검색

이 콘텐츠는 선택한 언어로 제공되지 않습니다.

Chapter 3. Monitoring

download PDF

Once you have a running cluster, you may begin monitoring the storage cluster to ensure that the Ceph Monitor and OSD daemons are running, at a high-level. Ceph storage cluster clients must connect to a Ceph monitor and receive the latest version of the Ceph cluster map before they can read and write data to the Ceph pools of the storage cluster. So the monitor cluster must have agreement on the state of the cluster before Ceph clients can read and write data.

Ceph OSDs must peer the placement groups on the primary OSD with the copies of the placement groups on secondary OSDs. If faults arise, peering will reflect something other than the active + clean state.

3.1. High-level Monitoring

High level monitoring of a storage cluster typically involves checking the status of Ceph OSD and Monitor daemons to ensure that they are up and running. High level monitoring also involves checking the storage cluster capacity to ensure that the cluster doesn’t exceed its full ratio. The Calamari instance on the Ansible Tower or Red Hat Storage Console node is the most common way to conduct high-level monitoring. However, you may also use the command line, the admin socket or the Ceph API to monitor the storage cluster.

3.1.1. Interactive Mode

To run the ceph utility in interactive mode, type ceph at the command line with no arguments, for example:

# ceph
ceph> health
ceph> status
ceph> quorum_status
ceph> mon_status

3.1.2. Checking Cluster Health

After you start the Ceph storage cluster, and before you start reading and/or writing data, check the storage cluster’s health first. You can check on the health of the Ceph storage cluster with the following:

# ceph health

If you specified non-default locations for the configuration or keyring, you may specify their locations:

# ceph -c /path/to/conf -k /path/to/keyring health

Upon starting the Ceph cluster, you will likely encounter a health warning such as HEALTH_WARN XXX num placement groups stale. Wait a few moments and check it again. When the storage cluster is ready, ceph health should return a message such as HEALTH_OK. At that point, it is okay to begin using the cluster.

3.1.3. Watching a Cluster

To watch the cluster’s ongoing events on the command line, open a new terminal. Then, enter:

# ceph -w

Ceph will print each event. For example, a tiny Ceph cluster consisting of one monitor and two OSDs may print the following:

cluster b370a29d-9287-4ca3-ab57-3d824f65e339
 health HEALTH_OK
 monmap e1: 1 mons at {ceph1=10.0.0.8:6789/0}, election epoch 2, quorum 0 ceph1
 osdmap e63: 2 osds: 2 up, 2 in
  pgmap v41338: 952 pgs, 20 pools, 17130 MB data, 2199 objects
        115 GB used, 167 GB / 297 GB avail
             952 active+clean

2014-06-02 15:45:21.655871 osd.0 [INF] 17.71 deep-scrub ok
2014-06-02 15:45:47.880608 osd.1 [INF] 1.0 scrub ok
2014-06-02 15:45:48.865375 osd.1 [INF] 1.3 scrub ok
2014-06-02 15:45:50.866479 osd.1 [INF] 1.4 scrub ok
2014-06-02 15:45:01.345821 mon.0 [INF] pgmap v41339: 952 pgs: 952 active+clean; 17130 MB data, 115 GB used, 167 GB / 297 GB avail
2014-06-02 15:45:05.718640 mon.0 [INF] pgmap v41340: 952 pgs: 1 active+clean+scrubbing+deep, 951 active+clean; 17130 MB data, 115 GB used, 167 GB / 297 GB avail
2014-06-02 15:45:53.997726 osd.1 [INF] 1.5 scrub ok
2014-06-02 15:45:06.734270 mon.0 [INF] pgmap v41341: 952 pgs: 1 active+clean+scrubbing+deep, 951 active+clean; 17130 MB data, 115 GB used, 167 GB / 297 GB avail
2014-06-02 15:45:15.722456 mon.0 [INF] pgmap v41342: 952 pgs: 952 active+clean; 17130 MB data, 115 GB used, 167 GB / 297 GB avail
2014-06-02 15:46:06.836430 osd.0 [INF] 17.75 deep-scrub ok
2014-06-02 15:45:55.720929 mon.0 [INF] pgmap v41343: 952 pgs: 1 active+clean+scrubbing+deep, 951 active+clean; 17130 MB data, 115 GB used, 167 GB / 297 GB avail

The output provides:

  • Cluster ID
  • Cluster health status
  • The monitor map epoch and the status of the monitor quorum
  • The OSD map epoch and the status of OSDs
  • The placement group map version
  • The number of placement groups and pools
  • The notional amount of data stored and the number of objects stored
  • The total amount of data stored

How Ceph Calculates Data Usage

The used value reflects the actual amount of raw storage used. The xxx GB / xxx GB value means the amount available, the lesser of the two numbers, of the overall storage capacity of the cluster. The notional number reflects the size of the stored data before it is replicated, cloned or snapshotted. Therefore, the amount of data actually stored typically exceeds the notional amount stored, because Ceph creates replicas of the data and may also use storage capacity for cloning and snapshotting.

3.1.4. Checking a Cluster’s Usage Statistics

To check a cluster’s data usage and data distribution among pools, you can use the df option. It is similar to Linux df. Execute the following:

# ceph df

The GLOBAL section of the output provides an overview of the amount of storage the storage cluster uses for data.

  • SIZE: The overall storage capacity of the storage cluster.
  • AVAIL: The amount of free space available in the storage cluster.
  • RAW USED: The amount of raw storage used.
  • % RAW USED: The percentage of raw storage used. Use this number in conjunction with the full ratio and near full ratio to ensure that you are not reaching the storage cluster’s capacity.

The POOLS section of the output provides a list of pools and the notional usage of each pool. The output from this section DOES NOT reflect replicas, clones or snapshots. For example, if you store an object with 1MB of data, the notional usage will be 1MB, but the actual usage may be 3MB or more depending on the number of replicas (e.g., size = 3, clones and snapshots.

  • NAME: The name of the pool.
  • ID: The pool ID.
  • USED: The notional amount of data stored in kilobytes, unless the number appends M for megabytes or G for gigabytes.
  • %USED: The notional percentage of storage used per pool.
  • Objects: The notional number of objects stored per pool.
Note

The numbers in the POOLS section are notional. They are not inclusive of the number of replicas, shapshots or clones. As a result, the sum of the USED and %USED amounts will not add up to the RAW USED and %RAW USED amounts in the GLOBAL section of the output. See How Ceph Calculates Data Usage for details.

3.1.5. Checking a Cluster’s Status

To check a cluster’s status, execute the following:

# ceph status

Or:

# ceph -s

In interactive mode, type status and press Enter. :

ceph> status

Ceph will print the cluster status. For example, a tiny Ceph cluster consisting of one monitor, and two OSDs may print the following:

cluster b370a29d-9287-4ca3-ab57-3d824f65e339
 health HEALTH_OK
 monmap e1: 1 mons at {ceph1=10.0.0.8:6789/0}, election epoch 2, quorum 0 ceph1
 osdmap e63: 2 osds: 2 up, 2 in
  pgmap v41332: 952 pgs, 20 pools, 17130 MB data, 2199 objects
        115 GB used, 167 GB / 297 GB avail
               1 active+clean+scrubbing+deep
             951 active+clean

3.1.6. Checking Monitor Status

If the storage cluster has multiple monitors, which is required for high availability for production Ceph storage clusters. You should check the Ceph Monitor quorum status after you start the Ceph storage cluster before reading and/or writing data. A quorum must be present when multiple monitors are running. You should also check Ceph Monitor status periodically to ensure that they are running. If there is a problem with the Monitor, that prevents an agreement on the state of the storage cluster, the fault may prevent Ceph clients from reading and writing data.

