ceph/doc/mgr/prometheus.rst

   1 .. _mgr-prometheus:
   2
   3 =================
   4 Prometheus Module
   5 =================
   6
   7 Provides a Prometheus exporter to pass on Ceph performance counters
   8 from the collection point in ceph-mgr.  Ceph-mgr receives MMgrReport
   9 messages from all MgrClient processes (mons and OSDs, for instance)
  10 with performance counter schema data and actual counter data, and keeps
  11 a circular buffer of the last N samples.  This module creates an HTTP
  12 endpoint (like all Prometheus exporters) and retrieves the latest sample
  13 of every counter when polled (or "scraped" in Prometheus terminology).
  14 The HTTP path and query parameters are ignored; all extant counters
  15 for all reporting entities are returned in text exposition format.
  16 (See the Prometheus `documentation <https://prometheus.io/docs/instrumenting/exposition_formats/#text-format-details>`_.)
  17
  18 Enabling prometheus output
  19 ==========================
  20
  21 The *prometheus* module is enabled with::
  22
  23   ceph mgr module enable prometheus
  24
  25 Configuration
  26 -------------
  27
  28 By default the module will accept HTTP requests on port ``9283`` on all
  29 IPv4 and IPv6 addresses on the host.  The port and listen address are both
  30 configurable with ``ceph config-key set``, with keys
  31 ``mgr/prometheus/server_addr`` and ``mgr/prometheus/server_port``.
  32 This port is registered with Prometheus's `registry <https://github.com/prometheus/prometheus/wiki/Default-port-allocations>`_.
  33
  34 .. _prometheus-rbd-io-statistics:
  35
  36 RBD IO statistics
  37 -----------------
  38
  39 The module can optionally collect RBD per-image IO statistics by enabling
  40 dynamic OSD performance counters. The statistics are gathered for all images
  41 in the pools that are specified in the ``mgr/prometheus/rbd_stats_pools``
  42 configuration parameter. The parameter is a comma or space separated list
  43 of ``pool[/namespace]`` entries. If the namespace is not specified the
  44 statistics are collected for all namespaces in the pool.
  45
  46 Example to activate the RBD-enabled pools ``pool1``, ``pool2`` and ``poolN``::
  47
  48   ceph config set mgr mgr/prometheus/rbd_stats_pools "pool1,pool2,poolN"
  49
  50 The module makes the list of all available images scanning the specified
  51 pools and namespaces and refreshes it periodically. The period is
  52 configurable via the ``mgr/prometheus/rbd_stats_pools_refresh_interval``
  53 parameter (in sec) and is 300 sec (5 minutes) by default. The module will
  54 force refresh earlier if it detects statistics from a previously unknown
  55 RBD image.
  56
  57 Example to turn up the sync interval to 10 minutes::
  58
  59   ceph config set mgr mgr/prometheus/rbd_stats_pools_refresh_interval 600
  60
  61 Statistic names and labels
  62 ==========================
  63
  64 The names of the stats are exactly as Ceph names them, with
  65 illegal characters ``.``, ``-`` and ``::`` translated to ``_``,
  66 and ``ceph_`` prefixed to all names.
  67
  68
  69 All *daemon* statistics have a ``ceph_daemon`` label such as "osd.123"
  70 that identifies the type and ID of the daemon they come from.  Some
  71 statistics can come from different types of daemon, so when querying
  72 e.g. an OSD's RocksDB stats, you would probably want to filter
  73 on ceph_daemon starting with "osd" to avoid mixing in the monitor
  74 rocksdb stats.
  75
  76
  77 The *cluster* statistics (i.e. those global to the Ceph cluster)
  78 have labels appropriate to what they report on.  For example,
  79 metrics relating to pools have a ``pool_id`` label.
  80
  81
  82 The long running averages that represent the histograms from core Ceph
  83 are represented by a pair of ``<name>_sum`` and ``<name>_count`` metrics.
  84 This is similar to how histograms are represented in `Prometheus <https://prometheus.io/docs/concepts/metric_types/#histogram>`_
  85 and they can also be treated `similarly <https://prometheus.io/docs/practices/histograms/>`_.
  86
  87 Pool and OSD metadata series
  88 ----------------------------
  89
  90 Special series are output to enable displaying and querying on
  91 certain metadata fields.
  92
  93 Pools have a ``ceph_pool_metadata`` field like this:
  94
  95 ::
  96
  97     ceph_pool_metadata{pool_id="2",name="cephfs_metadata_a"} 1.0
  98
  99 OSDs have a ``ceph_osd_metadata`` field like this:
 100
 101 ::
 102
 103     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
 104
 105
 106 Correlating drive statistics with node_exporter
 107 -----------------------------------------------
 108
 109 The prometheus output from Ceph is designed to be used in conjunction
 110 with the generic host monitoring from the Prometheus node_exporter.
