domains). The easiest way to create and modify a CRUSH hierarchy is with the Ceph CLI; however, you can also decompile a CRUSH map, edit it, recompile it, and activate it.
When declaring a bucket instance with the Ceph CLI, you must specify its type and give it a unique name (string). Ceph automatically assigns a bucket ID, set the algorithm to straw
, set the hash to 0
reflecting rjenkins1
and set a weight. When modifying a decompiled CRUSH map, assign the bucket a unique ID expressed as a negative integer (optional), specify a weight relative to the total capacity/capability of its item(s), specify the bucket algorithm (usually straw
), and the hash (usually 0
, reflecting hash algorithm rjenkins1
).
A bucket can have one or more items. The items can consist of node buckets (for example, racks, rows, hosts) or leaves (for example, an OSD disk). Items can have a weight that reflects the relative weight of the item.
When modifying a decompiled CRUSH map, you can declare a node bucket with the following syntax:
[bucket-type] [bucket-name] { id [a unique negative numeric ID] weight [the relative capacity/capability of the item(s)] alg [the bucket type: uniform | list | tree | straw ] hash [the hash type: 0 by default] item [item-name] weight [weight] }
For example, using the diagram above, we would define two host buckets and one rack bucket. The OSDs are declared as items within the host buckets:
host node1 { id -1 alg straw hash 0 item osd.0 weight 1.00 item osd.1 weight 1.00 } host node2 { id -2 alg straw hash 0 item osd.2 weight 1.00 item osd.3 weight 1.00 } rack rack1 { id -3 alg straw hash 0 item node1 weight 2.00 item node2 weight 2.00 }
In the foregoing example, note that the rack bucket does not contain any OSDs. Rather it contains lower level host buckets, and includes the sum total of their weight in the item entry.
2.2.1. CRUSH Location
A CRUSH location is the position of an OSD in terms of the CRUSH map’s hierarchy. When you express a CRUSH location on the command line interface, a CRUSH location specifier takes the form of a list of name/value pairs describing the OSD’s position. For example, if an OSD is in a particular row, rack, chassis and host, and is part of the default
CRUSH tree, its crush location could be described as:
root=default row=a rack=a2 chassis=a2a host=a2a1
Note:
- The order of the keys does not matter.
-
The key name (left of
=
) must be a valid CRUSHtype
. By default these includeroot
,datacenter
,room
,row
,pod
,pdu
,rack
,chassis
andhost
. You may edit the CRUSH map to change the types to suit your needs. -
You do not need to specify all the buckets/keys. For example, by default, Ceph automatically sets a
ceph-osd
daemon’s location to beroot=default host={HOSTNAME}
(based on the output fromhostname -s
).
2.2.1.1. Default ceph-crush-location
Hook
Upon startup, Ceph gets the CRUSH location of each daemon using the ceph-crush-location
tool by default. The ceph-crush-location
utility returns the CRUSH location of a given daemon. Its CLI usage consists of:
ceph-crush-location --cluster {cluster-name} --id {ID} --type {daemon-type}
For example, the following command returns the location of OSD.0
:
ceph-crush-location --cluster ceph --id 0 --type osd
By default, the ceph-crush-location
utility returns a CRUSH location string for a given daemon. The location returned in order of precedence is based on:
-
A
{TYPE}_crush_location
option in the Ceph configuration file. For example, for OSD daemons,{TYPE}
would beosd
and the setting would look likeosd_crush_location
. -
A
crush_location
option for a particular daemon in the Ceph configuration file. -
A default of
root=default host=HOSTNAME
where the host name is returned by thehostname -s
command.
In a typical deployment scenario, provisioning software (or the system administrator) can simply set the crush_location
field in a host’s Ceph configuration file to describe that machine’s location within the datacenter or cluster. This provides location awareness to Ceph daemons and clients alike.
2.2.1.2. Custom Location Hooks
A custom location hook can be used in place of the generic hook for OSD daemon placement in the hierarchy. (On startup, each OSD ensures its position is correct.):
osd_crush_location_hook = /path/to/script
This hook is passed several arguments (below) and should output a single line to stdout
with the CRUSH location description.:
ceph-crush-location --cluster {cluster-name} --id {ID} --type {daemon-type}
where the --cluster
name is typically 'ceph', the --id
is the daemon identifier (the OSD number), and the daemon --type
is typically osd
.
2.2.2. Adding a Bucket
To add a bucket instance to your CRUSH hierarchy, specify the bucket name and its type. Bucket names must be unique in the CRUSH map.
ceph osd crush add-bucket {name} {type}
If you plan to use multiple hierarchies, for example, for different hardware performance profiles, consider naming buckets based on their type of hardware or use case.
For example, you could create a hierarchy for solid state drives (ssd
), a hierarchy for SAS disks with SSD journals (hdd-journal
), and another hierarchy for SATA drives (hdd
):
ceph osd crush add-bucket ssd-root root ceph osd crush add-bucket hdd-journal-root root ceph osd crush add-bucket hdd-root root
The Ceph CLI outputs:
added bucket ssd-root type root to crush map added bucket hdd-journal-root type root to crush map added bucket hdd-root type root to crush map
Using colons (:) in bucket names is not supported.
Add an instance of each bucket type you need for your hierarchy. The following example demonstrates adding buckets for a row with a rack of SSD hosts and a rack of hosts for object storage.
ceph osd crush add-bucket ssd-row1 row ceph osd crush add-bucket ssd-row1-rack1 rack ceph osd crush add-bucket ssd-row1-rack1-host1 host ceph osd crush add-bucket ssd-row1-rack1-host2 host ceph osd crush add-bucket hdd-row1 row ceph osd crush add-bucket hdd-row1-rack2 rack ceph osd crush add-bucket hdd-row1-rack1-host1 host ceph osd crush add-bucket hdd-row1-rack1-host2 host ceph osd crush add-bucket hdd-row1-rack1-host3 host ceph osd crush add-bucket hdd-row1-rack1-host4 host
If you have already used the Ansible automation application or another tool to add OSDs to your cluster, the host nodes can already be in the CRUSH map.
Once you have completed these steps, view your tree.
ceph osd tree
Notice that the hierarchy remains flat. You must move your buckets into hierarchical position after you add them to the CRUSH map.
2.2.3. Moving a Bucket
When you create your initial cluster, Ceph has a default CRUSH map with a root bucket named default
and your initial OSD hosts appear under the default
bucket. When you add a bucket instance to your CRUSH map, it appears in the CRUSH hierarchy, but it does not necessarily appear under a particular bucket.
To move a bucket instance to a particular location in your CRUSH hierarchy, specify the bucket name and its type. For example:
ceph osd crush move ssd-row1 root=ssd-root ceph osd crush move ssd-row1-rack1 row=ssd-row1 ceph osd crush move ssd-row1-rack1-host1 rack=ssd-row1-rack1 ceph osd crush move ssd-row1-rack1-host2 rack=ssd-row1-rack1
Once you have completed these steps, you can view your tree.
ceph osd tree
You can also use ceph osd crush create-or-move
to create a location while moving an OSD.
2.2.4. Removing a Bucket
To remove a bucket instance from your CRUSH hierarchy, specify the bucket name. For example:
ceph osd crush remove {bucket-name}
Or:
ceph osd crush rm {bucket-name}
The bucket must be empty in order to remove it.
If you are removing higher level buckets (for example, a root like default
), check to see if a pool uses a CRUSH rule that selects that bucket. If so, you need to modify your CRUSH rules; otherwise, peering fails.
2.2.5. Bucket Algorithms
When you create buckets using the Ceph CLI, Ceph sets the algorithm to straw
by default. Ceph supports four bucket algorithms, each representing a tradeoff between performance and reorganization efficiency. If you are unsure of which bucket type to use, we recommend using a straw
bucket. The bucket algorithms are:
-
Uniform: Uniform buckets aggregate devices with exactly the same weight. For example, when firms commission or decommission hardware, they typically do so with many machines that have exactly the same physical configuration (for example, bulk purchases). When storage devices have exactly the same weight, you can use the
uniform
bucket type, which allows CRUSH to map replicas into uniform buckets in constant time. With non-uniform weights, you should use another bucket algorithm. - List: List buckets aggregate their content as linked lists. Based on the RUSH (Replication Under Scalable Hashing) P algorithm, a list is a natural and intuitive choice for an expanding cluster: either an object is relocated to the newest device with some appropriate probability, or it remains on the older devices as before. The result is optimal data migration when items are added to the bucket. Items removed from the middle or tail of the list, however, can result in a significant amount of unnecessary movement, making list buckets most suitable for circumstances in which they never (or very rarely) shrink.
- Tree: Tree buckets use a binary search tree. They are more efficient than list buckets when a bucket contains a larger set of items. Based on the RUSH (Replication Under Scalable Hashing) R algorithm, tree buckets reduce the placement time to O(log n), making them suitable for managing much larger sets of devices or nested buckets.
- Straw (default): List and Tree buckets use a divide and conquer strategy in a way that either gives certain items precedence (for example, those at the beginning of a list) or obviates the need to consider entire subtrees of items at all. That improves the performance of the replica placement process, but can also introduce suboptimal reorganization behavior when the contents of a bucket change due an addition, removal, or re-weighting of an item. The straw bucket type allows all items to fairly “compete” against each other for replica placement through a process analogous to a draw of straws.
2.3. Ceph OSDs in CRUSH
Once you have a CRUSH hierarchy for the OSDs, add OSDs to the CRUSH hierarchy. You can also move or remove OSDs from an existing hierarchy. The Ceph CLI usage has the following values:
- id
- Description
- The numeric ID of the OSD.
- Type
- Integer
- Required
- Yes
- Example
-
0
- name
- Description
- The full name of the OSD.
- Type
- String
- Required
- Yes
- Example
-
osd.0
- weight
- Description
- The CRUSH weight for the OSD.
- Type
- Double
- Required
- Yes
- Example
-
2.0
- root
- Description
- The name of the root bucket of the hierarchy or tree in which the OSD resides.
- Type
- Key-value pair.
- Required
- Yes
- Example
-
root=default
,root=replicated_rule
, and so on
- bucket-type
- Description
- One or more name-value pairs, where the name is the bucket type and the value is the bucket’s name. You can specify a CRUSH location for an OSD in the CRUSH hierarchy.
- Type
- Key-value pairs.
- Required
- No
- Example
-
datacenter=dc1 room=room1 row=foo rack=bar host=foo-bar-1
2.3.1. Viewing OSDs in CRUSH
The ceph osd crush tree
command prints CRUSH buckets and items in a tree view. Use this command to determine a list of OSDs in a particular bucket. It will print output similar to ceph osd tree
.
To return additional details, execute the following:
# ceph osd crush tree -f json-pretty
The command returns an output similar to the following:
[ { "id": -2, "name": "ssd", "type": "root", "type_id": 10, "items": [ { "id": -6, "name": "dell-per630-11-ssd", "type": "host", "type_id": 1, "items": [ { "id": 6, "name": "osd.6", "type": "osd", "type_id": 0, "crush_weight": 0.099991, "depth": 2 } ] }, { "id": -7, "name": "dell-per630-12-ssd", "type": "host", "type_id": 1, "items": [ { "id": 7, "name": "osd.7", "type": "osd", "type_id": 0, "crush_weight": 0.099991, "depth": 2 } ] }, { "id": -8, "name": "dell-per630-13-ssd", "type": "host", "type_id": 1, "items": [ { "id": 8, "name": "osd.8", "type": "osd", "type_id": 0, "crush_weight": 0.099991, "depth": 2 } ] } ] }, { "id": -1, "name": "default", "type": "root", "type_id": 10, "items": [ { "id": -3, "name": "dell-per630-11", "type": "host", "type_id": 1, "items": [ { "id": 0, "name": "osd.0", "type": "osd", "type_id": 0, "crush_weight": 0.449997, "depth": 2 }, { "id": 3, "name": "osd.3", "type": "osd", "type_id": 0, "crush_weight": 0.289993, "depth": 2 } ] }, { "id": -4, "name": "dell-per630-12", "type": "host", "type_id": 1, "items": [ { "id": 1, "name": "osd.1", "type": "osd", "type_id": 0, "crush_weight": 0.449997, "depth": 2 }, { "id": 4, "name": "osd.4", "type": "osd", "type_id": 0, "crush_weight": 0.289993, "depth": 2 } ] }, { "id": -5, "name": "dell-per630-13", "type": "host", "type_id": 1, "items": [ { "id": 2, "name": "osd.2", "type": "osd", "type_id": 0, "crush_weight": 0.449997, "depth": 2 }, { "id": 5, "name": "osd.5", "type": "osd", "type_id": 0, "crush_weight": 0.289993, "depth": 2 } ] } ] } ]
In RHCS 3 and later, an OSD object will also have a device_class
attribute.
2.3.2. Adding an OSD to CRUSH
Adding an OSD to a CRUSH hierarchy is the final step before you may start an OSD (rendering it up
and in
) and Ceph assigns placement groups to the OSD.