  • To display the monitor map, execute the following:

    # ceph mon stat

    or

    # ceph mon dump
  • To check the quorum status for the storage cluster, execute the following:

    # ceph quorum_status -f json-pretty

    Ceph will return the quorum status. For example, a Ceph storage cluster consisting of three monitors may return the following:

    { "election_epoch": 10,
      "quorum": [
            0,
            1,
            2],
      "monmap": { "epoch": 1,
          "fsid": "444b489c-4f16-4b75-83f0-cb8097468898",
          "modified": "2011-12-12 13:28:27.505520",
          "created": "2011-12-12 13:28:27.505520",
          "mons": [
                { "rank": 0,
                  "name": "a",
                  "addr": "127.0.0.1:6789\/0"},
                { "rank": 1,
                  "name": "b",
                  "addr": "127.0.0.1:6790\/0"},
                { "rank": 2,
                  "name": "c",
                  "addr": "127.0.0.1:6791\/0"}
               ]
        }
    }

3.1.7. Using the Administration Socket

Use the administration socket to interact with a given daemon directly by using a UNIX socket file. For example, the socket enables you to:

  • List the Ceph configuration at runtime
  • Set configuration values at runtime directly without relaying on Monitors. This is useful when Monitors are down.
  • Dump historic operations
  • Dump the operation priority queue state
  • Dump operations without rebooting
  • Dump performance counters

In addition, using the socket is helpful when troubleshooting problems related to Monitors or OSDs. For details, see the Troubleshooting Guide for Red Hat Ceph Storage 3.

To use the socket:

ceph daemon <type>.<id> <command>

Replace:

  • <type> with the type of the Ceph daemon (mon, osd, mds).
  • <id> with the daemon ID
  • <command> with the command to run. Use help to list the available commands for a given daemon.

For example, to view a Monitor status of a Monitor named mon.0:

# ceph daemon mon.0 mon_status

Alternatively, specify the daemon by using its socket file.

ceph daemon /var/run/ceph/<socket-file> <command>

For example, to view the status of an OSD named osd.2:

# ceph daemon /var/run/ceph/ceph-osd.2.asok status

To list all socket files for the Ceph processes:

$ ls /var/run/ceph

3.1.8. Checking OSD Status

An OSD’s status is either in the cluster, in, or out of the cluster, out; and, it is either up and running, up, or it is down and not running, or down. If an OSD is up, it may be either in the storage cluster, which data can be read and written, or it is out of the storage cluster. If it was in the cluster and recently moved out of the cluster, Ceph will migrate placement groups to other OSDs. If an OSD is out of the cluster, CRUSH will not assign placement groups to the OSD. If an OSD is down, it should also be out.

Note

If an OSD is down and in, there is a problem and the cluster will not be in a healthy state.

OSD States

If you execute a command such as ceph health, ceph -s or ceph -w, you may notice that the cluster does not always echo back HEALTH OK. Don’t panic. With respect to OSDs, you should expect that the cluster will NOT echo HEALTH OK in a few expected circumstances:

  • You haven’t started the cluster yet, it won’t respond.
  • You have just started or restarted the cluster and it’s not ready yet, because the placement groups are getting created and the OSDs are in the process of peering.
  • You just added or removed an OSD.
  • You just have modified the cluster map.

An important aspect of monitoring OSDs is to ensure that when the cluster is up and running that all OSDs that are in the cluster are up and running, too. To see if all OSDs are running, execute:

# ceph osd stat

or

# ceph osd dump

The result should tell you the map epoch, eNNNN, the total number of OSDs, x, how many, y, are up, and how many, z, are in:

eNNNN: x osds: y up, z in

If the number of OSDs that are in the cluster is more than the number of OSDs that are up, execute the following command to identify the ceph-osd daemons that aren’t running:

# ceph osd tree

Example output:

# id    weight  type name   up/down reweight
-1  3   pool default
-3  3       rack mainrack
-2  3           host osd-host
0   1               osd.0   up  1
1   1               osd.1   up  1
2   1               osd.2   up  1
Tip

The ability to search through a well-designed CRUSH hierarchy may help you troubleshoot the storage cluster by identifying the physical locations faster.

If an OSD is down, connect to the node and start it. You can use Red Hat Storage Console to restart the OSD node, or you can use the command line, for example:

# systemctl start ceph-osd@<osd_id>

3.2. Low-level Monitoring

Lower-level monitoring typically involves ensuring that OSDs are peering. When faults occur, placement groups operate in a degraded state. This can be due to many things such as failed hardware, hung or crashed daemon, network latency or outage among other things.

3.2.1. Placement Group Sets

When CRUSH assigns placement groups to OSDs, it looks at the number of replicas for the pool and assigns the placement group to OSDs such that each replica of the placement group gets assigned to a different OSD. For example, if the pool requires three replicas of a placement group, CRUSH may assign them to osd.1, osd.2 and osd.3 respectively. CRUSH actually seeks a pseudo-random placement that will take into account failure domains you set in the CRUSH map, so you will rarely see placement groups assigned to nearest neighbor OSDs in a large cluster. We refer to the set of OSDs that should contain the replicas of a particular placement group as the Acting Set. In some cases, an OSD in the Acting Set is down or otherwise not able to service requests for objects in the placement group. When these situations arise, don’t panic. Common examples include:

  • You added or removed an OSD. Then, CRUSH reassigned the placement group to other OSDs—​thereby changing the composition of the Acting Set and spawning the migration of data with a "backfill" process.
  • An OSD was down, was restarted and is now recovering.
  • An OSD in the Acting Set is down or unable to service requests, and another OSD has temporarily assumed its duties.

Ceph processes a client request using the Up Set, which is the set of OSDs that will actually handle the requests. In most cases, the Up Set and the Acting Set are virtually identical. When they are not, it may indicate that Ceph is migrating data, an OSD is recovering, or that there is a problem, that is, Ceph usually echoes a HEALTH WARN state with a "stuck stale" message in such scenarios.

  • To retrieve a list of placement groups:

    # ceph pg dump
  • To view which OSDs are in the Acting Set or in the Up Set for a given placement group:

    # ceph pg map <pg-num>

    The result should tell you the osdmap epoch, eNNN, the placement group number, <pg-num>, the OSDs in the Up Set up[], and the OSDs in the acting set, acting[]:

    osdmap eNNN pg <pg-num> -> up [0,1,2] acting [0,1,2]
    Note

    If the Up Set and Acting Set do not match, this may be an indicator that the cluster rebalancing itself or of a potential problem with the cluster.

3.2.2. Peering

Before you can write data to a placement group, it must be in an active state, and it should be in a clean state. For Ceph to determine the current state of a placement group, the primary OSD of the placement group (i.e., the first OSD in the acting set), peers with the secondary and tertiary OSDs to establish agreement on the current state of the placement group (assuming a pool with 3 replicas of the PG).