 111
 112 To enable correlation of Ceph OSD statistics with node_exporter's
 113 drive statistics, special series are output like this:
 114
 115 ::
 116
 117     ceph_disk_occupation{ceph_daemon="osd.0",device="sdd", exported_instance="myhost"}
 118
 119 To use this to get disk statistics by OSD ID, use either the ``and`` operator or
 120 the ``*`` operator in your prometheus query. All metadata metrics (like ``
 121 ceph_disk_occupation`` have the value 1 so they act neutral with ``*``. Using ``*``
 122 allows to use ``group_left`` and ``group_right`` grouping modifiers, so that
 123 the resulting metric has additional labels from one side of the query.
 124
 125 See the
 126 `prometheus documentation`__ for more information about constructing queries.
 127
 128 __ https://prometheus.io/docs/prometheus/latest/querying/basics
 129
 130 The goal is to run a query like
 131
 132 ::
 133
 134     rate(node_disk_bytes_written[30s]) and on (device,instance) ceph_disk_occupation{ceph_daemon="osd.0"}
 135
 136 Out of the box the above query will not return any metrics since the ``instance`` labels of
 137 both metrics don't match. The ``instance`` label of ``ceph_disk_occupation``
 138 will be the currently active MGR node.
 139
 140  The following two section outline two approaches to remedy this.
 141
 142 Use label_replace
 143 =================
 144
 145 The ``label_replace`` function (cp.
 146 `label_replace documentation <https://prometheus.io/docs/prometheus/latest/querying/functions/#label_replace>`_)
 147 can add a label to, or alter a label of, a metric within a query.
 148
 149 To correlate an OSD and its disks write rate, the following query can be used:
 150
 151 ::
 152
 153     label_replace(rate(node_disk_bytes_written[30s]), "exported_instance", "$1", "instance", "(.*):.*") and on (device,exported_instance) ceph_disk_occupation{ceph_daemon="osd.0"}
 154
 155 Configuring Prometheus server
 156 =============================
 157
 158 honor_labels
 159 ------------
 160
 161 To enable Ceph to output properly-labeled data relating to any host,
 162 use the ``honor_labels`` setting when adding the ceph-mgr endpoints
 163 to your prometheus configuration.
 164
 165 This allows Ceph to export the proper ``instance`` label without prometheus
 166 overwriting it. Without this setting, Prometheus applies an ``instance`` label
 167 that includes the hostname and port of the endpoint that the series came from.
 168 Because Ceph clusters have multiple manager daemons, this results in an
 169 ``instance`` label that changes spuriously when the active manager daemon
 170 changes.
 171
 172 If this is undesirable a custom ``instance`` label can be set in the
 173 Prometheus target configuration: you might wish to set it to the hostname
 174 of your first mgr daemon, or something completely arbitrary like "ceph_cluster".
 175
 176 node_exporter hostname labels
 177 -----------------------------
 178
 179 Set your ``instance`` labels to match what appears in Ceph's OSD metadata
 180 in the ``instance`` field.  This is generally the short hostname of the node.
 181
 182 This is only necessary if you want to correlate Ceph stats with host stats,
 183 but you may find it useful to do it in all cases in case you want to do
 184 the correlation in the future.
 185
 186 Example configuration
 187 ---------------------
 188
 189 This example shows a single node configuration running ceph-mgr and
 190 node_exporter on a server called ``senta04``. Note that this requires to add the
 191 appropriate instance label to every ``node_exporter`` target individually.
 192
 193 This is just an example: there are other ways to configure prometheus
 194 scrape targets and label rewrite rules.
 195
 196 prometheus.yml
 197 ~~~~~~~~~~~~~~
 198
 199 ::
 200
 201     global:
 202       scrape_interval:     15s
 203       evaluation_interval: 15s
 204
 205     scrape_configs:
 206       - job_name: 'node'
 207         file_sd_configs:
 208           - files:
 209             - node_targets.yml
 210       - job_name: 'ceph'
 211         honor_labels: true
 212         file_sd_configs:
 213           - files:
 214             - ceph_targets.yml
 215
 216
 217 ceph_targets.yml
 218 ~~~~~~~~~~~~~~~~
 219
 220
 221 ::
 222
 223     [
 224         {
 225             "targets": [ "senta04.mydomain.com:9283" ],
 226             "labels": {}
 227         }
 228     ]
 229
 230
 231 node_targets.yml
 232 ~~~~~~~~~~~~~~~~
 233
 234 ::
 235
 236     [
 237         {
 238             "targets": [ "senta04.mydomain.com:9100" ],
 239             "labels": {
 240                 "instance": "senta04"
 241             }
 242         }
 243     ]
 244
 245
 246 Notes
 247 =====
 248
 249 Counters and gauges are exported; currently histograms and long-running
 250 averages are not.  It's possible that Ceph's 2-D histograms could be
 251 reduced to two separate 1-D histograms, and that long-running averages
 252 could be exported as Prometheus' Summary type.
 253
 254 Timestamps, as with many Prometheus exporters, are established by
 255 the server's scrape time (Prometheus expects that it is polling the
 256 actual counter process synchronously).  It is possible to supply a
 257 timestamp along with the stat report, but the Prometheus team strongly
 258 advises against this.  This means that timestamps will be delayed by
 259 an unpredictable amount; it's not clear if this will be problematic,
 260 but it's worth knowing about.