In RHCS 3, you may also add a device class.
You must prepare an OSD before you add it to the CRUSH hierarchy. Deployment utilities, such the Ansible automation application, performs this step for you. See Adding/Removing OSDs for additional details.
The CRUSH hierarchy is notional, so the ceph osd crush add
command allows you to add OSDs to the CRUSH hierarchy wherever you wish. The location you specify should reflect its actual location. If you specify at least one bucket, the command places the OSD into the most specific bucket you specify, and it moves that bucket underneath any other buckets you specify.
To add an OSD to a CRUSH hierarchy:
ceph osd crush add {id-or-name} {weight} [{bucket-type}={bucket-name} ...]
If you specify only the root bucket, the command attaches the OSD directly to the root. However, CRUSH rules expect OSDs to be inside of hosts or chassis, and host or chassis should be inside of other buckets reflecting your cluster topology.
The following example adds osd.0
to the hierarchy:
ceph osd crush add osd.0 1.0 root=default datacenter=dc1 room=room1 row=foo rack=bar host=foo-bar-1
You can also use ceph osd crush set
or ceph osd crush create-or-move
to add an OSD to the CRUSH hierarchy.
2.3.3. Moving an OSD within a CRUSH Hierarchy
If the deployment utility, such as the Ansible automation application added an OSD to the CRUSH map at a sub-optimal CRUSH location, or if your cluster topology changes, you can move an OSD in the CRUSH hierarchy to reflect its actual location.
Moving an OSD in the CRUSH hierarchy means that Ceph will recompute which placement groups get assigned to the OSD, potentially resulting in significant redistribution of data.
To move an OSD within the CRUSH hierarchy:
ceph osd crush set {id-or-name} {weight} root={pool-name} [{bucket-type}={bucket-name} ...]
You can also use ceph osd crush create-or-move
to move an OSD within the CRUSH hierarchy.
2.3.4. Removing an OSD from a CRUSH Hierarchy
Removing an OSD from a CRUSH hierarchy is the first step when you want to remove an OSD from your cluster. When you remove the OSD from the CRUSH map, CRUSH recomputes which OSDs get the placement groups and data re-balances accordingly. See Adding/Removing OSDs for additional details.
To remove an OSD from the CRUSH map of a running cluster, execute the following:
ceph osd crush remove {name}
2.4. Device Classes
Ceph’s CRUSH map provides extraordinary flexibility in controlling data placement. This is one of Ceph’s greatest strengths. Early Ceph deployments used hard disk drives almost exclusively. Today, Ceph clusters are frequently built with multiple types of storage devices: HDD, SSD, NVMe, or even various classes of the foregoing. For example, it is common in Ceph Object Gateway deployments to have storage policies where clients can store data on slower HDDs and other storage policies for storing data on fast SSDs. Ceph Object Gateway deployments may even have a pool backed by fast SSDs for bucket indices. Additionally, OSD nodes also frequently have SSDs used exclusively for journals or write-ahead logs that do NOT appear in the CRUSH map. These complex hardware scenarios historically required manually editing the CRUSH map, which can be time-consuming and tedious. In RHCS 3, Red Hat has added a new 'device class', which dramatically simplifies creating CRUSH hierarchies, In RHCS 3 and beyond, it is no longer required to have different CRUSH hierarchies for different classes of storage devices.
CRUSH rules work in terms of the CRUSH hierarchy. However, if different classes of storage devices reside in the same hosts, the process became more complicated—requiring users to create multiple CRUSH hierarchies for each class of device, and then disable the osd crush update on start
option that automates much of the CRUSH hierarchy management. Device classes eliminate this tediousness by telling the CRUSH rule what class of device to use, dramatically simplifying CRUSH management tasks.
In RHCS 3 and later, the ceph osd tree
command has a column reflecting a device class.
The following sections detail device class usage. See Using Different Device Classes in RHCS 3 and Later and CRUSH Storage Strategy Examples for additional examples.
2.4.1. Setting a Device Class
To set a device class for an OSD, execute the following:
# ceph osd crush set-device-class <class> <osdId> [<osdId>...]
For example:
# ceph osd crush set-device-class hdd osd.0 osd.1 # ceph osd crush set-device-class ssd osd.2 osd.3 # ceph osd crush set-device-class bucket-index osd.4
Ceph may assign a class to a device automatically. However, class names are simply arbitrary strings. There is no requirement to adhere to hdd
, ssd
or nvme
. In the foregoing example, a device class named bucket-index
may indicate an SSD device that a Ceph Object Gatway pool uses exclusively bucket index workloads. To change a device class that was already set, use ceph osd crush rm-device-class
first.
2.4.2. Removing a Device Class
To remove a device class for an OSD, execute the following:
# ceph osd crush rm-device-class <class> <osdId> [<osdId>...]
For example:
# ceph osd crush rm-device-class hdd osd.0 osd.1 # ceph osd crush rm-device-class ssd osd.2 osd.3 # ceph osd crush rm-device-class bucket-index osd.4
2.4.3. Renaming a Device Class
To rename a device class for all OSDs that use that class, execute the following:
# ceph osd crush class rename <oldName> <newName>
For example:
# ceph osd crush class rename hdd sas15k
2.4.4. Listing Device Classes
To list device classes in the CRUSH map, execute the following:
# ceph osd crush class ls
The output will look something like this:
[ "hdd", "ssd", "bucket-index" ]
2.4.5. Listing OSDs of a Device Class
To list all OSDs that belong to a particular class, execute the following:
# ceph osd crush class ls-osd <class>
For example:
# ceph osd crush class ls-osd hdd
The output is simply a list of OSD numbers. For example:
0 1 2 3 4 5 6
2.4.6. Listing CRUSH Rules by Class
To list all crush rules that reference the same class, execute the following:
# ceph osd crush rule ls-by-class <class>
For example:
# ceph osd crush rule ls-by-class hdd
2.5. CRUSH Weights
The CRUSH algorithm assigns a weight value in terabytes (by convention) per OSD device with the objective of approximating a uniform probability distribution for write requests that assign new data objects to PGs and PGs to OSDs. For this reason, as a best practice, we recommend creating CRUSH hierarchies with devices of the same type and size, and assigning the same weight. We also recommend using devices with the same I/O and throughput characteristics so that you will also have uniform performance characteristics in your CRUSH hierarchy, even though performance characteristics do not affect data distribution.
Since using uniform hardware is not always practical, you may incorporate OSD devices of different sizes and use a relative weight so that Ceph will distribute more data to larger devices and less data to smaller devices.
2.5.1. Set an OSD’s Weight in Terabytes
To set an OSD CRUSH weight in Terabytes within the CRUSH map, execute the following command
ceph osd crush reweight {name} {weight}
Where:
- name
- Description
- The full name of the OSD.
- Type
- String
- Required
- Yes
- Example
-
osd.0
- weight
- Description
-
The CRUSH weight for the OSD. This should be the size of the OSD in Terabytes, where
1.0
is 1 Terabyte. - Type
- Double
- Required
- Yes
- Example
-
2.0
This setting is used when creating an OSD or adjusting the CRUSH weight immediately after adding the OSD. It usually does not change over the life of the OSD.
2.5.2. Set a Bucket’s OSD Weights
Using ceph osd crush reweight
can be time-consuming. You can set (or reset) all Ceph OSD weights under a bucket (row, rack, node, and so on) by executing:
osd crush reweight-subtree <name>
Where,
name
is the name of the CRUSH bucket.
2.5.3. Set an OSD’s in
Weight
For the purposes of ceph osd in
and ceph osd out
, an OSD is either in
the cluster or out
of the cluster. That is how a monitor records an OSD’s status. However, even though an OSD is in
the cluster, it may be experiencing a malfunction such that you do not want to rely on it as much until you fix it (for example, replace a storage drive, change out a controller, and so on).
You can increase or decrease the in
weight of a particular OSD (that is, without changing its weight in Terabytes) by executing:
ceph osd reweight {id} {weight}
Where:
-
id
is the OSD number. -
weight
is a range from 0.0-1.0, where0
is notin
the cluster (that is, it does not have any PGs assigned to it) and 1.0 isin
the cluster (that is, the OSD receives the same number of PGs as other OSDs).
2.5.4. Set an OSD’s Weight by Utilization
CRUSH is designed to approximate a uniform probability distribution for write requests that assign new data objects PGs and PGs to OSDs. However, a cluster may become imbalanced anyway. This can happen for a number of reasons. For example:
- Multiple Pools: You can assign multiple pools to a CRUSH hierarchy, but the pools may have different numbers of placement groups, size (number of replicas to store), and object size characteristics.
-
Custom Clients: Ceph clients such as block device, object gateway and filesystem shard data from their clients and stripe the data as objects across the cluster as uniform-sized smaller RADOS objects. So except for the foregoing scenario, CRUSH usually achieves its goal. However, there is another case where a cluster can become imbalanced: namely, using
librados
to store data without normalizing the size of objects. This scenario can lead to imbalanced clusters (for example, storing 100 1MB objects and 10 4MB objects will make a few OSDs have more data than the others). - Probability: A uniform distribution will result in some OSDs with more PGs and some with less. For clusters with a large number of OSDs, the statistical outliers will be further out.
You can reweight OSDs by utilization by executing the following:
ceph osd reweight-by-utilization [threshold] [weight_change_amount] [number_of_OSDs] [--no-increasing]
For example:
ceph osd test-reweight-by-utilization 110 .5 4 --no-increasing
Where:
-
threshold
is a percentage of utilization such that OSDs facing higher data storage loads will receive a lower weight and thus fewer PGs assigned to them. The default value is120
, reflecting 120%. Any value from100+
is a valid threshold. Optional. -
weight_change_amount
is the amount to change the weight. Valid values are greater than0.0 - 1.0
. The default value is0.05
. Optional. -
number_of_OSDs
is the maximum number of OSDs to reweight. For large clusters, limiting the number of OSDs to reweight prevents significant rebalancing. Optional. -
no-increasing
is off by default. Increasing the osd weight is allowed when using thereweight-by-utilization
ortest-reweight-by-utilization
commands. If this option is used with these commands, it prevent the increasing the OSD weight, even if the OSD is underutilized. Optional.
Executing reweight-by-utilization
is recommended and somewhat inevitable for large clusters. Utilization rates may change over time, and as your cluster size or hardware changes, the weightings may need to be updated to reflect changing utilization. If you elect to reweight by utilization, you may need to re-run this command as utilization, hardware or cluster size change.
Executing this or other weight commands that assign a weight will override the weight assigned by this command (for example, osd reweight-by-utilization
, osd crush weight
, osd weight
, in
or out
).
2.5.5. Set an OSD’s Weight by PG Distribution
In CRUSH hierarchies with a smaller number of OSDs, it’s possible for some OSDs to get more PGs than other OSDs, resulting in a higher load. You can reweight OSDs by PG distribution to address this situation by executing the following:
osd reweight-by-pg <poolname>
Where:
-
poolname
is the name of the pool. Ceph will examine how the pool assigns PGs to OSDs and reweight the OSDs according to this pool’s PG distribution. Note that multiple pools could be assigned to the same CRUSH hierarchy. Reweighting OSDs according to one pool’s distribution could have unintended effects for other pools assigned to the same CRUSH hierarchy if they do not have the same size (number of replicas) and PGs.
2.5.6. Recalculate a CRUSH Tree’s Weights
CRUSH tree buckets should be the sum of their leaf weights. If you manually edit the CRUSH map weights, you should execute the following to ensure that the CRUSH bucket tree accurately reflects the sum of the leaf OSDs under the bucket.
osd crush reweight-all
2.6. Primary Affinity
When a Ceph Client reads or writes data, it always contacts the primary OSD in the acting set. For set [2, 3, 4]
, osd.2
is the primary. Sometimes an OSD is not well suited to act as a primary compared to other OSDs (for example, it has a slow disk or a slow controller). To prevent performance bottlenecks (especially on read operations) while maximizing utilization of your hardware, you can set a Ceph OSD’s primary affinity so that CRUSH is less likely to use the OSD as a primary in an acting set. :
ceph osd primary-affinity <osd-id> <weight>
Primary affinity is 1
by default (that is, an OSD may act as a primary). You may set the OSD primary range from 0-1
, where 0
means that the OSD may NOT be used as a primary and 1
means that an OSD may be used as a primary. When the weight is < 1
, it is less likely that CRUSH will select the Ceph OSD Daemon to act as a primary.
2.7. CRUSH Rules
CRUSH rules define how a Ceph client selects buckets and the primary OSD within them to store object, and how the primary OSD selects buckets and the secondary OSDs to store replicas or coding chunks. For example, you might create a rule that selects a pair of target OSDs backed by SSDs for two object replicas, and another rule that select three target OSDs backed by SAS drives in different data centers for three replicas.