Peering

3.2.3. Monitoring Placement Group States

If you execute a command such as ceph health, ceph -s or ceph -w, you may notice that the cluster does not always echo back HEALTH OK. After you check to see if the OSDs are running, you should also check placement group states. You should expect that the cluster will NOT echo HEALTH OK in a number of placement group peering-related circumstances:

  • You have just created a pool and placement groups haven’t peered yet.
  • The placement groups are recovering.
  • You have just added an OSD to or removed an OSD from the cluster.
  • You have just modified the CRUSH map and the placement groups are migrating.
  • There is inconsistent data in different replicas of a placement group.
  • Ceph is scrubbing a placement group’s replicas.
  • Ceph doesn’t have enough storage capacity to complete backfilling operations.

If one of the foregoing circumstances causes Ceph to echo HEALTH WARN, don’t panic. In many cases, the cluster will recover on its own. In some cases, you may need to take action. An important aspect of monitoring placement groups is to ensure that when the cluster is up and running that all placement groups are active, and preferably in the clean state. To see the status of all placement groups, execute:

# ceph pg stat

The result should tell you the placement group map version, vNNNNNN, the total number of placement groups, x, and how many placement groups, y, are in a particular state such as active+clean:

vNNNNNN: x pgs: y active+clean; z bytes data, aa MB used, bb GB / cc GB avail
Note

It is common for Ceph to report multiple states for placement groups.

Snapshot Trimming PG States

When snapshots exist, two additional PG states will be reported.

  • snaptrim : The PGs are currently being trimmed
  • snaptrim_wait : The PGs are waiting to be trimmed

Example Output:

244 active+clean+snaptrim_wait
 32 active+clean+snaptrim
Note

See the miscellaneous OSD settings in the Red Hat Ceph Storage 3 Configuration Guide for more details on the snapshot trimming settings.

In addition to the placement group states, Ceph will also echo back the amount of data used, aa, the amount of storage capacity remaining, bb, and the total storage capacity for the placement group. These numbers can be important in a few cases:

  • You are reaching the near full ratio or full ratio.
  • Your data isn’t getting distributed across the cluster due to an error in the CRUSH configuration.

Placement Group IDs

Placement group IDs consist of the pool number, and not the pool name, followed by a period (.) and the placement group ID—​a hexadecimal number. You can view pool numbers and their names from the output of ceph osd lspools. The default pool names data, metadata and rbd correspond to pool numbers 0, 1 and 2 respectively. A fully qualified placement group ID has the following form:

<pool_num>.<pg_id>

Example output:

0.1f
  • To retrieve a list of placement groups:

    # ceph pg dump
  • To format the output in JSON format and save it to a file:

    # ceph pg dump -o <file_name> --format=json
  • To query a particular placement group:

    # ceph pg <pool_num>.<pg_id> query

    Example output in JSON format:

    {
      "state": "active+clean",
      "up": [
        1,
        0
      ],
      "acting": [
        1,
        0
      ],
      "info": {
        "pgid": "1.e",
        "last_update": "4'1",
        "last_complete": "4'1",
        "log_tail": "0'0",
        "last_backfill": "MAX",
        "purged_snaps": "[]",
        "history": {
          "epoch_created": 1,
          "last_epoch_started": 537,
          "last_epoch_clean": 537,
          "last_epoch_split": 534,
          "same_up_since": 536,
          "same_interval_since": 536,
          "same_primary_since": 536,
          "last_scrub": "4'1",
          "last_scrub_stamp": "2013-01-25 10:12:23.828174"
        },
        "stats": {
          "version": "4'1",
          "reported": "536'782",
          "state": "active+clean",
          "last_fresh": "2013-01-25 10:12:23.828271",
          "last_change": "2013-01-25 10:12:23.828271",
          "last_active": "2013-01-25 10:12:23.828271",
          "last_clean": "2013-01-25 10:12:23.828271",
          "last_unstale": "2013-01-25 10:12:23.828271",
          "mapping_epoch": 535,
          "log_start": "0'0",
          "ondisk_log_start": "0'0",
          "created": 1,
          "last_epoch_clean": 1,
          "parent": "0.0",
          "parent_split_bits": 0,
          "last_scrub": "4'1",
          "last_scrub_stamp": "2013-01-25 10:12:23.828174",
          "log_size": 128,
          "ondisk_log_size": 128,
          "stat_sum": {
            "num_bytes": 205,
            "num_objects": 1,
            "num_object_clones": 0,
            "num_object_copies": 0,
            "num_objects_missing_on_primary": 0,
            "num_objects_degraded": 0,
            "num_objects_unfound": 0,
            "num_read": 1,
            "num_read_kb": 0,
            "num_write": 3,
            "num_write_kb": 1
          },
          "stat_cat_sum": {
    
          },
          "up": [
            1,
            0
          ],
          "acting": [
            1,
            0
          ]
        },
        "empty": 0,
        "dne": 0,
        "incomplete": 0
      },
      "recovery_state": [
        {
          "name": "Started\/Primary\/Active",
          "enter_time": "2013-01-23 09:35:37.594691",
          "might_have_unfound": [
    
          ],
          "scrub": {
            "scrub_epoch_start": "536",
            "scrub_active": 0,
            "scrub_block_writes": 0,
            "finalizing_scrub": 0,
            "scrub_waiting_on": 0,
            "scrub_waiting_on_whom": [
    
            ]
          }
        },
        {
          "name": "Started",
          "enter_time": "2013-01-23 09:35:31.581160"
        }
      ]
    }

The following subsections describe common states in greater detail.

3.2.3.1. Creating

When you create a pool, it will create the number of placement groups you specified. Ceph will echo creating when it is creating one or more placement groups. Once they are created, the OSDs that are part of a placement group’s Acting Set will peer. Once peering is complete, the placement group status should be active+clean, which means a Ceph client can begin writing to the placement group.

Creating PGs

3.2.3.2. Peering

When Ceph is Peering a placement group, Ceph is bringing the OSDs that store the replicas of the placement group into agreement about the state of the objects and metadata in the placement group. When Ceph completes peering, this means that the OSDs that store the placement group agree about the current state of the placement group. However, completion of the peering process does NOT mean that each replica has the latest contents.

Authoritative History

Ceph will NOT acknowledge a write operation to a client, until all OSDs of the acting set persist the write operation. This practice ensures that at least one member of the acting set will have a record of every acknowledged write operation since the last successful peering operation.

With an accurate record of each acknowledged write operation, Ceph can construct and disseminate a new authoritative history of the placement group—​a complete, and fully ordered set of operations that, if performed, would bring an OSD’s copy of a placement group up to date.

3.2.3.3. Active

Once Ceph completes the peering process, a placement group may become active. The active state means that the data in the placement group is generally available in the primary placement group and the replicas for read and write operations.

3.2.3.4. Clean

When a placement group is in the clean state, the primary OSD and the replica OSDs have successfully peered and there are no stray replicas for the placement group. Ceph replicated all objects in the placement group the correct number of times.

3.2.3.5. Degraded

When a client writes an object to the primary OSD, the primary OSD is responsible for writing the replicas to the replica OSDs. After the primary OSD writes the object to storage, the placement group will remain in a degraded state until the primary OSD has received an acknowledgement from the replica OSDs that Ceph created the replica objects successfully.

