Chapter 3. Monitoring

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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

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

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.

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.

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.

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

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:

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

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

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

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"}

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

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:

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:

Or..

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

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

Additional Resources