A rule takes the following form:
rule <rulename> { ruleset <ruleset> type [ replicated | raid4 ] min_size <min-size> max_size <max-size> step take <bucket-type> [class <class-name>] step [choose|chooseleaf] [firstn|indep] <N> <bucket-type> step emit }
- ruleset
- Description
- (Deprecated) A means of classifying a rule as belonging to a set of rules. Activated by setting the ruleset in a pool. Supported in RHCS 2 and earlier releases. Not supported in RHCS 3 and later releases.
- Purpose
- A component of the rule mask.
- Type
- Integer
- Required
- Yes
- Default
-
0
- type
- Description
- Describes a rule for either a storage drive (replicated) or a RAID.
- Purpose
- A component of the rule mask.
- Type
- String
- Required
- Yes
- Default
-
replicated
- Valid Values
-
Currently only
replicated
- min_size
- Description
- If a pool makes fewer replicas than this number, CRUSH will not select this rule.
- Type
- Integer
- Purpose
- A component of the rule mask.
- Required
- Yes
- Default
-
1
- max_size
- Description
- If a pool makes more replicas than this number, CRUSH will not select this rule.
- Type
- Integer
- Purpose
- A component of the rule mask.
- Required
- Yes
- Default
-
10
- step take <bucket-name> [class <class-name>]
- Description
- Takes a bucket name, and begins iterating down the tree. May take a device class name in RHCS 3 and later.
- Purpose
- A component of the rule.
- Required
- Yes
- Example
-
step take data
step take data class ssd
- step choose firstn <num> type <bucket-type>
- Description
Selects the number of buckets of the given type. The number is usually the number of replicas in the pool (that is, pool size).
-
If
<num> == 0
, choosepool-num-replicas
buckets (all available). -
If
<num> > 0 && < pool-num-replicas
, choose that many buckets. -
If
<num> < 0
, it meanspool-num-replicas - {num}
.
-
If
- Purpose
- A component of the rule.
- Prerequisite
-
Follows
step take
orstep choose
. - Example
-
step choose firstn 1 type row
- step chooseleaf firstn <num> type <bucket-type>
- Description
Selects a set of buckets of
{bucket-type}
and chooses a leaf node from the subtree of each bucket in the set of buckets. The number of buckets in the set is usually the number of replicas in the pool (that is, pool size).-
If
<num> == 0
, choosepool-num-replicas
buckets (all available). -
If
<num> > 0 && < pool-num-replicas
, choose that many buckets. -
If
<num> < 0
, it meanspool-num-replicas - <num>
.
-
If
- Purpose
- A component of the rule. Usage removes the need to select a device using two steps.
- Prerequisite
-
Follows
step take
orstep choose
. - Example
-
step chooseleaf firstn 0 type row
- step emit
- Description
- Outputs the current value and empties the stack. Typically used at the end of a rule, but may also be used to pick from different trees in the same rule.
- Purpose
- A component of the rule.
- Prerequisite
-
Follows
step choose
. - Example
-
step emit
- firstn versus indep
- Description
-
Controls the replacement strategy CRUSH uses when OSDs are marked down in the CRUSH map. If this rule is to be used with replicated pools it should be
firstn
and if it is for erasure-coded pools it should beindep
. - Example
-
You have a PG stored on OSDs 1, 2, 3, 4, 5 in which 3 goes down.. In the first scenario, with the
firstn
mode, CRUSH adjusts its calculation to select 1 and 2, then selects 3 but discovers it is down, so it retries and selects 4 and 5, and then goes on to select a new OSD 6. The final CRUSH mapping change is from 1, 2, 3, 4, 5 to 1, 2, 4, 5, 6. In the second scenario, withindep
mode on an erasure-coded pool, CRUSH attempts to select the failed OSD 3, tries again and picks out 6, for a final transformation from 1, 2, 3, 4, 5 to 1, 2, 6, 4, 5.
A given CRUSH rule can be assigned to multiple pools, but it is not possible for a single pool to have multiple CRUSH rules.
2.7.1. Listing Rules
To list CRUSH rules from the command line, execute the following:
ceph osd crush rule list ceph osd crush rule ls
2.7.2. Dumping a Rule
To dump the contents of a specific CRUSH rule, execute the following:
ceph osd crush rule dump {name}
2.7.3. Adding a Simple Rule
To add a CRUSH rule, you must specify a rule name, the root node of the hierarchy you wish to use, the type of bucket you want to replicate across (for example, 'rack', 'row', and so on and the mode for choosing the bucket.
ceph osd crush rule create-simple {rulename} {root} {bucket-type} {firstn|indep}
Ceph will create a rule with chooseleaf
and 1 bucket of the type you specify.
For example:
ceph osd crush rule create-simple deleteme default host firstn
Create the following rule:
{ "rule_id": 1, "rule_name": "deleteme", "ruleset": 1, "type": 1, "min_size": 1, "max_size": 10, "steps": [ { "op": "take", "item": -1, "item_name": "default"}, { "op": "chooseleaf_firstn", "num": 0, "type": "host"}, { "op": "emit"}]}
The ruleset
is not supported in RHCS 3 and later releases. It is only present for backward compatibility for RHCS 2 and earlier releases of Ceph.
2.7.4. Adding a Replicated Rule
To create a CRUSH rule for a replicated pool, execute the following:
# ceph osd crush rule create-replicated <name> <root> <failure-domain> <class>
Where:
-
<name>
: The name of the rule. -
<root>
: The root of the CRUSH hierarchy. -
<failure-domain>
: The failure domain. For example:host
orrack
. -
<class>
: The storage device class. For example:hdd
orssd
. RHCS 3 and later only.
For example:
# ceph osd crush rule create-replicated fast default host ssd
2.7.5. Adding an Erasure Code Rule
To add a CRUSH rule for use with an erasure coded pool, you may specify a rule name and an erasure code profile.
ceph osd crush rule create-erasure {rulename} {profilename}
2.7.6. Removing a Rule
To remove a rule, execute the following and specify the CRUSH rule name:
ceph osd crush rule rm {name}
2.8. CRUSH Tunables
The Ceph project has grown exponentially with many changes and many new features. Beginning with the first commercially supported major release of Ceph, v0.48 (Argonaut), Ceph provides the ability to adjust certain parameters of the CRUSH algorithm, that is, the settings are not frozen in the source code.
A few important points to consider:
- Adjusting CRUSH values may result in the shift of some PGs between storage nodes. If the Ceph cluster is already storing a lot of data, be prepared for some fraction of the data to move.
-
The
ceph-osd
andceph-mon
daemons will start requiring the feature bits of new connections as soon as they receive an updated map. However, already-connected clients are effectively grandfathered in, and will misbehave if they do not support the new feature. Make sure when you upgrade your Ceph Storage Cluster daemons that you also update your Ceph clients. -
If the CRUSH tunables are set to non-legacy values and then later changed back to the legacy values,
ceph-osd
daemons will not be required to support the feature. However, the OSD peering process requires examining and understanding old maps. Therefore, you should not run old versions of theceph-osd
daemon if the cluster has previously used non-legacy CRUSH values, even if the latest version of the map has been switched back to using the legacy defaults.
2.8.1. The Evolution of CRUSH Tunables
Ceph clients and daemons prior to version 0.48 do not detect for tunables and are not compatible with version 0.48 and beyond, you must upgrade. The ability to adjust tunable CRUSH values has also evolved with major Ceph releases.
Legacy Values
Legacy values deployed in newer clusters with CRUSH Tunables may misbehave. Issues include:
- In Hierarchies with a small number of devices in the leaf buckets, some PGs map to fewer than the desired number of replicas. This commonly happens for hierarchies with "host" nodes with a small number (1-3) of OSDs nested beneath each one.
- For large clusters, some small percentages of PGs map to less than the desired number of OSDs. This is more prevalent when there are several layers of the hierarchy. For example, row, rack, host, osd.
- When some OSDs are marked out, the data tends to get redistributed to nearby OSDs instead of across the entire hierarchy.
Red Hat strongly encourage upgrading both Ceph clients and Ceph daemons to major supported releases to take advantage of CRUSH tunables. Red Hat recommends that all cluster daemons and clients use the same release version.
Argonaut (Legacy)
This is the first commercially supported release of Ceph.
Version Requirements:
- Ceph 0.48, 0.49 and later
- Linux kernel 3.6 or later, including the RBD kernel clients
Supported CRUSH Tunables:
-
choose_local_tries
: Number of local retries. Legacy value is 2, optimal value is 0. -
choose_local_fallback_tries
: Legacy value is 5, optimal value is 0. -
choose_total_tries
: Total number of attempts to choose an item. Legacy value was 19, subsequent testing indicates that a value of 50 is more appropriate for typical clusters. For extremely large clusters, a larger value might be necessary.
-
Bobtail
Version Requirements:
- Ceph 0.55, 0.56.x and later
- Linux kernel 3.9 or later, including the RBD kernel clients
Supported CRUSH Tunable:
-
chooseleaf_descend_once
: Whether a recursive chooseleaf attempt will retry, or only try once and allow the original placement to retry. Legacy default is 0, optimal value is 1.
-
Firefly
This is the first Red Hat supported version of Ceph.
Version Requirements:
- Red Hat Ceph Storage 1.2.3 and later
- Red Hat Enterprise Linux 7.1 or later, including the RBD kernel clients
Supported CRUSH Tunables:
-
chooseleaf_vary_r
: Whether a recursive chooseleaf attempt will start with a non-zero value of r, based on how many attempts the parent has already made. Legacy default is 0, but with this value CRUSH is sometimes unable to find a mapping. The optimal value, in terms of computational cost and correctness is 1. However, for legacy clusters that have lots of existing data, changing from 0 to 1 will cause a lot of data to move; a value of 4 or 5 will allow CRUSH to find a valid mapping but will make less data move. -
straw_calc_version
: There were some problems with the internal weights calculated and stored in the CRUSH map for straw buckets. Specifically, when there were items with a CRUSH weight of 0 or both a mix of weights and some duplicated weights CRUSH would distribute data incorrectly, that is, not in proportion to the weights. A value of 0 preserves the old, broken internal weight calculation; a value of 1 fixes the behavior. Setting thestraw_calc_version
to 1 and then adjusting a straw bucket, by either adding, removing, reweighting an item, or by using the reweight-all command, can trigger a small to moderate amount of data movement if the cluster has hit one of the problematic conditions. This tunable option is special because it has absolutely no impact concerning the required kernel version on the client side.
-
Hammer
The hammer
tunable profile does not affect the mapping of existing CRUSH maps simply by changing the tunable profile, but a new bucket type, straw2
is now supported.
Version Requirements:
- Red Hat Ceph Storage 1.3 and later
- Red Hat Enterprise Linux 7.1 or later, including the RBD kernel clients
New Bucket Type:
-
The new
straw2
bucket type fixes several limitations in the original straw bucket. Specifically, the old straw buckets would change some mappings that should have changed when a weight was adjusted, while straw2 buckets achieves the original goal of only changing mappings to or from the bucket item whose weight has changed. Thestraw2
bucket is the default for any newly created buckets. Changing a bucket type from straw to straw2 will result in a reasonably small amount of data movement, depending on how much the bucket item weights vary from each other. When the weights are all the same no data will move, and when item weights vary significantly there will be more movement.
-
The new
Jewel
Red Hat Ceph Storage 2 is supported on Red Hat Enterprise Linux 7.2 or above, but the jewel
tunable profile is only supported on Red Hat Enterprise Linux 7.3 or above. The jewel
tunable profile improves the overall behavior of CRUSH, such that it significantly reduces mapping changes when an OSD is marked out of the cluster.
Version Requirements:
- Red Hat Ceph Storage 2 and later
- Red Hat Enterprise Linux 7.3 or later, including the RBD kernel clients and the CephFS kernel clients
Supported CRUSH Tunable:
-
chooseleaf_stable
: Whether a recursive chooseleaf attempt will use a better value for an inner loop that greatly reduces the number of mapping changes when an OSD is marked out. The legacy value is 0, while the new value of 1 uses the new approach. Changing this value on an existing cluster will result in a very large amount of data movement as almost every PG mapping is likely to change.
-
2.8.2. Tuning CRUSH
Before you tune CRUSH, you should ensure that all Ceph clients and all Ceph daemons use the same version. If you have recently upgraded, ensure that you have restarted daemons and reconnected clients.
The simplest way to adjust the CRUSH tunables is by changing to a known profile. Those are:
-
legacy
: The legacy behavior from v0.47 (pre-Argonaut) and earlier. -
argonaut
: The legacy values supported by v0.48 (Argonaut) release. -
bobtail
: The values supported by the v0.56 (Bobtail) release. -
firefly
: The values supported by the v0.80 (Firefly) release. -
hammer
: The values supported by the v0.94 (Hammer) release. -
jewel
: The values supported by the v10.0.2 (Jewel) release. -
optimal
: The current best values. -
default
: The current default values for a new cluster.
You can select a profile on a running cluster with the command:
# ceph osd crush tunables <profile>
This may result in some data movement.