The reason a placement group can be active+degraded is that an OSD may be active even though it doesn’t hold all of the objects yet. If an OSD goes down, Ceph marks each placement group assigned to the OSD as degraded. The OSDs must peer again when the OSD comes back online. However, a client can still write a new object to a degraded placement group if it is active.

If an OSD is down and the degraded condition persists, Ceph may mark the down OSD as out of the cluster and remap the data from the down OSD to another OSD. The time between being marked down and being marked out is controlled by mon_osd_down_out_interval, which is set to 600 seconds by default.

A placement group can also be degraded, because Ceph cannot find one or more objects that Ceph thinks should be in the placement group. While you cannot read or write to unfound objects, you can still access all of the other objects in the degraded placement group.

Let’s say there are 9 OSDs with three copies of an object. If OSD number 9 goes down, the PGs assigned to OSD 9 go in a degraded state. If OSD 9 doesn’t recover, it goes out of the cluster and the cluster rebalances. In that scenario, the PGs are degraded and then recover to an active state.

3.2.3.6. Recovering

Ceph was designed for fault-tolerance at a scale where hardware and software problems are ongoing. When an OSD goes down, its contents may fall behind the current state of other replicas in the placement groups. When the OSD is back up, the contents of the placement groups must be updated to reflect the current state. During that time period, the OSD may reflect a recovering state.

Recovery isn’t always trivial, because a hardware failure might cause a cascading failure of multiple OSDs. For example, a network switch for a rack or cabinet may fail, which can cause the OSDs of a number of host machines to fall behind the current state of the cluster. Each one of the OSDs must recover once the fault is resolved.

Ceph provides a number of settings to balance the resource contention between new service requests and the need to recover data objects and restore the placement groups to the current state. The osd recovery delay start setting allows an OSD to restart, re-peer and even process some replay requests before starting the recovery process. The osd recovery threads setting limits the number of threads for the recovery process, by default one thread. The osd recovery thread timeout sets a thread timeout, because multiple OSDs may fail, restart and re-peer at staggered rates. The osd recovery max active setting limits the number of recovery requests an OSD will entertain simultaneously to prevent the OSD from failing to serve . The osd recovery max chunk setting limits the size of the recovered data chunks to prevent network congestion.

3.2.3.7. Backfilling

When a new OSD joins the cluster, CRUSH will reassign placement groups from OSDs in the cluster to the newly added OSD. Forcing the new OSD to accept the reassigned placement groups immediately can put excessive load on the new OSD. Backfilling the OSD with the placement groups allows this process to begin in the background. Once backfilling is complete, the new OSD will begin serving requests when it is ready.

During the backfill operations, you may see one of several states: backfill_wait indicates that a backfill operation is pending, but isn’t underway yet; backfill indicates that a backfill operation is underway; and, backfill_too_full indicates that a backfill operation was requested, but couldn’t be completed due to insufficient storage capacity. When a placement group cannot be backfilled, it may be considered incomplete.

Ceph provides a number of settings to manage the load spike associated with reassigning placement groups to an OSD, especially a new OSD. By default, osd_max_backfills sets the maximum number of concurrent backfills to or from an OSD to 10. The osd backfill full ratio enables an OSD to refuse a backfill request if the OSD is approaching its full ratio, by default 85%. If an OSD refuses a backfill request, the osd backfill retry interval enables an OSD to retry the request, by default after 10 seconds. OSDs can also set osd backfill scan min and osd backfill scan max to manage scan intervals, by default 64 and 512.

For some workloads, it is beneficial to avoid regular recovery entirely and use backfill instead. Since backfilling occurs in the background, this allows I/O to proceed on the objects in the OSD. To force backfill rather than recovery, set osd_min_pg_log_entries to 1, and set osd_max_pg_log_entries to 2. Contact your Red Hat Support account team for details on when this situation is appropriate for your workload.

3.2.3.8. Forcing the priority of recovery or backfill operations

You might encounter a situation where some placement groups (PGs) require recovery and/or backfill, and some of those placement groups contain more important data than do others. Use the pg force-recovery or pg force-backfill command to ensure that the PGs with the higher-priority data undergo recovery or backfill first.

Prerequisites

  • A running Red Hat Ceph Storage cluster.
  • Root-level access to the node.

Procedure

  1. Issue the pg force-recovery or pg force-backfill command and specify the order of priority for the PGs with the higher-priority data:

    Syntax

    # ceph pg force-recovery PG1 [PG2] [PG3 ...]
    # ceph pg force-backfill PG1 [PG2] [PG3 ...]

    Example

    [root@node]# ceph pg force-recovery group1 group2
    [root@node]# ceph pg force-backfill group1 group2

    This command causes Red Hat Ceph Storage to perform recovery or backfill on specified placement groups (PGs) first, before processing other placement groups. Issuing the command does not interrupt backfill or recovery operations that are currently executing. After the currently running operations have finished, recovery or backfill takes place as soon as possible for the specified PGs.

3.2.3.9. Changing or canceling the priority of recovery or backfill operations

If you cancel a high-priority force-recovery or force-backfill operation on certain placement groups (PGs) in a storage cluster, operations for those PGs revert to the default recovery or backfill settings.

Prerequisites

  • A running Red Hat Ceph Storage cluster.
  • Root-level access to the node.

Procedure

  1. To change or cancel a recovery or backfill operation on specified placement groups:

    Syntax

    ceph pg cancel-force-recovery PG1 [PG2] [PG3 ...]
    ceph pg cancel-force-backfill PG1 [PG2] [PG3 ...]

    Example

    [root@node]# ceph pg cancel-force-recovery group1 group2
    [root@node]# ceph pg cancel-force-backfill group1 group2

    This cancels the force flag and processes the PGs in the default order.

    After recovery or backfill operations for the specified PGs have completed, processing order reverts to the default.

3.2.3.10. Forcing high-priority recovery or backfill operations for pools

If all of the placement groups in a pool require high-priority recovery or backfill, use the force-recovery or force-backfill options to initiate the operation.

Prerequisites

  • A running Red Hat Ceph Storage cluster.
  • Root-level access to the node.

Procedure

  1. To force high-priority recovery or backfill on all placement groups in a specified pool:

    Syntax

    ceph osd pool force-recovery POOL_NAME
    ceph osd pool force-backfill POOL_NAME

    Example

    [root@node]# ceph osd pool force-recovery pool1
    [root@node]# ceph osd pool force-backfill pool1

    Note

    Use the force-recovery and force-backfill commands with caution. Changing the priority of these operations might break the ordering of Ceph’s internal priority computations.

3.2.3.11. Canceling high-priority recovery or backfill operations for pools

If you cancel a high-priority force-recovery or force-backfill operation on all placement groups in a pool, operations for the PGs in that pool revert to the default recovery or backfill settings.

Prerequisites

  • A running Red Hat Ceph Storage cluster.
  • Root-level access to the node.

Procedure

  1. To cancel a high-priority recovery or backfill operation on all placement groups in a specified pool:

    Syntax

    ceph osd pool cancel-force-recovery POOL_NAME
    ceph osd pool cancel-force-backfill POOL_NAME

    Example

    [root@node]# ceph osd pool cancel-force-recovery pool1
    [root@node]# ceph osd pool cancel-force-backfill pool1

3.2.3.12. Rearranging the priority of recovery and backfill operations for pools

If you have multiple pools that currently use the same underlying OSDs and some of the pools contain high-priority data, you can rearrange the order in which the operations execute. Use the recovery_priority option to assign a higher priority value to the pools with the higher-priority data. Those pools will execute before pools with lower priority values, or pools that are set to default priority.