Generally, you should set the CRUSH tunables after you upgrade, or if you receive a warning. Starting with version v0.74, Ceph will issue a health warning if the CRUSH tunables are not set to their optimal values, the optimal values are the default as of v0.73. To make this warning go away, you have two options:
Adjust the tunables on the existing cluster. Note that this will result in some data movement (possibly as much as 10%). This is the preferred route, but should be taken with care on a production cluster where the data movement may affect performance. You can enable optimal tunables with:
# ceph osd crush tunables optimal
If things go poorly (for example, too much load) and not very much progress has been made, or there is a client compatibility problem (old kernel cephfs or rbd clients, or pre-bobtail librados clients), you can switch back to an earlier profile:
# ceph osd crush tunables <profile>
For example, to restore the pre-v0.48 (Argonaut) values, execute:
# ceph osd crush tunables legacy
You can make the warning go away without making any changes to CRUSH by adding the following option to the
[mon]
section of theceph.conf
file:mon warn on legacy crush tunables = false
For the change to take effect, restart the monitors, or apply the option to running monitors with:
# ceph tell mon.\* injectargs --no-mon-warn-on-legacy-crush-tunables
2.8.3. Tuning CRUSH, The Hard Way
If you can ensure that all clients are running recent code, you can adjust the tunables by extracting the CRUSH map, modifying the values, and reinjecting it into the cluster.
Extract the latest CRUSH map:
ceph osd getcrushmap -o /tmp/crush
Adjust tunables. These values appear to offer the best behavior for both large and small clusters we tested with. You will need to additionally specify the
--enable-unsafe-tunables
argument tocrushtool
for this to work. Please use this option with extreme care.:crushtool -i /tmp/crush --set-choose-local-tries 0 --set-choose-local-fallback-tries 0 --set-choose-total-tries 50 -o /tmp/crush.new
Reinject modified map:
ceph osd setcrushmap -i /tmp/crush.new
2.8.4. Legacy Values
For reference, the legacy values for the CRUSH tunables can be set with:
crushtool -i /tmp/crush --set-choose-local-tries 2 --set-choose-local-fallback-tries 5 --set-choose-total-tries 19 --set-chooseleaf-descend-once 0 --set-chooseleaf-vary-r 0 -o /tmp/crush.legacy
Again, the special --enable-unsafe-tunables
option is required. Further, as noted above, be careful running old versions of the ceph-osd
daemon after reverting to legacy values as the feature bit is not perfectly enforced.
2.9. Editing a CRUSH Map
Generally, modifying your CRUSH map at runtime with the Ceph CLI is more convenient than editing the CRUSH map manually. However, there are times when you may choose to edit it, such as changing the default bucket types, or using a bucket algorithm other than straw
.
To edit an existing CRUSH map:
- Get the CRUSH map.
- Decompile the CRUSH map.
- Edit at least one of devices, and Buckets and rules.
- Recompile the CRUSH map.
- Set the CRUSH map.
To activate a CRUSH Map rule for a specific pool, identify the common rule number and specify that rule number for the pool when creating the pool.
2.9.1. Getting a CRUSH Map
To get the CRUSH map for your cluster, execute the following:
ceph osd getcrushmap -o {compiled-crushmap-filename}
Ceph will output (-o) a compiled CRUSH map to the file name you specified. Since the CRUSH map is in a compiled form, you must decompile it first before you can edit it.
2.9.2. Decompiling a CRUSH Map
To decompile a CRUSH map, execute the following:
crushtool -d {compiled-crushmap-filename} -o {decompiled-crushmap-filename}
Ceph will decompile (-d) the compiled CRUSH map and output (-o) it to the file name you specified.
2.9.3. Compiling a CRUSH Map
To compile a CRUSH map, execute the following:
crushtool -c {decompiled-crush-map-filename} -o {compiled-crush-map-filename}
Ceph will store a compiled CRUSH map to the file name you specified.
2.9.4. Setting a CRUSH Map
To set the CRUSH map for your cluster, execute the following:
ceph osd setcrushmap -i {compiled-crushmap-filename}
Ceph will input the compiled CRUSH map of the file name you specified as the CRUSH map for the cluster.
2.10. CRUSH Storage Strategy Examples
Suppose you want to have most pools default to OSDs backed by large hard drives, but have some pools mapped to OSDs backed by fast solid-state drives (SSDs). CRUSH can handle these scenarios easily.
RHCS 2 and Earlier
In RHCS 2 and earlier releases, it’s possible to have multiple independent CRUSH hierarchies within the same CRUSH map to reflect different performance domains. Define two hierarchies with two different root nodes—one for hard disks (for example, "root platter") and one for SSDs (for example, "root ssd") as shown below:
device 0 osd.0 device 1 osd.1 device 2 osd.2 device 3 osd.3 device 4 osd.4 device 5 osd.5 device 6 osd.6 device 7 osd.7 host ceph-osd-ssd-server-1 { id -1 alg straw hash 0 item osd.0 weight 1.00 item osd.1 weight 1.00 } host ceph-osd-ssd-server-2 { id -2 alg straw hash 0 item osd.2 weight 1.00 item osd.3 weight 1.00 } host ceph-osd-platter-server-1 { id -3 alg straw hash 0 item osd.4 weight 1.00 item osd.5 weight 1.00 } host ceph-osd-platter-server-2 { id -4 alg straw hash 0 item osd.6 weight 1.00 item osd.7 weight 1.00 } root platter { id -5 alg straw hash 0 item ceph-osd-platter-server-1 weight 2.00 item ceph-osd-platter-server-2 weight 2.00 } root ssd { id -6 alg straw hash 0 item ceph-osd-ssd-server-1 weight 2.00 item ceph-osd-ssd-server-2 weight 2.00 } rule data { ruleset 0 type replicated min_size 2 max_size 2 step take platter step chooseleaf firstn 0 type host step emit } rule metadata { ruleset 1 type replicated min_size 0 max_size 10 step take platter step chooseleaf firstn 0 type host step emit } rule rbd { ruleset 2 type replicated min_size 0 max_size 10 step take platter step chooseleaf firstn 0 type host step emit } rule platter { ruleset 3 type replicated min_size 0 max_size 10 step take platter step chooseleaf firstn 0 type host step emit } rule ssd { ruleset 4 type replicated min_size 0 max_size 4 step take ssd step chooseleaf firstn 0 type host step emit } rule ssd-primary { ruleset 5 type replicated min_size 5 max_size 10 step take ssd step chooseleaf firstn 1 type host step emit step take platter step chooseleaf firstn -1 type host step emit }
You can then set a pool to use the SSD rule by executing:
ceph osd pool set <poolname> crush_ruleset 4
- NOTE
-
Red Hat does not support the
ruleset
andcrush_ruleset
settings in RHCS 3 and later releases.
Your SSD pool can serve as the fast storage pool. Similarly, you could use the ssd-primary
rule to cause each placement group in the pool to be placed with an SSD as the primary and platters as the replicas.
RHCS 3 and Later
In RHCS 3 and later releases, use device classes. The process is much simpler: add a class to each device. For example:
# ceph osd crush set-device-class <class> <osdId> [<osdId>] # ceph osd crush set-device-class hdd osd.0 osd.1 osd.4 osd.5 # ceph osd crush set-device-class ssd osd.2 osd.3 osd.6 osd.7
Then, create rules to use the devices.
# ceph osd crush rule create-replicated <rule-name> <root> <failure-domain-type> <device-class>: # ceph osd crush rule create-replicated cold default host hdd # ceph osd crush rule create-replicated hot default host ssd
Finally, set pools to use the rules.
ceph osd pool set <poolname> crush_rule <rule-name> ceph osd pool set cold crush_rule hdd ceph osd pool set hot crush_rule ssd
There is no need to manually edit the CRUSH map, because one hierarchy can serve multiple classes of devices. Compared to the foregoing example for RHCS 2, the CRUSH map is much simpler for RHCS 3 when using device classes.
device 0 osd.0 class hdd device 1 osd.1 class hdd device 2 osd.2 class ssd device 3 osd.3 class ssd device 4 osd.4 class hdd device 5 osd.5 class hdd device 6 osd.6 class ssd device 7 osd.7 class ssd host ceph-osd-server-1 { id -1 alg straw hash 0 item osd.0 weight 1.00 item osd.1 weight 1.00 item osd.2 weight 1.00 item osd.3 weight 1.00 } host ceph-osd-server-2 { id -2 alg straw hash 0 item osd.4 weight 1.00 item osd.5 weight 1.00 item osd.6 weight 1.00 item osd.7 weight 1.00 } root default { id -3 alg straw hash 0 item ceph-osd-server-1 weight 4.00 item ceph-osd-server-2 weight 4.00 } rule cold { ruleset 0 type replicated min_size 2 max_size 11 step take default class hdd step chooseleaf firstn 0 type host step emit } rule hot { ruleset 1 type replicated min_size 2 max_size 11 step take default class ssd step chooseleaf firstn 0 type host step emit }
Chapter 3. Placement Groups (PGs)
Placement Groups (PGs) are invisible to Ceph clients, but they play an important role in Ceph Storage Clusters.
A Ceph Storage Cluster may require many thousands of OSDs to reach an exabyte level of storage capacity. Ceph clients store objects in pools, which are a logical subset of the overall cluster. The number of objects stored in a pool may easily run into the millions and beyond. A system with millions of objects or more cannot realistically track placement on a per-object basis and still perform well. Ceph assigns objects to placement groups, and placement groups to OSDs to make re-balancing dynamic and efficient.
All problems in computer science can be solved by another level of indirection, except of course for the problem of too many indirections. | ||
-- David Wheeler |
3.1. About Placement Groups
Tracking object placement on a per-object basis within a pool is computationally expensive at scale. To facilitate high performance at scale, Ceph subdivides a pool into placement groups, assigns each individual object to a placement group, and assigns the placement group to a primary OSD. If an OSD fails or the cluster re-balances, Ceph can move or replicate an entire placement group—i.e., all of the objects in the placement groups—without having to address each object individually. This allows a Ceph cluster to re-balance or recover efficiently.
When CRUSH assigns a placement group to an OSD, it calculates a series of OSDs—the first being the primary. The osd_pool_default_size
setting minus 1
for replicated pools, and the number of coding chunks M
for erasure-coded pools determine the number of OSDs storing a placement group that can fail without losing data permanently. Primary OSDs use CRUSH to identify the secondary OSDs and copy the placement group’s contents to the secondary OSDs. For example, if CRUSH assigns an object to a placement group, and the placement group is assigned to OSD 5 as the primary OSD, if CRUSH calculates that OSD 1 and OSD 8 are secondary OSDs for the placement group, the primary OSD 5 will copy the data to OSDs 1 and 8. By copying data on behalf of clients, Ceph simplifies the client interface and reduces the client workload. The same process allows the Ceph cluster to recover and rebalance dynamically.
When the primary OSD fails and gets marked out of the cluster, CRUSH assigns the placement group to another OSD, which receives copies of objects in the placement group. Another OSD in the Up Set
will assume the role of the primary OSD.
When you increase the number of object replicas or coding chunks, CRUSH will assign each placement group to additional OSDs as required.
PGs do not own OSDs. CRUSH assigns many placement groups to each OSD pseudo-randomly to ensure that data gets distributed evenly across the cluster.
3.2. Placement Group Tradeoffs
Data durability and data distribution among all OSDs call for more placement groups but their number should be reduced to the minimum required for maximum performance to conserve CPU and memory resources.
3.2.1. Data Durability
Ceph strives to prevent the permanent loss of data. However, after an OSD fails, the risk of permanent data loss increases until the data it contained is fully recovered. Permanent data loss, though rare, is still possible. The following scenario describes how Ceph could permanently lose data in a single placement group with three copies of the data:
- An OSD fails and all copies of the object it contains are lost. For all objects within a placement group stored on the OSD, the number of replicas suddenly drops from three to two.
- Ceph starts recovery for each placement group stored on the failed OSD by choosing a new OSD to re-create the third copy of all objects for each placement group.
- The second OSD containing a copy of the same placement group fails before the new OSD is fully populated with the third copy. Some objects will then only have one surviving copy.
- Ceph picks yet another OSD and keeps copying objects to restore the desired number of copies.
- The third OSD containing a copy of the same placement group fails before recovery is complete. If this OSD contained the only remaining copy of an object, the object is lost permanently.
Hardware failure isn’t an exception, but an expectation. To prevent the foregoing scenario, ideally the recovery process should be as fast as reasonably possible. The size of your cluster, your hardware configuration and the number of placement groups play an important role in total recovery time.
Small clusters don’t recover as quickly.
In a cluster containing 10 OSDs with 512 placement groups in a three replica pool, CRUSH will give each placement group three OSDs. Each OSD will end up hosting (512 * 3) / 10 = ~150
placement groups. When the first OSD fails, the cluster will start recovery for all 150 placement groups simultaneously.
It is likely that Ceph stored the remaining 150 placement groups randomly across the 9 remaining OSDs. Therefore, each remaining OSD is likely to send copies of objects to all other OSDs and also receive some new objects, because the remaining OSDs become responsible for some of the 150 placement groups now assigned to them.
The total recovery time depends upon the hardware supporting the pool. For example, in a 10 OSD cluster, if a host contains one OSD with a 1TB SSD, and a 10GB/s switch connects each of the 10 hosts, the recovery time will take M
minutes. By contrast, if a host contains two SATA OSDs and a 1GB/s switch connects the five hosts, recovery will take substantially longer. Interestingly, in a cluster of this size, the number of placement groups has almost no influence on data durability. The placement group count could be 128 or 8192 and the recovery would not be slower or faster.
However, growing the same Ceph cluster to 20 OSDs instead of 10 OSDs is likely to speed up recovery and therefore improve data durability significantly. Why? Each OSD now participates in only 75 placement groups instead of 150. The 20 OSD cluster will still require all 19 remaining OSDs to perform the same amount of copy operations in order to recover. In the 10 OSD cluster, each OSDs had to copy approximately 100GB. In the 20 OSD cluster each OSD only has to copy 50GB each. If the network was the bottleneck, recovery will happen twice as fast. In other words, recovery time decreases as the number of OSDs increases.
In large clusters, PG count is important!
If the exemplary cluster grows to 40 OSDs, each OSD will only host 35 placement groups. If an OSD dies, recovery time will decrease unless another bottleneck precludes improvement. However, if this cluster grows to 200 OSDs, each OSD will only host approximately 7 placement groups. If an OSD dies, recovery will happen between at most of 21 (7 * 3)
OSDs in these placement groups: recovery will take longer than when there were 40 OSDs, meaning the number of placement groups should be increased!
No matter how short the recovery time, there is a chance for another OSD storing the placement group to fail while recovery is in progress.
In the 10 OSD cluster described above, if any OSD fails, then approximately 8 placement groups (i.e. 75 pgs / 9 osds
being recovered) will only have one surviving copy. And if any of the 8 remaining OSDs fail, the last objects of one placement group are likely to be lost (i.e. 8 pgs / 8 osds
with only one remaining copy being recovered). This is why starting with a somewhat larger cluster is preferred (e.g., 50 OSDs).
When the size of the cluster grows to 20 OSDs, the number of placement groups damaged by the loss of three OSDs drops. The second OSD lost will degrade approximately 2 (i.e. 35 pgs / 19 osds
being recovered) instead of 8 and the third OSD lost will only lose data if it is one of the two OSDs containing the surviving copy. In other words, if the probability of losing one OSD is 0.0001%
during the recovery time frame, it goes from 8 * 0.0001%
in the cluster with 10 OSDs to 2 * 0.0001%
in the cluster with 20 OSDs. Having 512 or 4096 placement groups is roughly equivalent in a cluster with less than 50 OSDs as far as data durability is concerned.
In a nutshell, more OSDs means faster recovery and a lower risk of cascading failures leading to the permanent loss of a placement group and its objects.
When you add an OSD to the cluster, it may take a long time top populate the new OSD with placement groups and objects. However there is no degradation of any object and adding the OSD has no impact on data durability.
3.2.2. Data Distribution
Ceph seeks to avoid hot spots—i.e., some OSDs receive substantially more traffic than other OSDs. Ideally, CRUSH assigns objects to placement groups evenly so that when the placement groups get assigned to OSDs (also pseudo randomly), the primary OSDs store objects such that they are evenly distributed across the cluster and hot spots and network over-subscription problems cannot develop because of data distribution.
Since CRUSH computes the placement group for each object, but does not actually know how much data is stored in each OSD within this placement group, the ratio between the number of placement groups and the number of OSDs may influence the distribution of the data significantly.
For instance, if there was only one a placement group with ten OSDs in a three replica pool, Ceph would only use three OSDs to store data because CRUSH would have no other choice. When more placement groups are available, CRUSH is more likely to be evenly spread objects across OSDs. CRUSH also evenly assigns placement groups to OSDs.
As long as there are one or two orders of magnitude more placement groups than OSDs, the distribution should be even. For instance, 300 placement groups for 3 OSDs, 1000 placement groups for 10 OSDs etc.
The ratio between OSDs and placement groups usually solves the problem of uneven data distribution for Ceph clients that implement advanced features like object striping. For example, a 4TB block device might get sharded up into 4MB objects.
The ratio between OSDs and placement groups does not address uneven data distribution in other cases, because CRUSH does not take object size into account. Using the librados
interface to store some relatively small objects and some very large objects can lead to uneven data distribution. For example, one million 4K objects totaling 4GB are evenly spread among 1000 placement groups on 10 OSDs. They will use 4GB / 10 = 400MB
on each OSD. If one 400MB object is added to the pool, the three OSDs supporting the placement group in which the object has been placed will be filled with 400MB + 400MB = 800MB
while the seven others will remain occupied with only 400MB.
3.2.3. Resource Usage
For each placement group, OSDs and Ceph monitors need memory, network and CPU at all times, and even more during recovery. Sharing this overhead by clustering objects within a placement group is one of the main reasons placement groups exist.
Minimizing the number of placement groups saves significant amounts of resources.
3.3. PG Count
The number of placement groups in a pool plays a significant role in how a cluster peers, distributes data and rebalances. Small clusters don’t see as many performance improvements compared to large clusters by increasing the number of placement groups. However, clusters that have many pools accessing the same OSDs may need to carefully consider PG count so that Ceph OSDs use resources efficiently.
Red Hat recommends 100 to 200 PGs per OSD.
3.3.1. PG Calculator
The PG calculator calculates the number of placement groups for you and addresses specific use cases. The PG calculator is especially helpful when using Ceph clients like the Ceph Object Gateway where there are many pools typically using same rule (CRUSH hierarchy). You may still calculate PGs manually using the guidelines in PG Count for Small Clusters and Calculating PG Count. However, the PG calculator is the preferred method of calculating PGs.
See Ceph Placement Groups (PGs) per Pool Calculator on the Red Hat Customer Portal for details.
3.3.2. Configuring Default PG Counts
When you create a pool, you also create a number of placement groups for the pool. If you don’t specify the number of placement groups, Ceph will use the default value of 8
, which is unacceptably low. You can increase the number of placement groups for a pool, but we recommend setting reasonable default values in your Ceph configuration file too.
osd pool default pg num = 100 osd pool default pgp num = 100
You need to set both the number of placement groups (total), and the number of placement groups used for objects (used in PG splitting). They should be equal.
3.3.3. PG Count for Small Clusters
Small clusters don’t benefit from large numbers of placement groups. As the number of OSDs increase, choosing the right value for pg_num
and pgp_num
becomes more important because it has a significant influence on the behavior of the cluster as well as the durability of the data when something goes wrong (i.e. the probability that a catastrophic event leads to data loss). It is important to use the PG calculator with small clusters.
3.3.4. Calculating PG Count
If you have more than 50 OSDs, we recommend approximately 50-100 placement groups per OSD to balance out resource usage, data durability and distribution. If you have less than 50 OSDs, choosing among the PG Count for Small Clusters is ideal. For a single pool of objects, you can use the following formula to get a baseline:
(OSDs * 100) Total PGs = ------------ pool size
Where pool size is either the number of replicas for replicated pools or the K+M
sum for erasure coded pools (as returned by ceph osd erasure-code-profile get
).
You should then check if the result makes sense with the way you designed your Ceph cluster to maximize data durability, data distribution and minimize resource usage.
The result should be rounded up to the nearest power of two. Rounding up is optional, but recommended for CRUSH to evenly balance the number of objects among placement groups.
For a cluster with 200 OSDs and a pool size of 3 replicas, you would estimate your number of PGs as follows:
(200 * 100) ----------- = 6667. Nearest power of 2: 8192 3
With 8192 placement groups distributed across 200 OSDs, that evaluates to approximately 41 placement groups per OSD. You also need to consider the number of pools you are likely to use in your cluster, since each pool will create placement groups too. Ensure that you have a reasonable maximum PG count.
3.3.5. Maximum PG Count
When using multiple data pools for storing objects, you need to ensure that you balance the number of placement groups per pool with the number of placement groups per OSD so that you arrive at a reasonable total number of placement groups. The aim is to achieve reasonably low variance per OSD without taxing system resources or making the peering process too slow.
In an exemplary Ceph Storage Cluster consisting of 10 pools, each pool with 512 placement groups on ten OSDs, there is a total of 5,120 placement groups spread over ten OSDs, or 512 placement groups per OSD. That may not use too many resources depending on your hardware configuration. By contrast, if you create 1,000 pools with 512 placement groups each, the OSDs will handle ~50,000 placement groups each and it would require significantly more resources. Operating with too many placement groups per OSD can significantly reduce performance, especially during rebalancing or recovery.
The Ceph Storage Cluster has a default maximum value of 300 placement groups per OSD. You can set a different maximum value in your Ceph configuration file.
mon pg warn max per osd
Ceph Object Gateways deploy with 10-15 pools, so you may consider using less than 100 PGs per OSD to arrive at a reasonable maximum number.
3.4. PG Command-Line Reference
The ceph
CLI allows you to set and get the number of placement groups for a pool, view the PG map and retrieve PG statistics.
3.4.1. Set the Number of PGs
To set the number of placement groups in a pool, you must specify the number of placement groups at the time you create the pool. See Create a Pool for details. Once you set placement groups for a pool, you can increase the number of placement groups (but you cannot decrease the number of placement groups). To increase the number of placement groups, execute the following:
ceph osd pool set {pool-name} pg_num {pg_num}
Once you increase the number of placement groups, you must also increase the number of placement groups for placement (pgp_num
) before your cluster will rebalance. The pgp_num
should be equal to the pg_num
. To increase the number of placement groups for placement, execute the following:
ceph osd pool set {pool-name} pgp_num {pgp_num}
3.4.2. Get the Number of PGs
To get the number of placement groups in a pool, execute the following:
ceph osd pool get {pool-name} pg_num
3.4.3. Get a Cluster PG Statistics
To get the statistics for the placement groups in your cluster, execute the following:
ceph pg dump [--format {format}]
Valid formats are plain
(default) and json
.
3.4.4. Get Statistics for Stuck PGs
To get the statistics for all placement groups stuck in a specified state, execute the following:
ceph pg dump_stuck {inactive|unclean|stale|undersized|degraded [inactive|unclean|stale|undersized|degraded...]} {<int>}
Inactive Placement groups cannot process reads or writes because they are waiting for an OSD with the most up-to-date data to come up and in.
Unclean Placement groups contain objects that are not replicated the desired number of times. They should be recovering.
Stale Placement groups are in an unknown state - the OSDs that host them have not reported to the monitor cluster in a while (configured by mon_osd_report_timeout
).
Valid formats are plain
(default) and json
. The threshold defines the minimum number of seconds the placement group is stuck before including it in the returned statistics (default 300 seconds).
3.4.5. Get a PG Map
To get the placement group map for a particular placement group, execute the following:
ceph pg map {pg-id}
For example:
ceph pg map 1.6c
Ceph will return the placement group map, the placement group, and the OSD status:
osdmap e13 pg 1.6c (1.6c) -> up [1,0] acting [1,0]
3.4.6. Get a PGs Statistics
To retrieve statistics for a particular placement group, execute the following:
ceph pg {pg-id} query
3.4.7. Scrub a Placement Group
To scrub a placement group, execute the following:
ceph pg scrub {pg-id}
Ceph checks the primary and any replica nodes, generates a catalog of all objects in the placement group and compares them to ensure that no objects are missing or mismatched, and their contents are consistent. Assuming the replicas all match, a final semantic sweep ensures that all of the snapshot-related object metadata is consistent. Errors are reported via logs.
3.4.8. Revert Lost
If the cluster has lost one or more objects, and you have decided to abandon the search for the lost data, you must mark the unfound objects as lost
.
If all possible locations have been queried and objects are still lost, you may have to give up on the lost objects. This is possible given unusual combinations of failures that allow the cluster to learn about writes that were performed before the writes themselves are recovered.
Currently the only supported option is "revert", which will either roll back to a previous version of the object or (if it was a new object) forget about it entirely. To mark the "unfound" objects as "lost", execute the following:
ceph pg {pg-id} mark_unfound_lost revert|delete
Use this feature with caution, because it may confuse applications that expect the object(s) to exist.
Chapter 4. Pools
Ceph clients store data in pools. When you create pools, you are creating an I/O interface for clients to store data. From the perspective of a Ceph client (i.e., block device, gateway, etc.), interacting with the Ceph storage cluster is remarkably simple: create a cluster handle and connect to the cluster; then, create an I/O context for reading and writing objects and their extended attributes.
Create a Cluster Handle and Connect to the Cluster
To connect to the Ceph storage cluster, the Ceph client needs the cluster name (usually ceph
by default) and an initial monitor address. Ceph clients usually retrieve these parameters using the default path for the Ceph configuration file and then read it from the file, but a user may also specify the parameters on the command line too. The Ceph client also provides a user name and secret key (authentication is on
by default). Then, the client contacts the Ceph monitor cluster and retrieves a recent copy of the cluster map, including its monitors, OSDs and pools.