Prerequisites

  • A running Red Hat Ceph Storage cluster.
  • Root-level access to the node.

Procedure

  1. To rearrange the recovery/backfill priority for the pools:

    Syntax

    ceph osd pool set POOL_NAME recovery_priority VALUE

    VALUE sets the order of priority. For example, if you have 10 pools, the pool with a priority value of 10 gets processed first, followed by the pool with priority 9, and so on. If only some pools have high priority, you can set priority values for just those pools. The pools without set priority values are processed in the default order.

    Example

    ceph osd pool set pool1 recovery_priority 10

3.2.3.13. Priority of placement group recovery in RADOS

This section describes the relative priority values for the recovery and backfilling of placement groups (PGs) in RADOS. Higher values are processed first. Inactive PGs receive higher priority values than active or degraded PGs.

OperationValueDescription

OSD_RECOVERY_PRIORITY_MIN

0

Minimum recovery value

OSD_BACKFILL_PRIORITY_BASE

100

Base backfill priority for MBackfillReserve

OSD_BACKFILL_DEGRADED_PRIORITY_BASE

140

Base backfill priority for MBackfillReserve (degraded PG)

OSD_RECOVERY_PRIORITY_BASE

180

Base recovery priority for MBackfillReserve

OSD_BACKFILL_INACTIVE_PRIORITY_BASE

220

Base backfill priority for MBackfillReserve (inactive PG)

OSD_RECOVERY_INACTIVE_PRIORITY_BASE

220

Base recovery priority for MRecoveryReserve (inactive PG)

OSD_RECOVERY_PRIORITY_MAX

253

Max manually/automatically set recovery priority for MBackfillReserve

OSD_BACKFILL_PRIORITY_FORCED

254

Backfill priority for MBackfillReserve, when forced manually

OSD_RECOVERY_PRIORITY_FORCED

255

Recovery priority for MRecoveryReserve, when forced manually

OSD_DELETE_PRIORITY_NORMAL

179

Priority for PG deletion when the OSD is not fullish

OSD_DELETE_PRIORITY_FULLISH

219

Priority for PG deletion when the OSD is approaching full

OSD_DELETE_PRIORITY_FULL

255

Priority for deletion when the OSD is full

3.2.3.14. Remapped

When the Acting Set that services a placement group changes, the data migrates from the old acting set to the new acting set. It may take some time for a new primary OSD to service requests. So it may ask the old primary to continue to service requests until the placement group migration is complete. Once data migration completes, the mapping uses the primary OSD of the new acting set.

3.2.3.15. Stale

While Ceph uses heartbeats to ensure that hosts and daemons are running, the ceph-osd daemons may also get into a stuck state where they aren’t reporting statistics in a timely manner, for example, a temporary network fault. By default, OSD daemons report their placement group, up thru, boot and failure statistics every half second, that is, 0.5, which is more frequent than the heartbeat thresholds. If the Primary OSD of a placement group’s acting set fails to report to the monitor or if other OSDs have reported the primary OSD down, the monitors will mark the placement group stale.

When you start the storage cluster, it is common to see the stale state until the peering process completes. After the storage cluster has been running for awhile, seeing placement groups in the stale state indicates that the primary OSD for those placement groups is down or not reporting placement group statistics to the monitor.

3.2.3.16. Misplaced

There are some temporary backfilling scenarios where a PG gets mapped temporarily to an OSD. When that temporary situation should no longer be the case, the PGs might still reside in the temporary location and not in the proper location. In which case, they are said to be misplaced. That’s because the correct number of extra copies actually exist, but one or more copies is in the wrong place.

Lets say there are 3 OSDs: 0,1,2 and all PGs map to some permutation of those three. If you add another OSD (OSD 3), some PGs will now map to OSD 3 instead of one of the others. However, until OSD 3 is backfilled, the PG will have a temporary mapping allowing it to continue to serve I/O from the old mapping. During that time, the PG is misplaced, because it has a temporary mapping, but not degraded, since there are 3 copies.

Example

pg 1.5: up=acting: [0,1,2]
<add osd 3>
pg 1.5: up: [0,3,1] acting: [0,1,2]

Here, [0,1,2] is a temporary mapping, so the up set is not equal to the acting set and the PG is misplaced but not degraded since [0,1,2] is still three copies.

Example

pg 1.5: up=acting: [0,3,1]

OSD 3 is now backfilled and the temporary mapping is removed, not degraded and not misplaced.

3.2.3.17. Incomplete

A PG goes into a incomplete state when there is incomplete content and peering fails, that is, when there are no complete OSDs which are current enough to perform recovery.

Lets say OSD 1, 2, and 3 are the acting OSD set and it switches to OSD 1, 4, and 3, then osd.1 will request a temporary acting set of OSD 1, 2, and 3 while backfilling 4. During this time, if OSD 1, 2, and 3 all go down, osd.4 will be the only one left which might not have fully backfilled all the data. At this time, the PG will go incomplete indicating that there are no complete OSDs which are current enough to perform recovery.

Alternately, if osd.4 is not involved and the acting set is simply OSD 1, 2, and 3 when OSD 1, 2, and 3 go down, the PG would likely go stale indicating that the mons have not heard anything on that PG since the acting set changed. The reason being there are no OSDs left to notify the new OSDs.

3.2.4. Identifying Troubled Placement Groups

As previously noted, a placement group isn’t necessarily problematic just because its state isn’t active+clean. Generally, Ceph’s ability to self repair may not be working when placement groups get stuck. The stuck states include:

  • Unclean: Placement groups contain objects that are not replicated the desired number of times. They should be recovering.
  • Inactive: Placement groups cannot process reads or writes because they are waiting for an OSD with the most up-to-date data to come back up.
  • Stale: Placement groups are in an unknown state, because the OSDs that host them have not reported to the monitor cluster in a while, and can be configured with the mon osd report timeout setting.

To identify stuck placement groups, execute the following:

# ceph pg dump_stuck {inactive|unclean|stale|undersized|degraded [inactive|unclean|stale|undersized|degraded...]} {<int>}

3.2.5. Finding an Object Location

To store object data in the Ceph Object Store, a Ceph client must:

  1. Set an object name
  2. Specify a pool

The Ceph client retrieves the latest cluster map and the CRUSH algorithm calculates how to map the object to a placement group, and then calculates how to assign the placement group to an OSD dynamically. To find the object location, all you need is the object name and the pool name. For example:

# ceph osd map <pool_name> <object_name>

3.3. Monitoring a Ceph storage cluster with the Red Hat Ceph Storage Dashboard

The Red Hat Ceph Storage Dashboard provides a monitoring dashboard to visualize the state of a Ceph Storage Cluster. Also, the Red Hat Ceph Storage Dashboard architecture provides a framework for additional modules to add functionality to the storage cluster.

Prerequisites

  • A running Red Hat Ceph Storage cluster

3.3.1. The Red Hat Ceph Storage Dashboard

The Red Hat Ceph Storage Dashboard provides a monitoring dashboard for Ceph clusters to visualize the storage cluster state. The dashboard is accessible from a web browser and provides a number of metrics and graphs about the state of the cluster, Monitors, OSDs, Pools, or the network.

With the previous releases of Red Hat Ceph Storage, monitoring data was sourced through a collectd plugin, which sent the data to an instance of the Graphite monitoring utility. Starting with Red Hat Ceph Storage 3.3, monitoring data is sourced directly from the ceph-mgr daemon, using the ceph-mgr Prometheus plugin.

The introduction of Prometheus as the monitoring data source simplifies deployment and operational management of the Red Hat Ceph Storage Dashboard solution, along with reducing the overall hardware requirements. By sourcing the Ceph monitoring data directly, the Red Hat Ceph Storage Dashboard solution is better able to support Ceph clusters deployed in containers.

Note

With this change in architecture, there is no migration path for monitoring data from Red Hat Ceph Storage 2.x and 3.0 to Red Hat Ceph Storage 3.3.

The Red Hat Ceph Storage Dashboard uses the following utilities:

  • The Ansible automation application for deployment.
  • The embedded Prometheus ceph-mgr plugin.
  • The Prometheus node-exporter daemon, running on each node of the storage cluster.
  • The Grafana platform to provide a user interface and alerting.

The Red Hat Ceph Storage Dashboard supports the following features:

General Features
  • Support for Red Hat Ceph Storage 3.1 and higher
  • SELinux support
  • Support for FileStore and BlueStore OSD back ends
  • Support for encrypted and non-encrypted OSDs
  • Support for Monitor, OSD, the Ceph Object Gateway, and iSCSI roles
  • Initial support for the Metadata Servers (MDS)
  • Drill down and dashboard links
  • 15 second granularity
  • Support for Hard Disk Drives (HDD), Solid-state Drives (SSD), Non-volatile Memory Express (NVMe) interface, and Intel® Cache Acceleration Software (Intel® CAS)
Node Metrics
  • CPU and RAM usage
  • Network load
Configurable Alerts
  • Out-of-Band (OOB) alerts and triggers
  • Notification channel is automatically defined during the installation
  • The Ceph Health Summary dashboard created by default

    See the Red Hat Ceph Storage Dashboard Alerts section for details.

Cluster Summary
  • OSD configuration summary
  • OSD FileStore and BlueStore summary
  • Cluster versions breakdown by role
  • Disk size summary
  • Host size by capacity and disk count
  • Placement Groups (PGs) status breakdown
  • Pool counts
  • Device class summary, HDD vs. SSD
Cluster Details
  • Cluster flags status (noout, nodown, and others)
  • OSD or Ceph Object Gateway hosts up and down status
  • Per pool capacity usage
  • Raw capacity utilization
  • Indicators for active scrub and recovery processes
  • Growth tracking and forecast (raw capacity)
  • Information about OSDs that are down or near full, including the OSD host and disk
  • Distribution of PGs per OSD
  • OSDs by PG counts, highlighting the over or under utilized OSDs
OSD Performance
  • Information about I/O operations per second (IOPS) and throughput by pool
  • OSD performance indicators
  • Disk statistics per OSD
  • Cluster wide disk throughput
  • Read/write ratio (client IOPS)
  • Disk utilization heat map
  • Network load by Ceph role
The Ceph Object Gateway Details
  • Aggregated load view
  • Per host latency and throughput
  • Workload breakdown by HTTP operations
The Ceph iSCSI Gateway Details
  • Aggregated views
  • Configuration
  • Performance
  • Per Gateway resource utilization
  • Per client load and configuration
  • Per Ceph Block Device image performance

3.3.2. Installing the Red Hat Ceph Storage Dashboard

The Red Hat Ceph Storage Dashboard provides a visual dashboard to monitor various metrics in a running Ceph Storage Cluster.

Note

For information on upgrading the Red Hat Ceph Storage Dashboard see Upgrading Red Hat Ceph Storage Dashboard in the Installation Guide for Red Hat Enterprise Linux.

Prerequisites

  • The storage cluster nodes use Red Hat Enterprise Linux 7.
  • A separate node, the Red Hat Ceph Storage Dashboard node, for receiving data from the cluster nodes and providing the Red Hat Ceph Storage Dashboard.
  • Prepare the Red Hat Ceph Storage Dashboard node:

    • Enable the Tools repository on all nodes.

      For details, see the Enabling the Red Hat Ceph Storage Repositories section in the Red Hat Ceph Storage 3 Installation Guide for Red Hat Enterprise Linux.

    • If using a firewall, then ensure that the following TCP ports are open:

      Table 3.1. TCP Port Requirements
      PortUseWhere?

      3000

      Grafana

      The Red Hat Ceph Storage Dashboard node.

      9090

      Basic Prometheus graphs

      The Red Hat Ceph Storage Dashboard node.

      9100

      Prometheus' node-exporter daemon

      All storage cluster nodes.

      9283

      Gathering Ceph data

      All ceph-mgr nodes.

      9287

      Ceph iSCSI gateway data

      All Ceph iSCSI gateway nodes.

      For more details see the Using Firewalls chapter in the Security Guide for Red Hat Enterprise Linux 7.

Procedure

Run the following commands on the Ansible administration node as the root user.

  1. Install the cephmetrics-ansible package.

    [root@admin ~]# yum install cephmetrics-ansible
  2. Using the Ceph Ansible inventory as a base, add the Red Hat Ceph Storage Dashboard node under the [ceph-grafana] section of the Ansible inventory file, by default located at /etc/ansible/hosts.

    [ceph-grafana]
    $HOST_NAME

    Replace:

    • $HOST_NAME with the name of the Red Hat Ceph Storage Dashboard node

    For example:

    [ceph-grafana]
    node0
  3. Change to the /usr/share/cephmetrics-ansible/ directory.

    [root@admin ~]# cd /usr/share/cephmetrics-ansible
  4. Run the Ansible playbook.

    [root@admin cephmetrics-ansible]# ansible-playbook -v playbook.yml
    Important

    Every time you update the cluster configuration, for example, you add or remove a MON or OSD node, you must re-run the cephmetrics Ansible playbook.

    Note

    The cephmetrics Ansible playbook does the following actions:

    • Updates the ceph-mgr instance to enable the prometheus plugin and opens TCP port 9283.
    • Deploys the Prometheus node-exporter daemon to each node in the storage cluster.

      • Opens TCP port 9100.
      • Starts the node-exporter daemon.
    • Deploys Grafana and Prometheus containers under Docker/systemd on the Red Hat Ceph Storage Dashboard node.

      • Prometheus is configured to gather data from the ceph-mgr nodes and the node-exporters running on each ceph host
      • Opens TCP port 3000.
      • The dashboards, themes and user accounts are all created in Grafana.
      • Outputs the URL of Grafana for the administrator.

3.3.3. Accessing the Red Hat Ceph Storage Dashboard

Accessing the Red Hat Ceph Storage Dashboard gives you access to the web-based management tool for administrating Red Hat Ceph Storage clusters.

Prerequisites
Procedure
  1. Enter the following URL to a web browser:

    http://$HOST_NAME:3000

    Replace:

    • $HOST_NAME with the name of the Red Hat Ceph Storage Dashboard node

    For example:

    http://cephmetrics:3000
  2. Enter the password for the admin user. If you did not set the password during the installation, use admin, which is the default password.