Create a Pool I/O Context
To read and write data, the Ceph client creates an i/o context to a specific pool in the Ceph storage cluster. If the specified user has permissions for the pool, the Ceph client can read from and write to the specified pool.
Ceph’s architecture enables the storage cluster to provide this remarkably simple interface to Ceph clients so that clients may select one of the sophisticated storage strategies you define simply by specifying a pool name and creating an I/O context. Storage strategies are invisible to the Ceph client in all but capacity and performance. Similarly, the complexities of Ceph clients (mapping objects into a block device representation, providing an S3/Swift RESTful service) are invisible to the Ceph storage cluster.
A pool provides you with:
-
Resilience: You can set how many OSD are allowed to fail without losing data. For replicated pools, it is the desired number of copies/replicas of an object. A typical configuration stores an object and one additional copy (i.e.,
size = 2
), but you can determine the number of copies/replicas. For erasure coded pools, it is the number of coding chunks (i.e.m=2
in the erasure code profile) - Placement Groups: You can set the number of placement groups for the pool. A typical configuration uses approximately 50-100 placement groups per OSD to provide optimal balancing without using up too many computing resources. When setting up multiple pools, be careful to ensure you set a reasonable number of placement groups for both the pool and the cluster as a whole.
- CRUSH Rules: When you store data in a pool, a CRUSH rule mapped to the pool enables CRUSH to identify the rule for the placement of each object and its replicas (or chunks for erasure coded pools) in your cluster. You can create a custom CRUSH rule for your pool.
-
Snapshots: When you create snapshots with
ceph osd pool mksnap
, you effectively take a snapshot of a particular pool. -
Quotas: When you set quotas on a pool with
ceph osd pool set-quota
you may limit the maximum number of objects or the maximum number of bytes stored in the specified pool.
4.1. Pools and Storage Strategies
To manage pools, you can list, create, and remove pools. You can also view the utilization statistics for each pool.
4.2. List Pools
To list your cluster’s pools, execute:
ceph osd lspools
4.3. Create a Pool
Before creating pools, see the Pool, PG and CRUSH Configuration Reference chapter in the Configuration Guide for Red Hat Ceph Storage 3.
In Red Hat Ceph Storage 3 and later releases, system administrators must expressly enable a pool to receive I/O operations from Ceph clients. See Enable Application for details. Failure to enable a pool will result in a HEALTH_WARN
status.
It is better to adjust the default value for the number of placement groups in the Ceph configuration file, as the default value does not have to suit your needs. For example:
osd pool default pg num = 100 osd pool default pgp num = 100
To create a replicated pool, execute:
ceph osd pool create <pool-name> <pg-num> <pgp-num> [replicated] \ [crush-rule-name] [expected-num-objects]
To create an erasure-coded pool, execute:
ceph osd pool create <pool-name> <pg-num> <pgp-num> erasure \ [erasure-code-profile] [crush-rule-name] [expected-num-objects]
Where:
- pool-name
- Description
- The name of the pool. It must be unique.
- Type
- String
- Required
- Yes. If not specified, it is set to the value listed in the Ceph configuration file or to the default value.
- Default
-
ceph
- pg-num
- Description
-
The total number of placement groups for the pool. See the Placement Groups section and the Ceph Placement Groups (PGs) per Pool Calculator for details on calculating a suitable number. The default value
8
is not suitable for most systems. - Type
- Integer
- Required
- Yes
- Default
-
8
- pgp-num
- Description
- The total number of placement groups for placement purposes. This value must be equal to the total number of placement groups, except for placement group splitting scenarios.
- Type
- Integer
- Required
- Yes. If not specified it is set to the value listed in the Ceph configuration file or to the default value.
- Default
-
8
- replicated or erasure
- Description
-
The pool type which can be either
replicated
to recover from lost OSDs by keeping multiple copies of the objects orerasure
to get a kind of generalized RAID5 capability. The replicated pools require more raw storage but implement all Ceph operations. The erasure-coded pools require less raw storage but only implement a subset of the available operations. - Type
- String
- Required
- No
- Default
-
replicated
- crush-rule-name
- Description
-
The name of the crush rule for the pool. The rule MUST exist. For replicated pools, the name is the rule specified by the
osd_pool_default_crush_rule
configuration setting. For erasure-coded pools the name iserasure-code
if you specify the default erasure code profile or{pool-name}
otherwise. Ceph creates this rule with the specified name implicitly if the rule doesn’t already exist. - Type
- String
- Required
- No
- Default
-
Uses
erasure-code
for an erasure-coded pool. For replicated pools, it uses the value of theosd_pool_default_crush_rule
variable from the Ceph configuration.
- expected-num-objects
- Description
-
The expected number of objects for the pool. By setting this value together with a negative
filestore_merge_threshold
variable, Ceph splits the placement groups at pool creation time to avoid the latency impact to perform runtime directory splitting. - Type
- Integer
- Required
- No
- Default
-
0
, no splitting at the pool creation time
- erasure-code-profile
- Description
-
For erasure-coded pools only. Use the erasure code profile. It must be an existing profile as defined by the
osd erasure-code-profile set
variable in the Ceph configuration file. For further information, see the Erasure Code Profiles section. - Type
- String
- Required
- No
When you create a pool, set the number of placement groups to a reasonable value (for example to 100
). Consider the total number of placement groups per OSD too. Placement groups are computationally expensive, so performance will degrade when you have many pools with many placement groups, for example, 50 pools with 100 placement groups each. The point of diminishing returns depends upon the power of the OSD host.
See the Placement Groups section and Ceph Placement Groups (PGs) per Pool Calculator for details on calculating an appropriate number of placement groups for your pool.
4.4. Set Pool Quotas
You can set pool quotas for the maximum number of bytes or the maximum number of objects per pool or for both.
ceph osd pool set-quota <pool-name> [max_objects <obj-count>] [max_bytes <bytes>]
For example:
ceph osd pool set-quota data max_objects 10000
To remove a quota, set its value to 0
.
In-flight write operations may overrun pool quotas for a short time until Ceph propagates the pool usage across the cluster. This is normal behavior. Enforcing pool quotas on in-flight write operations would impose significant performance penalties.
4.5. Delete a Pool
To delete a pool, execute:
ceph osd pool delete <pool-name> [<pool-name> --yes-i-really-really-mean-it]
To protect data, in RHCS 3 and later releases, administrators cannot delete pools by default. Set the mon_allow_pool_delete
configuration option before deleting pools.
If a pool has its own rule, consider removing it after deleting the pool. If a pool has users strictly for its own use, consider deleting those users after deleting the pool.
4.6. Rename a Pool
To rename a pool, execute:
ceph osd pool rename <current-pool-name> <new-pool-name>
If you rename a pool and you have per-pool capabilities for an authenticated user, you must update the user’s capabilities (i.e., caps) with the new pool name.
4.7. Show Pool Statistics
To show a pool’s utilization statistics, execute:
rados df
4.8. Make a Snapshot of a Pool
To make a snapshot of a pool, execute:
ceph osd pool mksnap <pool-name> <snap-name>
If you create a pool snapshot, you will never be able to take RBD image snapshots within the pool and it will be irreversible.
4.9. Remove a Snapshot of a Pool
To remove a snapshot of a pool, execute:
ceph osd pool rmsnap <pool-name> <snap-name>
4.10. Set Pool Values
To set a value to a pool, execute the following command:
ceph osd pool set <pool-name> <key> <value>
The Pool Values section lists all key-values pairs that you can set.
4.11. Get Pool Values
To get a value from a pool, execute the following command:
ceph osd pool get <pool-name> <key>
The Pool Values section lists all key-values pairs that you can get.
4.12. Enable Application
RHCS 3 and later releases provide additional protection for pools to prevent unauthorized types of clients from writing data to the pool. This means that system administrators must expressly enable pools to receive I/O operations from Ceph Block Device, Ceph Object Gateway, Ceph Filesystem or for a custom application.
To enable a client application to conduct I/O operations on a pool, execute the following:
[root@host ~]# ceph osd pool application enable <poolname> <app> {--yes-i-really-mean-it}
Where <app>
is:
-
cephfs
for the Ceph Filesystem. -
rbd
for the Ceph Block Device -
rgw
for the Ceph Object Gateway
Specify a different <app>
value for a custom application.
A pool that is not enabled will generate a HEALTH_WARN
status. In that scenario, the output for ceph health detail -f json-pretty
will output the following:
{ "checks": { "POOL_APP_NOT_ENABLED": { "severity": "HEALTH_WARN", "summary": { "message": "application not enabled on 1 pool(s)" }, "detail": [ { "message": "application not enabled on pool '<pool-name>'" }, { "message": "use 'ceph osd pool application enable <pool-name> <app-name>', where <app-name> is 'cephfs', 'rbd', 'rgw', or freeform for custom applications." } ] } }, "status": "HEALTH_WARN", "overall_status": "HEALTH_WARN", "detail": [ "'ceph health' JSON format has changed in luminous. If you see this your monitoring system is scraping the wrong fields. Disable this with 'mon health preluminous compat warning = false'" ] }
- NOTE
-
Initialize pools for the Ceph Block Device with
rbd pool init <pool-name>
.
4.13. Disable Application
To disable a client application from conducting I/O operations on a pool, execute the following:
[root@host ~]# ceph osd pool application disable <poolname> <app> {--yes-i-really-mean-it}
Where <app>
is:
-
cephfs
for the Ceph Filesystem. -
rbd
for the Ceph Block Device -
rgw
for the Ceph Object Gateway
Specify a different <app>
value for a custom application.
4.14. Set Application Metadata
RHCS 3 and later releases provide functionality to set key-value pairs describing attributes of the client application.
To set client application metadata on a pool, execute the following:
[root@host ~]# ceph osd pool application set <poolname> <app> <key> <value>
Where <app>
is:
-
cephfs
for the Ceph Filesystem. -
rbd
for the Ceph Block Device -
rgw
for the Ceph Object Gateway
Specify a different <app>
value for a custom application.
4.15. Remove Application Metadata
To remove client application metadata on a pool, execute the following:
[root@host ~]# ceph osd pool application set <poolname> <app> <key>
Where <app>
is:
-
cephfs
for the Ceph Filesystem. -
rbd
for the Ceph Block Device -
rgw
for the Ceph Object Gateway
Specify a different <app>
value for a custom application.
4.16. Set the Number of Object Replicas
To set the number of object replicas on a replicated pool, execute the following command:
ceph osd pool set <poolname> size <num-replicas>
The <num-replicas>
parameter includes the object itself. If you want to include the object and two copies of the object for a total of three instances of the object, specify 3
.
For example:
ceph osd pool set data size 3
You can execute this command for each pool.
An object might accept I/O operations in degraded mode with fewer replicas than specified by the pool size
setting. To set a minimum number of required replicas for I/O, use the min_size
setting. For example:
ceph osd pool set data min_size 2
This ensures that no object in the data pool will receive I/O with fewer replicas than specified by the min_size
setting.
4.17. Get the Number of Object Replicas
To get the number of object replicas, execute the following command:
ceph osd dump | grep 'replicated size'
Ceph will list the pools, with the replicated size
attribute highlighted. By default, Ceph creates two replicas of an object, that is a total of three copies, or a size of 3
.
4.18. Pool Values
The following list contains key-values pairs that you can set or get. For further information, see the Set Pool Values and Get Pool Values sections.
- size
- Description
- Specifies the number of replicas for objects in the pool. See the Set the Number of Object Replicas section for further details. Applicable for the replicated pools only.
- Type
- Integer
- min_size
- Description
- Specifies the minimum number of replicas required for I/O. See the Set the Number of Object Replicas section for further details. Applicable for the replicated pools only.
- Type
- Integer
- crash_replay_interval
- Description
- Specifies the number of seconds to allow clients to replay acknowledged, but uncommitted requests.
- Type
- Integer
- pg-num
- Description
-
The total number of placement groups for the pool. See the Pool, PG and CRUSH Configuration Reference section in the Red Hat Ceph Storage 3 Configuration Guide for details on calculating a suitable number. The default value
8
is not suitable for most systems. - Type
- Integer
- Required
- Yes.
- Default
- 8
- pgp-num
- Description
- The total number of placement groups for placement purposes. This should be equal to the total number of placement groups, except for placement group splitting scenarios.
- Type
- Integer
- Required
- Yes. Picks up default or Ceph configuration value if not specified.
- Default
- 8
- Valid Range
-
Equal to or less than what specified by the
pg_num
variable.
- crush_rule
- Description
- The rule to use for mapping object placement in the cluster.
- Type
- String
- hashpspool
- Description
-
Enable or disable the
HASHPSPOOL
flag on a given pool. With this option enabled, pool hashing and placement group mapping are changed to improve the way pools and placement groups overlap. - Type
- Integer
- Valid Range
-
1
enables the flag,0
disables the flag.
Do not enable this option on production pools of a cluster with a large amount of OSDs and data. All placement groups in the pool would have to be remapped causing too much data movement.
- fast_read
- Description
-
On a pool that uses erasure coding, if this flag is enabled, the read request issues subsequent reads to all shards, and wait until it receives enough shards to decode to serve the client. In the case of the
jerasure
andisa erasure
plug-ins, once the first K replies return, client’s request is served immediately using the data decoded from these replies. This helps to allocate some resources for better performance. Currently this flag is only supported for erasure coding pools. - Type
- Boolean
- Defaults
-
0
- allow_ec_overwrites
- Description
- Whether writes to an erasure coded pool can update part of an object, so the Ceph Filesystem and Ceph Block Device can use it.
- Type
- Boolean
- Version
- RHCS 3 and later.
- compression_algorithm
- Description
-
Sets inline compression algorithm to use with the BlueStore storage backend. This setting overrides the
bluestore_compression_algorithm
configuration setting. - Type
- String
- Valid Settings
-
lz4
,snappy
,zlib
,zstd
- compression_mode
- Description
-
Sets the policy for the inline compression algorithm for the BlueStore storage backend. This setting overrides the
bluestore_compression_mode
configuration setting. - Type
- String
- Valid Settings
-
none
,passive
,aggressive
,force
- compression_min_blob_size
- Description
-
BlueStore will not compress chunks smaller than this size. This setting overrides the
bluestore_compression_min_blob_size
configuration setting. - Type
- Unsigned Integer
- compression_max_blob_size
- Description
-
BlueStore will break chunks larger than this size into smaller blobs of
compression_max_blob_size
before compressing the data. - Type
- Unsigned Integer
- nodelete
- Description
-
Set or unset the
NODELETE
flag on a given pool. - Type
- Integer
- Valid Range
-
1
sets flag.0
unsets flag.
- nopgchange
- Description
-
Set or unset the
NOPGCHANGE
flag on a given pool. - Type
- Integer
- Valid Range
-
1
sets the flag.0
unsets the flag.
- nosizechange
- Description
-
Set or unset the
NOSIZECHANGE
flag on a given pool. - Type
- Integer
- Valid Range
-
1
sets the flag.0
unsets the flag.
- write_fadvise_dontneed
- Description
-
Set or unset the
WRITE_FADVISE_DONTNEED
flag on a given pool. - Type
- Integer
- Valid Range
-
1
sets the flag.0
unsets the flag.
- noscrub
- Description
-
Set or unset the
NOSCRUB
flag on a given pool. - Type
- Integer
- Valid Range
-
1
sets the flag.0
unsets the flag.
nodeep-scrub
- Description
-
Set or unset the
NODEEP_SCRUB
flag on a given pool. - Type
- Integer
- Valid Range
1
sets the flag.0
unsets the flag.- scrub_min_interval
- Description
-
The minimum interval in seconds for pool scrubbing when load is low. If it is
0
, Ceph uses theosd_scrub_min_interval
configuration setting. - Type
- Double
- Default
0
- scrub_max_interval
- Description
-
The maximum interval in seconds for pool scrubbing irrespective of cluster load. If it is
0
, Ceph uses theosd_scrub_max_interval
configuration setting. - Type
- Double
- Default
0
- deep_scrub_interval
- Description
-
The interval in seconds for pool 'deep' scrubbing. If it is
0
, Ceph uses theosd_deep_scrub_interval
configuration setting. - Type
- Double
- Default
-
0
Chapter 5. Erasure Code Pools
Ceph storage strategies involve defining data durability requirements. Data durability means the ability to sustain the loss of one or more OSDs without losing data.
Ceph stores data in pools and there are two types of the pools:
- replicated
- erasure-coded
Ceph uses the replicated pools by default, meaning the Ceph copies every object from a primary OSD node to one or more secondary OSDs.
The erasure-coded pools reduces the amount of disk space required to ensure data durability but it is computationally a bit more expensive than replication.
Erasure coding is a method of storing an object in the Ceph storage cluster durably where the erasure code algorithm breaks the object into data chunks (k
) and coding chunks (m
), and stores those chunks in different OSDs.
In the event of the failure of an OSD, Ceph retrieves the remaining data (k
) and coding (m
) chunks from the other OSDs and the erasure code algorithm restores the object from those chunks.
Red Hat recommends min_size
for erasure-coded pools to be K+2
or more to prevent loss of writes and data.
Erasure coding uses storage capacity more efficiently than replication. The n-replication approach maintains n
copies of an object (3x by default in Ceph), whereas erasure coding maintains only k
+ m
chunks. For example, 3 data and 2 coding chunks use 1.5x the storage space of the original object.
While erasure coding uses less storage overhead than replication, the erasure code algorithm uses more RAM and CPU than replication when it accesses or recovers objects. Erasure coding is advantageous when data storage must be durable and fault tolerant, but do not require fast read performance (for example, cold storage, historical records, and so on).
For the mathematical and detailed explanation on how erasure code works in Ceph, see the Erasure Coded I/O section in the Architecture Guide for Red Hat Ceph Storage 3.
Ceph creates a default erasure code profile when initializing a cluster with k=2 and m=1, This mean that Ceph will spread the object data over three OSDs (k+m == 3) and Ceph can lose one of those OSDs without losing data. To know more about erasure code profiling see the Erasure Code Profiles section.
Configure only the .rgw.buckets
pool as erasure-coded and all other Ceph Object Gateway pools as replicated, otherwise an attempt to create a new bucket fails with the following error:
set_req_state_err err_no=95 resorting to 500
The reason for this is that erasure-coded pools do not support the omap
operations and certain Ceph Object Gateway metadata pools require the omap
support.
5.1. Creating a Sample Erasure-coded Pool
The simplest erasure coded pool is equivalent to RAID5 and requires at least three hosts:
$ ceph osd pool create ecpool 50 50 erasure pool 'ecpool' created $ echo ABCDEFGHI | rados --pool ecpool put NYAN - $ rados --pool ecpool get NYAN - ABCDEFGHI
The 50 in pool create stands for the number of placement groups.
5.2. Erasure Code Profiles
Ceph defines an erasure-coded pool with a profile. Ceph uses a profile when creating an erasure-coded pool and the associated CRUSH rule.
Ceph creates a default erasure code profile when initializing a cluster and it provides the same level of redundancy as two copies in a replicated pool. However, it uses 25% less storage capacity. The default profiles defines k=2 and m=1, meaning Ceph will spread the object data over three OSDs (k+m=3) and Ceph can lose one of those OSDs without losing data.
The default erasure code profile can sustain the loss of a single OSD. It is equivalent to a replicated pool with a size two, but requires 1.5 TB instead of 2 TB to store 1 TB of data. To display the default profile use the following command:
$ ceph osd erasure-code-profile get default directory=.libs k=2 m=1 plugin=jerasure crush-failure-domain=host technique=reed_sol_van
You can create a new profile to improve redundancy without increasing raw storage requirements. For instance, a profile with k=8 and m=4 can sustain the loss of four (m=4) OSDs by distributing an object on 12 (k+m=12) OSDs. Ceph divides the object into 8 chunks and computes 4 coding chunks for recovery. For example, if the object size is 8 MB, each data chunk is 1 MB and each coding chunk has the same size as the data chunk, that is also 1 MB. The object will not be lost even if four OSDs fail simultaneously.
The most important parameters of the profile are k, m and crush-failure-domain, because they define the storage overhead and the data durability.
Choosing the correct profile is important because you cannot change the profile after you create the pool. To modify a profile, you must create a new pool with a different profile and migrate the objects from the old pool to the new pool.
For instance, if the desired architecture must sustain the loss of two racks with a storage overhead of 40% overhead, the following profile can be defined:
$ ceph osd erasure-code-profile set myprofile \ k=4 \ m=2 \ crush-failure-domain=rack $ ceph osd pool create ecpool 12 12 erasure *myprofile* $ echo ABCDEFGHIJKL | rados --pool ecpool put NYAN - $ rados --pool ecpool get NYAN - ABCDEFGHIJKL
The primary OSD will divide the NYAN object into four (k=4) data chunks and create two additional chunks (m=2). The value of m defines how many OSDs can be lost simultaneously without losing any data. The crush-failure-domain=rack will create a CRUSH rule that ensures no two chunks are stored in the same rack.
Red Hat supports the following erasure coding values for k, and m:
- k=8 m=3
- k=8 m=4
- k=4 m=2
If the number of OSDs lost equals the number of coding chunks (m
), some placement groups in the erasure coding pool will go into incomplete state. If the number of OSDs lost is less than m
, no placement groups will go into incomplete state. In either situation, no data loss will occur. If placement groups are in incomplete state, temporarily reducing min_size
of an erasure coded pool will allow recovery.
5.2.1. OSD erasure-code-profile
Set
To create a new erasure code profile:
ceph osd erasure-code-profile set <name> \ [<directory=directory>] \ [<plugin=plugin>] \ [<stripe_unit=stripe_unit>] \ [<crush-device-class>]\ [<crush-failure-domain>]\ [<key=value> ...] \ [--force]
Where:
- directory
- Description
- Set the directory name from which the erasure code plug-in is loaded.
- Type
- String
- Required
- No.
- Default
-
/usr/lib/ceph/erasure-code
- plugin
- Description
- Use the erasure code plug-in to compute coding chunks and recover missing chunks. See the Erasure Code Plug-ins section for details.
- Type
- String
- Required
- No.
- Default
-
jerasure
- stripe_unit
- Description
-
The amount of data in a data chunk, per stripe. For example, a profile with 2 data chunks and
stripe_unit=4K
would put the range 0-4K in chunk 0, 4K-8K in chunk 1, then 8K-12K in chunk 0 again. This should be a multiple of 4K for best performance. The default value is taken from the monitor config optionosd_pool_erasure_code_stripe_unit
when a pool is created. The stripe_width of a pool using this profile will be the number of data chunks multiplied by thisstripe_unit
. - Type
- String
- Required
- No.
- Default
-
4K
- crush-device-class
- Description
-
The device class, such as
hdd
orssd
. Available only in RHCS 3 and later. - Type
- String
- Required
- No
- Default
-
none
, meaning CRUSH uses all devices regardless of class.
- crush-failure-domain
- Description
-
The failure domain, such as
host
orrack
. - Type
- String
- Required
- No
- Default
-
host
- key
- Description
- The semantic of the remaining key-value pairs is defined by the erasure code plug-in.
- Type
- String
- Required
- No.
- --force
- Description
- Override an existing profile by the same name.
- Type
- String
- Required
- No.
5.2.2. OSD erasure-code-profile
Remove
To remove an erasure code profile:
ceph osd erasure-code-profile rm <name>
If the profile is referenced by a pool, the deletion will fail.
5.2.3. OSD erasure-code-profile
Get
To display an erasure code profile:
ceph osd erasure-code-profile get <name>
5.2.4. OSD erasure-code-profile
List
To list the names of all erasure code profiles:
ceph osd erasure-code-profile ls
5.3. Erasure Coding with Overwrites (Technology Preview)
By default, erasure coded pools only work with the Ceph Object Gateway, which performs full object writes and appends. With Red Hat Ceph Storage 3, partial writes for an erasure coded pool can be enabled on a per-pool basis.
Using erasure coded pools with overwrites allows Ceph Block Devices and CephFS store their data in an erasure coded pool:
Syntax
ceph osd pool set <erasure_coded_pool_name> allow_ec_overwrites true
Example
$ ceph osd pool set ec_pool allow_ec_overwrites true
Enabling erasure coded pools with overwrites can only reside in a pool using BlueStore OSDs. Since BlueStore’s checksumming is used to detect bitrot or other corruption during deep scrubs. Using FileStore with erasure coded overwrites is unsafe, and yields lower performance when compared to BlueStore.
Erasure coded pools do not support omap. To use erasure coded pools with Ceph Block Devices and CephFS, store the data in an erasure coded pool, and the metadata in a replicated pool.
For Ceph Block Devices, use the --data-pool
option during image creation:
Syntax
rbd create --size <image_size>M|G|T --data-pool <erasure_coded_pool_name> <replicated_pool_name>/<image_name>
Example
$ rbd create --size 1G --data-pool ec_pool rep_pool/image01
If using erasure coded pools for CephFS, then setting the overwrites must be done in a file layout.
5.4. Erasure Code Plugins
Ceph supports erasure coding with a plug-in architecture, which means you can create erasure coded pools using different types of algorithms. Ceph supports:
- Jerasure (Default)
- Locally Repairable
- ISA (Intel only)
The following sections describe these plug-ins in greater detail.
5.4.1. Jerasure Erasure Code Plugin
The jerasure plug-in is the most generic and flexible plug-in. It is also the default for Ceph erasure coded pools.
The jerasure plug-in encapsulates the JerasureH library. For detailed information about the parameters, see the jerasure documentation.