    Once logged in, you are automatically placed on the Ceph At a Glance dashboard. The Ceph At a Glance dashboard provides a high-level overviews of capacity, performance, and node-level performance information.

    Example

    RHCS Dashboard Grafana Ceph At a Glance page

Additional Resources

3.3.4. Changing the default Red Hat Ceph Storage dashboard password

The default user name and password for accessing the Red Hat Ceph Storage Dashboard is set to admin and admin. For security reasons, you might want to change the password after the installation.

Note

To prevent the password from resetting to the default value, update the custom password in the /usr/share/cephmetrics-ansible/group_vars/all.yml file.

Procedure

  1. Click the Grafana icon in the upper-left corner.
  2. Hover over the user name you want to modify the password for. In this case admin.
  3. Click Profile.
  4. Click Change Password.
  5. Enter the new password twice and click Change Password.

Additional Resource

3.3.5. The Prometheus plugin for Red Hat Ceph Storage

As a storage administrator, you can gather performance data, export that data using the Prometheus plugin module for the Red Hat Ceph Storage Dashboard, and then perform queries on this data. The Prometheus module allows ceph-mgr to expose Ceph related state and performance data to a Prometheus server.

3.3.5.1. Prerequisites

  • Running Red Hat Ceph Storage 3.1 or higher.
  • Installation of the Red Hat Ceph Storage Dashboard.

3.3.5.2. The Prometheus plugin

The Prometheus plugin provides an exporter to pass on Ceph performance counters from the collection point in ceph-mgr. The Red Hat Ceph Storage Dashboard receives MMgrReport messages from all MgrClient processes, such as Ceph Monitors and OSDs. A circular buffer of the last number of samples contains the performance counter schema data and the actual counter data. This plugin creates an HTTP endpoint and retrieves the latest sample of every counter when polled. The HTTP path and query parameters are ignored; all extant counters for all reporting entities are returned in a text exposition format.

Additional Resources

3.3.5.3. Managing the Prometheus environment

To monitor a Ceph storage cluster with Prometheus you can configure and enable the Prometheus exporter so the metadata information about the Ceph storage cluster can be collected.

Prerequisites

  • A running Red Hat Ceph Storage 3.1 cluster
  • Installation of the Red Hat Ceph Storage Dashboard

Procedure

  1. As the root user, open and edit the /etc/prometheus/prometheus.yml file.

    1. Under the global section, set the scrape_interval and evaluation_interval options to 15 seconds.

      Example

      global:
        scrape_interval:     15s
        evaluation_interval: 15s

    2. Under the scrape_configs section, add the honor_labels: true option, and edit the targets, and instance options for each of the ceph-mgr nodes.

      Example

      scrape_configs:
        - job_name: 'node'
          honor_labels: true
          static_configs:
          - targets: [ 'node1.example.com:9100' ]
            labels:
              instance: "node1.example.com"
          - targets: ['node2.example.com:9100']
            labels:
              instance: "node2.example.com"

      Note

      Using the honor_labels option enables Ceph to output properly-labelled data relating to any node in the Ceph storage cluster. This allows Ceph to export the proper instance label without Prometheus overwriting it.

    3. To add a new node, simply add the targets, and instance options in the following format:

      Example

      - targets: [ 'new-node.example.com:9100' ]
        labels:
          instance: "new-node"

      Note

      The instance label has to match what appears in Ceph’s OSD metadata instance field, which is the short host name of the node. This helps to correlate Ceph stats with the node’s stats.

  2. Add Ceph targets to the /etc/prometheus/ceph_targets.yml file in the following format.

    Example

    [
        {
            "targets": [ "cephnode1.example.com:9283" ],
            "labels": {}
        }
    ]

  3. Enable the Prometheus module:

    # ceph mgr module enable prometheus

3.3.5.4. Using the Prometheus expression browser

Use the builtin Prometheus expression browser to run queries against the collected data.

Prerequisites

  • A running Red Hat Ceph Storage 3.1 cluster
  • Installation of the Red Hat Ceph Storage Dashboard

Procedure

  1. Enter the URL for the Prometh the web browser:

    http://$DASHBOARD_SEVER_NAME:9090/graph

    Replace…​

    • $DASHBOARD_SEVER_NAME with the name of the Red Hat Ceph Storage Dashboard server.
  2. Click on Graph, then type in or paste the query into the query window and press the Execute button.

    1. View the results in the console window.
  3. Click on Graph to view the rendered data.

Additional Resources

3.3.5.5. Working with the Prometheus data and queries

The statistic names are exactly as Ceph names them, with illegal characters translated to underscores, and ceph_ prefixed to all names. All Ceph daemon statistics have a ceph_daemon label that identifies the type and ID of the daemon they come from, for example: osd.123. Some statistics can come from different types of daemons, so when querying you will want to filter on Ceph daemons starting with osd to avoid mixing in the Ceph Monitor and RocksDB stats. The global Ceph storage cluster statistics have labels appropriate to what they report on. For example, metrics relating to pools have a pool_id label. The long running averages that represent the histograms from core Ceph are represented by a pair of sum and count performance metrics.

The following example queries can be used in the Prometheus expression browser:

Show the physical disk utilization of an OSD

(irate(node_disk_io_time_ms[1m]) /10) and on(device,instance) ceph_disk_occupation{ceph_daemon="osd.1"}

Show the physical IOPS of an OSD as seen from the operating system

irate(node_disk_reads_completed[1m]) + irate(node_disk_writes_completed[1m]) and on (device, instance) ceph_disk_occupation{ceph_daemon="osd.1"}

Pool and OSD metadata series

Special data series are output to enable the displaying and the querying on certain metadata fields. Pools have a ceph_pool_metadata field, for example:

ceph_pool_metadata{pool_id="2",name="cephfs_metadata_a"} 1.0

OSDs have a ceph_osd_metadata field, for example:

ceph_osd_metadata{cluster_addr="172.21.9.34:6802/19096",device_class="ssd",ceph_daemon="osd.0",public_addr="172.21.9.34:6801/19096",weight="1.0"} 1.0

Correlating drive statistics with node_exporter

The Prometheus output from Ceph is designed to be used in conjunction with the generic node monitoring from the Prometheus node exporter. Correlation of Ceph OSD statistics with the generic node monitoring drive statistics, special data series are output, for example:

ceph_disk_occupation{ceph_daemon="osd.0",device="sdd", exported_instance="node1"}

To get disk statistics by an OSD ID, use either the and operator or the asterisk (*) operator in the Prometheus query. All metadata metrics have the value of 1 so they act neutral with asterisk operator. Using asterisk operator allows to use group_left and group_right grouping modifiers, so that the resulting metric has additional labels from one side of the query. For example:

rate(node_disk_bytes_written[30s]) and on (device,instance) ceph_disk_occupation{ceph_daemon="osd.0"}

Using label_replace

The label_replace function can add a label to, or alter a label of, a metric within a query. To correlate an OSD and its disks write rate, the following query can be used:

label_replace(rate(node_disk_bytes_written[30s]), "exported_instance", "$1", "instance", "(.*):.*") and on (device,exported_instance) ceph_disk_occupation{ceph_daemon="osd.0"}

Additional Resources

  • See Prometheus querying basics for more information on constructing queries.
  • See Prometheus' label_replace documentation for more information.

3.3.5.6. Additional Resources

3.3.6. The Red Hat Ceph Storage Dashboard alerts

This section includes information about alerting in the Red Hat Ceph Storage Dashboard.

3.3.6.1. Prerequisites

3.3.6.2. About Alerts

The Red Hat Ceph Storage Dashboard supports alerting mechanism that is provided by the Grafana platform. You can configure the dashboard to send you a notification when a metric that you are interested in reaches certain value. Such metrics are in the Alert Status dashboard.

By default, Alert Status already includes certain metrics, such as Overall Ceph Health, OSDs Down, or Pool Capacity. You can add metrics that you are interested in to this dashboard or change their trigger values.

Here is a list of the pre-defined alerts that are included with Red Hat Ceph Storage Dashboard:

  • Overall Ceph Health
  • Disks Near Full (>85%)
  • OSD Down
  • OSD Host Down
  • PG’s Stuck Inactive
  • OSD Host Less - Free Capacity Check
  • OSD’s With High Response Times
  • Network Errors
  • Pool Capacity High
  • Monitors Down
  • Overall Cluster Capacity Low
  • OSDs With High PG Count

3.3.6.3. Accessing the Alert Status dashboard

Certain Red Hat Ceph Storage Dashboard alerts are configured by default in the Alert Status dashboard. This section shows two ways to access it.

Procedure

To access the dashboard:

  • In the main At the Glance dashboard, click the Active Alerts panel in the upper-right corner.

Or..

  • Click the dashboard menu from in the upper-left corner next to the Grafana icon. Select Alert Status.

3.3.6.4. Configuring the Notification Target

A notification channel called cephmetrics is automatically created during installation. All preconfigured alerts reference the cephmetrics channel but before you can receive the alerts, complete the notification channel definition by selecting the desired notification type. The Grafana platform supports a number of different notification types including email, Slack, and PagerDuty.

Procedure
  • To configure the notification channel, follow the instructions in the Alert Notifications section on the Grafana web page.

3.3.6.5. Changing the Default Alerts and Adding New Ones

This section explains how to change the trigger value on already configured alerts and how to add new alerts to the Alert Status dashboard.

Procedure
  • To change the trigger value on alerts or to add new alerts, follow the Alerting Engine & Rules Guide on the Grafana web pages.

    Important

    To prevent overriding custom alerts, the Alert Status dashboard will not be updated when upgrading the Red Hat Ceph Storage Dashboard packages when you change the trigger values or add new alerts.

Additional Resources

3.4. Using ceph-medic to diagnose a Ceph Storage Cluster

The ceph-medic utility performs checks against a running Ceph Storage Cluster to identify potential problems.

The ceph-medics utility for example checks:

  • The correct ownership of files and directories
  • If the fsid is the same for all nodes in the storage cluster
  • If the secret key in the keyring is different than other nodes in the storage cluster

3.4.1. Prerequisites

  • A working Red Hat Ceph Storage cluster
  • SSH and sudo access to storage nodes

3.4.2. Installing the ceph-medic Utility

Prerequisite
  • Access to the Red Hat Ceph Storage 3 software repositories
Procedure

Do the following steps on the Ansible administration node, as the root user.

  1. Install the ceph-medic package:

    [root@admin ~]# yum install ceph-medic
  2. Verify the installation of ceph-medic:

    [root@admin ~]# ceph-medic --help
Additional Resources

3.4.3. Running a Diagnostic Check

A basic check for potential problems with a Ceph Storage Cluster.

Prerequisites
  • A working Red Hat Ceph Storage cluster
  • SSH and sudo access to storage nodes
Procedure

Do the following step from the Ansible administration node, as a normal user.

  • Use the ceph-medic check command:

    [admin@admin ~]$ ceph-medic check
    Host: mon0                  connection: [connected  ]
    Host: mon1                  connection: [connected  ]
    Host: mon2                  connection: [connected  ]
    Host: osd0                  connection: [connected  ]
    Host: osd1                  connection: [connected  ]
    Host: osd2                  connection: [connected  ]
    Collection completed!
    
    =======================  Starting remote check session  ========================
    Version: 1.0.2    Cluster Name: "example"
    Total hosts: [6]
    OSDs:    3    MONs:    3     Clients:    0
    MDSs:    0    RGWs:    0     MGRs:       0
    
    ================================================================================
    
    ------------ osds ------------
     osd0
     osd1
     osd2
    
    ------------ mons ------------
     mon0
     mon1
     mon2
    
    17 passed, 0 errors, on 6 hosts
Additional Resources

3.4.4. Using a Custom Inventory File

The ceph-medic utility must know the storage cluster topology. By default, ceph-medic uses the Ansible inventory file (/etc/ansible/hosts) to detect nodes.

Prerequisites
  • A working Red Hat Ceph Storage cluster
  • SSH and sudo access to storage nodes
Procedure

To use a custom inventory file, do the following steps on the Ansible administration node, as a user.

  1. Create a custom hosts file:

    [admin@admin ~]$ touch ~/example/hosts
  2. Open the hosts file for editing. Add the nodes in the storage cluster under the appropriate node group type. The ceph-medic tool supports the following node group types: mons, osds, rgws, mdss, mgrs and clients.

    For example:

    [mons]
    mon0
    mon1
    mon2
    
    [osds]
    osd0
    osd1
    osd2
  3. When doing a diagnostic check, to specify a custom inventory file, use the --inventory option:

    ceph-medic --inventory $PATH_TO_HOSTS_FILE check
    Replace
    • $PATH_TO_HOSTS_FILE with the full path to the hosts file.

      For example:

      [admin@admin ~]$ ceph-medic --inventory ~/example/hosts check
Additional Resources

3.4.5. Configuring a Custom Logging Path

The ceph-medic log file contains more verbosity than the command output on the terminal. The ceph-medic tool by default writes logs to the current working directory.

Prerequisites
  • A working Red Hat Ceph Storage cluster
  • SSH and sudo access to storage nodes
Procedure

To change where these logs are written, do the following step on the Ansible administration node, as a normal user.

  1. Open the ~/.cephmedic.conf file for editing.
  2. Change the --log-path option, from a . to a custom log location.

    For example:

    --log-path = /var/log/ceph-medic/
Additional Resources
Red Hat logoGithubRedditYoutubeTwitter

자세한 정보

평가판, 구매 및 판매

커뮤니티

Red Hat 문서 정보

Red Hat을 사용하는 고객은 신뢰할 수 있는 콘텐츠가 포함된 제품과 서비스를 통해 혁신하고 목표를 달성할 수 있습니다.

보다 포괄적 수용을 위한 오픈 소스 용어 교체

Red Hat은 코드, 문서, 웹 속성에서 문제가 있는 언어를 교체하기 위해 최선을 다하고 있습니다. 자세한 내용은 다음을 참조하세요.Red Hat 블로그.

Red Hat 소개

Red Hat은 기업이 핵심 데이터 센터에서 네트워크 에지에 이르기까지 플랫폼과 환경 전반에서 더 쉽게 작업할 수 있도록 강화된 솔루션을 제공합니다.

© 2024 Red Hat, Inc.