To create a new erasure code profile using the jerasure plug-in, run the following command:
ceph osd erasure-code-profile set <name> \ plugin=jerasure \ k=<data-chunks> \ m=<coding-chunks> \ technique=<reed_sol_van|reed_sol_r6_op|cauchy_orig|cauchy_good|liberation|blaum_roth|liber8tion> \ [crush-root=<root>] \ [crush-failure-domain=<bucket-type>] \ [directory=<directory>] \ [--force]
Where:
- k
- Description
- Each object is split in data-chunks parts, each stored on a different OSD.
- Type
- Integer
- Required
- Yes.
- Example
-
4
- m
- Description
- Compute coding chunks for each object and store them on different OSDs. The number of coding chunks is also the number of OSDs that can be down without losing data.
- Type
- Integer
- Required
- Yes.
- Example
- 2
- technique
- Description
- The more flexible technique is reed_sol_van; it is enough to set k and m. The cauchy_good technique can be faster but you need to chose the packetsize carefully. All of reed_sol_r6_op, liberation, blaum_roth, liber8tion are RAID6 equivalents in the sense that they can only be configured with m=2.
- Type
- String
- Required
- No.
- Valid Setttings
-
reed_sol_van
reed_sol_r6_op
cauchy_orig
cauchy_good
liberation
blaum_roth
liber8tion
- Default
-
reed_sol_van
- packetsize
- Description
- The encoding will be done on packets of bytes size at a time. Choosing the correct packet size is difficult. The jerasure documentation contains extensive information on this topic.
- Type
- Integer
- Required
- No.
- Default
-
2048
- crush-root
- Description
- The name of the CRUSH bucket used for the first step of the rule. For instance step take default.
- Type
- String
- Required
- No.
- Default
- default
- crush-failure-domain
- Description
- Ensure that no two chunks are in a bucket with the same failure domain. For instance, if the failure domain is host no two chunks will be stored on the same host. It is used to create a rule step such as step chooseleaf host.
- Type
- String
- Required
- No.
- Default
-
host
- directory
- Description
- Set the directory name from which the erasure code plug-in is loaded.
- Type
- String
- Required
- No.
- Default
-
/usr/lib/ceph/erasure-code
- --force
- Description
- Override an existing profile by the same name.
- Type
- String
- Required
- No.
5.4.2. Locally Repairable Erasure Code (LRC) Plugin
With the jerasure plug-in, when Ceph stores an erasure-coded object on multiple OSDs, recovering from the loss of one OSD requires reading from all the others. For instance if you configure jerasure with k=8 and m=4, losing one OSD requires reading from the eleven others to repair.
The lrc erasure code plug-in creates local parity chunks to be able to recover using fewer OSDs. For instance if you configure lrc with k=8, m=4 and l=4, it will create an additional parity chunk for every four OSDs. When Ceph loses a single OSD, it can recover the object data with only four OSDs instead of eleven.
Although it is probably not an interesting use case when all hosts are connected to the same switch, you can actually observe reduced bandwidth usage between racks when using the lrc erasure code plug-in.
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ k=4 m=2 l=3 \ crush-failure-domain=host $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
In 1.2 version, you can only observe reduced bandwidth if the primary OSD is in the same rack as the lost chunk.:
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ k=4 m=2 l=3 \ crush-locality=rack \ crush-failure-domain=host $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
5.4.2.1. Create an LRC Profile
To create a new LRC erasure code profile, run the following command:
ceph osd erasure-code-profile set <name> \ plugin=lrc \ k=<data-chunks> \ m=<coding-chunks> \ l=<locality> \ [crush-root=<root>] \ [crush-locality=<bucket-type>] \ [crush-failure-domain=<bucket-type>] \ [directory=<directory>] \ [--force]
Where:
- k
- Description
- Each object is split in data-chunks parts, each stored on a different OSD.
- Type
- Integer
- Required
- Yes.
- Example
-
4
- m
- Description
- Compute coding chunks for each object and store them on different OSDs. The number of coding chunks is also the number of OSDs that can be down without losing data.
- Type
- Integer
- Required
- Yes.
- Example
-
2
- l
- Description
- Group the coding and data chunks into sets of size locality. For instance, for k=4 and m=2, when locality=3 two groups of three are created. Each set can be recovered without reading chunks from another set.
- Type
- Integer
- Required
- Yes.
- Example
-
3
- crush-root
- Description
- The name of the crush bucket used for the first step of the rule. For instance step take default.
- Type
- String
- Required
- No.
- Default
- default
- crush-locality
- Description
- The type of the crush bucket in which each set of chunks defined by l will be stored. For instance, if it is set to rack, each group of l chunks will be placed in a different rack. It is used to create a rule step such as step choose rack. If it is not set, no such grouping is done.
- Type
- String
- Required
- No.
- crush-failure-domain
- Description
- Ensure that no two chunks are in a bucket with the same failure domain. For instance, if the failure domain is host no two chunks will be stored on the same host. It is used to create a rule step such as step chooseleaf host.
- Type
- String
- Required
- No.
- Default
-
host
- directory
- Description
- Set the directory name from which the erasure code plug-in is loaded.
- Type
- String
- Required
- No.
- Default
-
/usr/lib/ceph/erasure-code
- --force
- Description
- Override an existing profile by the same name.
- Type
- String
- Required
- No.
5.4.2.2. Create a Low-level LRC Profile
The sum of k and m must be a multiple of the l parameter. The low level configuration parameters do not impose such a restriction and it may be more convenient to use it for specific purposes. It is for instance possible to define two groups, one with 4 chunks and another with 3 chunks. It is also possible to recursively define locality sets, for instance data centers and racks into data centers. The k/m/l are implemented by generating a low level configuration.
The lrc erasure code plug-in recursively applies erasure code techniques so that recovering from the loss of some chunks only requires a subset of the available chunks, most of the time.
For instance, when three coding steps are described as:
chunk nr 01234567 step 1 _cDD_cDD step 2 cDDD____ step 3 ____cDDD
where c are coding chunks calculated from the data chunks D, the loss of chunk 7 can be recovered with the last four chunks. And the loss of chun 2 chunk can be recovered with the first four chunks.
The miminal testing scenario is strictly equivalent to using the default erasure-code profile. The DD implies K=2, the c implies M=1 and uses the jerasure plug-in by default.
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ mapping=DD_ \ layers='[ [ "DDc", "" ] ]' $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
The lrc plug-in is particularly useful for reducing inter-rack bandwidth usage. Although it is probably not an interesting use case when all hosts are connected to the same switch, reduced bandwidth usage can actually be observed. It is equivalent to k=4, m=2 and l=3 although the layout of the chunks is different:
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ mapping=__DD__DD \ layers='[ [ "_cDD_cDD", "" ], [ "cDDD____", "" ], [ "____cDDD", "" ], ]' $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
In Firefly the reduced bandwidth will only be observed if the primary OSD is in the same rack as the lost chunk.:
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ mapping=__DD__DD \ layers='[ [ "_cDD_cDD", "" ], [ "cDDD____", "" ], [ "____cDDD", "" ], ]' \ crush-steps='[ [ "choose", "rack", 2 ], [ "chooseleaf", "host", 4 ], ]' $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
LRC now uses jerasure as the default EC backend. It is possible to specify the EC backend and algorithm on a per layer basis using the low level configuration. The second argument in layers='[ [ "DDc", "" ] ]' is actually an erasure code profile to be used for this level. The example below specifies the ISA backend with the Cauchy technique to be used in the lrcpool.:
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ mapping=DD_ \ layers='[ [ "DDc", "plugin=isa technique=cauchy" ] ]' $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
You could also use a different erasure code profile for for each layer.:
$ ceph osd erasure-code-profile set LRCprofile \ plugin=lrc \ mapping=__DD__DD \ layers='[ [ "_cDD_cDD", "plugin=isa technique=cauchy" ], [ "cDDD____", "plugin=isa" ], [ "____cDDD", "plugin=jerasure" ], ]' $ ceph osd pool create lrcpool 12 12 erasure LRCprofile
5.4.3. Controlling CRUSH Placement
The default CRUSH rule provides OSDs that are on different hosts. For instance:
chunk nr 01234567 step 1 _cDD_cDD step 2 cDDD____ step 3 ____cDDD
needs exactly 8 OSDs, one for each chunk. If the hosts are in two adjacent racks, the first four chunks can be placed in the first rack and the last four in the second rack. Recovering from the loss of a single OSD does not require using bandwidth between the two racks.
For instance:
crush-steps='[ [ "choose", "rack", 2 ], [ "chooseleaf", "host", 4 ] ]'
will create a rule that will select two crush buckets of type rack and for each of them choose four OSDs, each of them located in different bucket of type host.
The rule can also be created manually for finer control.
5.5. ISA Erasure Code Plugin
The isa plug-in is encapsulates the ISA library. It only runs on Intel processors.
To create a new erasure code profile using the isa plug-in, run the following command:
ceph osd erasure-code-profile set <name> \ plugin=isa \ technique=<reed_sol_van|cauchy> \ [k=<data-chunks>] \ [m=<coding-chunks>] \ [crush-root=<root>] \ [crush-failure-domain=<bucket-type>] \ [directory=<directory>] \ [--force]
Where:
- technique
- Description
- The ISA plug-in comes in two Reed Solomon forms. If reed_sol_van is set, it is Vandermonde, if cauchy is set, it is Cauchy.
- Type
- String
- Required
- No.
- Valid Settings
-
reed_sol_van
cauchy
- Default
-
reed_sol_van
- k
- Description
- Each object is split in data-chunks parts, each stored on a different OSD.
- Type
- Integer
- Required
- No.
- Default
-
7
- m
- Description
- Compute coding chunks for each object and store them on different OSDs. The number of coding chunks is also the number of OSDs that can be down without losing data.
- Type
- Integer
- Required
- No.
- Default
-
3
- crush-root
- Description
- The name of the crush bucket used for the first step of the rule. For intance step take default.
- Type
- String
- Required
- No.
- Default
- default
- crush-failure-domain
- Description
- Ensure that no two chunks are in a bucket with the same failure domain. For instance, if the failure domain is host no two chunks will be stored on the same host. It is used to create a rule step such as step chooseleaf host.
- Type
- String
- Required
- No.
- Default
-
host
- directory
- Description
- Set the directory name from which the erasure code plug-in is loaded.
- Type
- String
- Required
- No.
- Default
-
/usr/lib/ceph/erasure-code
- --force
- Description
- Override an existing profile by the same name.
- Type
- String
- Required
- No.
5.6. Pyramid Erasure Code
Pyramid erasure code is basically an improved algorithm based on the original erasure code algorithm to improve the access and recovery of data. It is simply a derivation of any existing MDS (Maximum Distance Separable) code like Reed-Solomon
for any given (n, k) where k
is the original number of data chunks and n
is the total number of data chunks after the coding process. As Pyramid code is based on standard MDS code, it has the same encoding/decoding approach. A pyramid coded data pool has the same write overhead and same recovery ability of arbitrary failures like a normal erasure coded pool. The only advantage of pyramid code is that it significantly reduces the read overhead in comparison to normal erasure code.
Let’s take an example of a (11, 8) erasure coded pool and apply the pyramid code approach on it. We will convert the (11, 8) erasure coded pool into a (12, 8) Pyramid coded pool. The erasure coded pool has 8
data chunks and 11 - 8 = 3
redundant chunks. The Pyramid Code separates the 8 data chunks into 2 equal size groups, say P1 = {a1, a2, a3, a4} and P2 = {a5, a6, a7, a8}. It keeps two of the redundant chunks from the erasure code unchanged (say b2 and b3). These two chunks are now called global redundant chunks, because they cover all the 8 data chunks. Next, a new redundant chunk is computed for group P1, which is denoted as group (or local) redundant chunk b1,1. The computation is done as if computing b1 in the original erasure code, except for setting all the data chunks in P2 to 0. Similarly, a group redundant chunks b1,2 is computed for P2. The group redundant chunks are only affected by data chunks in the corresponding groups and not by other groups at all. The group redundant chunks can be interpreted as the projection of the original redundant chunk in the erasure code onto each group i.e, b1,1 + b1,2 = b1. So, in this approach, when one data chunk fails, it can be recovered using a read overhead of 4 instead of 8 for the normal erasure coded pool. For instance, if a3
in P1
fails and a1
is the primary OSD in charge of scrubbing, it can use a1
, a2
, a4
and b1,1
to recover a3
instead of reading from P2
. Reading all chunks is only required when more than one chunk goes missing in the same group.
This Pyramid code has the same write overhead as the original erasure code. Whenever any data chunk is updated, the Pyramid Code needs to update 3 redundant chunks (both b2, b3, plus either b1,1 or b1,2), while the erasure code also updates 3 redundant chunks (b1, b2 and b3). Also, it can recover 3 arbitrary erasures same as the normal erasure code. The only benefit is lesser read overhead as mentioned above which comes at the cost of using one additional redundant chunk. So, Pyramid code trades storage space for access efficiency.
The below diagram shows the constructed pyramid code: