apache-ignite
484 строки · 20.4 Кб
1// Licensed to the Apache Software Foundation (ASF) under one or more
2// contributor license agreements. See the NOTICE file distributed with
3// this work for additional information regarding copyright ownership.
4// The ASF licenses this file to You under the Apache License, Version 2.0
5// (the "License"); you may not use this file except in compliance with
6// the License. You may obtain a copy of the License at
7//
8// http://www.apache.org/licenses/LICENSE-2.0
9//
10// Unless required by applicable law or agreed to in writing, software
11// distributed under the License is distributed on an "AS IS" BASIS,
12// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13// See the License for the specific language governing permissions and
14// limitations under the License.
15= Metrics System
16
17:javaFile: {javaCodeDir}/ConfiguringMetrics.java
18
19== Overview
20
21This section explains the metrics system and how you can use it to monitor your cluster.
22//the types of metrics and how to export them, but first let's explore the basic concepts of the metrics mechanism in Ignite.
23
24Let's explore the basic concepts of the metrics system in Ignite.
25First, there are different metrics.
26Each metric has a name and a return value.
27The return value can be a simple value like `String`, `long`, or `double`, or can represent a Java object.
28Some metrics represent <<histograms>>.
29
30And then there are different ways to export the metrics — what we call _exporters_.
31To put it another way, the exporter are different ways you can access the metrics.
32Each exporter always gives access to all available metrics.
33
34Ignite includes the following exporters:
35
36* JMX (default)
37* SQL Views
38* Log files
39* OpenCensus
40
41You can create a custom exporter by implementing the javadoc:org.apache.ignite.spi.metric.MetricExporterSpi[] interface.
42
43
44== Metric Registries [[registry]]
45
46Metrics are grouped into categories (called _registries_).
47Each registry has a name.
48The full name of a specific metric within the registry consists of the registry name followed by a dot, followed by the name of the metric: `<registry_name>.<metric_name>`.
49For example, the registry for data storage metrics is called `io.datastorage`.
50The metric that returns the storage size is called `io.datastorage.StorageSize`.
51
52The list of all registries and the metrics they contain are described link:monitoring-metrics/new-metrics[here].
53
54== Metric Exporters
55
56If you want to enable metrics, configure one or multiple metric exporters in the node configuration.
57This is a node-specific configuration, which means it enables metrics only on the node where it is specified.
58
59[tabs]
60--
61tab:XML[]
62[source, xml]
63----
64include::code-snippets/xml/metrics.xml[tags=ignite-config;!discovery, indent=0]
65----
66
67tab:Java[]
68
69[source, java]
70----
71include::{javaFile}[tags=new-metric-framework, indent=0]
72----
73
74tab:C#/.NET[]
75
76tab:C++[unsupported]
77--
78
79The following sections describe the exporters available in Ignite by default.
80
81
82=== JMX
83
84`org.apache.ignite.spi.metric.jmx.JmxMetricExporterSpi` exposes metrics via JMX beans.
85
86[tabs]
87--
88tab:Java[]
89[source, java]
90----
91include::{javaFile}[tags=metrics-filter, indent=0]
92----
93
94tab:C#/.NET[]
95
96tab:C++[unsupported]
97--
98
99[NOTE]
100====
101This exporter is enabled by default if nothing is set with `IgniteConfiguration.setMetricExporterSpi(...)`.
102====
103
104[tabs]
105--
106tab:Java[]
107[source, java]
108----
109include::{javaFile}[tags=disable-default-jmx-exporter, indent=0]
110----
111
112tab:C#/.NET[]
113
114tab:C++[unsupported]
115--
116
117==== Enabling JMX for Ignite
118
119By default, the JMX automatic configuration is disabled.
120To enable it, configure the following environment variables:
121
122* For `control.sh`, configure the `CONTROL_JVM_OPTS` variable
123* For `ignite.sh`, configure the `JVM_OPTS` variable
124
125For example:
126
127[source,shell]
128----
129JVM_OPTS="-Dcom.sun.management.jmxremote -Dcom.sun.management.jmxremote.port=${JMX_PORT} \
130-Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote.ssl=false"
131----
132
133
134// link:monitoring-metrics/configuring-metrics[Configuring Metrics]
135
136==== Understanding MBean's ObjectName
137
138Every JMX Mbean has an https://docs.oracle.com/javase/8/docs/api/javax/management/ObjectName.html[ObjectName,window=_blank].
139The ObjectName is used to identify the bean.
140The ObjectName consists of a domain and a list of key properties, and can be represented as a string as follows:
141
142domain: key1 = value1 , key2 = value2
143
144All Ignite metrics have the same domain: `org.apache.<classloaderId>` where the classloader ID is optional (omitted if you set `IGNITE_MBEAN_APPEND_CLASS_LOADER_ID=false`). In addition, each metric has two properties: `group` and `name`.
145For example:
146
147org.apache:group=SPIs,name=TcpDiscoverySpi
148
149This MBean provides various metrics related to node discovery.
150
151The MBean ObjectName can be used to identify the bean in UI tools like JConsole.
152For example, JConsole displays MBeans in a tree-like structure where all beans are first grouped by domain and then by the 'group' property:
153
154image::images/jconsole.png[]
155
156{sp}+
157
158=== SQL View
159
160`SqlViewMetricExporterSpi` is enabled by default, `SqlViewMetricExporterSpi` exposes metrics via the `SYS.METRICS` view.
161Each metric is displayed as a single record.
162You can use any supported SQL tool to view the metrics:
163
164[source, shell,subs="attributes"]
165----
166> select name, value from SYS.METRICS where name LIKE 'cache.myCache.%';
167+-----------------------------------+--------------------------------+
168| NAME | VALUE |
169+-----------------------------------+--------------------------------+
170| cache.myCache.CacheTxRollbacks | 0 |
171| cache.myCache.OffHeapRemovals | 0 |
172| cache.myCache.QueryCompleted | 0 |
173| cache.myCache.QueryFailed | 0 |
174| cache.myCache.EstimatedRebalancingKeys | 0 |
175| cache.myCache.CacheEvictions | 0 |
176| cache.myCache.CommitTime | [J@2eb66498 |
177....
178----
179
180=== Log
181
182`org.apache.ignite.spi.metric.log.LogExporterSpi` prints the metrics to the log file at regular intervals (1 min by default) at INFO level.
183
184[tabs]
185--
186tab:XML[]
187
188[source, xml]
189----
190include::code-snippets/xml/metrics.xml[tags=!*;ignite-config;log-exporter, indent=0]
191----
192
193
194tab:Java[]
195
196If you use programmatic configuration, you can change the print frequency as follows:
197
198[source, java]
199----
200include::{javaFile}[tags=log-exporter, indent=0]
201----
202
203tab:C#/.NET[]
204tab:C++[]
205--
206
207=== OpenCensus
208
209`org.apache.ignite.spi.metric.opencensus.OpenCensusMetricExporterSpi` adds integration with the OpenCensus library.
210
211To use the OpenCensus exporter:
212
213. link:setup#enabling-modules[Enable the 'ignite-opencensus' module].
214. Add `org.apache.ignite.spi.metric.opencensus.OpenCensusMetricExporterSpi` to the list of exporters in the node configuration.
215. Configure OpenCensus StatsCollector to export to a specific system. See link:{githubUrl}/examples/src/main/java/org/apache/ignite/examples/opencensus/OpenCensusMetricsExporterExample.java[OpenCensusMetricsExporterExample.java] for an example and OpenCensus documentation for additional information.
216
217
218Configuration parameters:
219
220* `filter` - predicate that filters metrics.
221* `period` - export period.
222* `sendInstanceName` - if enabled, a tag with the Ignite instance name is added to each metric.
223* `sendNodeId` - if enabled, a tag with the Ignite node id is added to each metric.
224* `sendConsistentId` - if enabled, a tag with the Ignite node consistent id is added to each metric.
225
226
227
228
229== Histograms
230
231The metrics that represent histograms are available in the JMX exporter only.
232Histogram metrics are exported as a set of values where each value corresponds to a specific bucket and is available through a separate JMX bean attribute.
233The attribute names of a histogram metric have the following format:
234
235```
236{metric_name}_{low_bound}_{high_bound}
237```
238
239where
240
241* `{metric_name}` - the name of the metric.
242* `{low_bound}` - start of the bound. `0` for the first bound.
243* `{high_bound}` - end of the bound. `inf` for the last bound.
244
245
246Example of the metric names if the bounds are [10,100]:
247
248* `histogram_0_10` - less than 10.
249* `histogram_10_100` - between 10 and 100.
250* `histogram_100_inf` - more than 100.
251
252== Common monitoring tasks
253=== Monitoring the Amount of Data
254
255If you do not use link:persistence/native-persistence[Native persistence] (i.e., all your data is kept in memory), you would want to monitor RAM usage.
256If you use Native persistence, in addition to RAM, you should monitor the size of the data storage on disk.
257
258The size of the data loaded into a node is available at different levels of aggregation. You can monitor for:
259
260* The total size of the data the node keeps on disk or in RAM. This amount is the sum of the size of each configured data region (in the simplest case, only the default data region) plus the sizes of the system data regions.
261* The size of a specific link:memory-configuration/data-regions[data region] on that node. The data region size is the sum of the sizes of all cache groups.
262* The size of a specific cache/cache group on that node, including the backup partitions.
263
264
265==== Allocated Space vs. Actual Size of Data
266
267There is no way to get the exact size of the data (neither in RAM nor on disk). Instead, there are two ways to estimate it.
268
269You can get the size of the space _allocated_ for storing the data.
270(The "space" here refers either to the space in RAM or on disk depending on whether you use Native persistence or not.)
271Space is allocated when the size of the storage gets full and more entries need to be added.
272However, when you remove entries from caches, the space is not deallocated.
273It is reused when new entries need to be added to the storage on subsequent write operations. Therefore, the allocated size does not decrease when you remove entries from the caches.
274The allocated size is available at the level of data storage, data region, and cache group metrics.
275The metric is called `TotalAllocatedSize`.
276
277You can also get an estimate of the actual size of data by multiplying the number of link:memory-centric-storage#data-pages[data pages] in use by the fill factor. The fill factor is the ratio of the size of data in a page to the page size, averaged over all pages. The number of pages in use and the fill factor are available at the level of data <<Data Region Size,region metrics>>.
278
279Add up the estimated size of all data regions to get the estimated total amount of data on the node.
280
281
282:allocsize_note: Note that when Native persistence is disabled, this metric shows the total size of the allocated space in RAM.
283
284==== Monitoring RAM Memory Usage
285The amount of data in RAM can be monitored for each data region through the following metrics:
286
287[{table_opts}]
288|===
289| Attribute | Type | Description | Scope
290
291| PagesFillFactor| float | The average size of data in pages as a ratio of the page size. When Native persistence is enabled, this metric is applicable only to the persistent storage (i.e. pages on disk). | Node
292| TotalUsedPages | long | The number of data pages that are currently in use. When Native persistence is enabled, this metric is applicable only to the persistent storage (i.e. pages on disk).| Node
293| PhysicalMemoryPages |long | The number of the allocated pages in RAM. | Node
294| PhysicalMemorySize |long |The size of the allocated space in RAM in bytes. | Node
295|===
296
297If you have multiple data regions, add up the sizes of all data regions to get the total size of the data on the node.
298
299==== Monitoring Storage Size
300
301Persistent storage, when enabled, saves all application data on disk.
302The total amount of data each node keeps on disk consists of the persistent storage (application data), the link:persistence/native-persistence#write-ahead-log[WAL files], and link:persistence/native-persistence#wal-archive[WAL Archive] files.
303
304===== Persistent Storage Size
305To monitor the size of the persistent storage on disk, use the following metrics:
306
307[{table_opts}]
308|===
309| Attribute | Type | Description | Scope
310| TotalAllocatedSize | long | The size of the space allocated on disk for the entire data storage (in bytes). {allocsize_note} | Node
311| WalTotalSize | long | Total size of the WAL files in bytes, including the WAL archive files. | Node
312| WalArchiveSegments | int | The number of WAL segments in the archive. | Node
313|===
314
315===== Data Region Size
316
317For each configured data region, Metrics collection for data regions are disabled by default. You can link:monitoring-metrics/configuring-metrics#enabling-data-region-metrics[enable it in the data region configuration.
318
319The size of the data region on a node comprises the size of all partitions (including backup partitions) that this node owns for all caches in that data region.
320
321[{table_opts}]
322|===
323| Attribute | Type | Description | Scope
324
325| TotalAllocatedSize | long | The size of the space allocated for this data region (in bytes). {allocsize_note} | Node
326| PagesFillFactor| float | The average amount of data in pages as a ratio of the page size. | Node
327| TotalUsedPages | long | The number of data pages that are currently in use. | Node
328| PhysicalMemoryPages |long |The number of data pages in this data region held in RAM. | Node
329| PhysicalMemorySize | long |The size of the allocated space in RAM in bytes.| Node
330|===
331
332===== Cache Group Size
333
334If you don't use link:configuring-caches/cache-groups[cache groups], each cache will be its own group.
335
336[{table_opts}]
337|===
338| Attribute | Type | Description | Scope
339|TotalAllocatedSize |long | The amount of space allocated for the cache group on this node. | Node
340|===
341
342=== Monitoring Checkpointing Operations
343Checkpointing may slow down cluster operations.
344You may want to monitor how much time each checkpoint operation takes, so that you can tune the properties that affect checkpointing.
345You may also want to monitor the disk performance to see if the slow-down is caused by external reasons.
346
347See link:persistence/persistence-tuning#pages-writes-throttling[Pages Writes Throttling] and link:persistence/persistence-tuning#adjusting-checkpointing-buffer-size[Checkpointing Buffer Size] for performance tips.
348
349[{table_opts}]
350|===
351| Attribute | Type | Description | Scope
352| DirtyPages | long | The number of pages in memory that have been changed but not yet synchronized to disk. Those will be written to disk during next checkpoint. | Node
353|LastCheckpointDuration | long | The time in milliseconds it took to create the last checkpoint. | Node
354|CheckpointBufferSize | long | The size of the checkpointing buffer. | Global
355|===
356
357=== Monitoring Rebalancing
358link:data-rebalancing[Rebalancing] is the process of moving partitions between the cluster nodes so that the data is always distributed in a balanced manner. Rebalancing is triggered when a new node joins, or an existing node leaves the cluster.
359
360If you have multiple caches, they will be rebalanced sequentially.
361There are several metrics that you can use to monitor the progress of the rebalancing process for a specific cache.
362
363In the metric system, link:monitoring-metrics/new-metrics#caches[Cache metrics]:
364[{table_opts}]
365|===
366| Attribute | Type | Description | Scope
367|RebalancingStartTime | long | This metric shows the time when rebalancing of local partitions started for the cache. This metric will return 0 if the local partitions do not participate in the rebalancing. The time is returned in milliseconds. | Node
368| EstimatedRebalancingFinishTime | long | Expected time of completion of the rebalancing process. | Node
369| KeysToRebalanceLeft | long | The number of keys on the node that remain to be rebalanced. You can monitor this metric to learn when the rebalancing process finishes.| Node
370|===
371
372=== Monitoring Topology
373Topology refers to the set of nodes in a cluster. There are a number of metrics that expose the information about the topology of the cluster. If the topology changes too frequently or has a size that is different from what you expect, you may want to look into whether there are network problems.
374
375[{table_opts}]
376|===
377| Attribute | Type | Description | Scope
378| TotalServerNodes| long |The number of server nodes in the cluster.| Global
379| TotalClientNodes| long |The number of client nodes in the cluster. | Global
380| TotalBaselineNodes | long | The number of nodes that are registered in the link:clustering/baseline-topology[baseline topology]. When a node goes down, it remains registered in the baseline topology and you need to remote it manually. | Global
381| ActiveBaselineNodes | long | The number of nodes that are currently active in the baseline topology. | Global
382|===
383
384[{table_opts}]
385|===
386| Attribute | Type | Description | Scope
387| Coordinator | String | The node ID of the current coordinator node.| Global
388| CoordinatorNodeFormatted|String a|
389Detailed information about the coordinator node.
390....
391TcpDiscoveryNode [id=e07ad289-ff5b-4a73-b3d4-d323a661b6d4,
392consistentId=fa65ff2b-e7e2-4367-96d9-fd0915529c25,
393addrs=[0:0:0:0:0:0:0:1%lo, 127.0.0.1, 172.25.4.200],
394sockAddrs=[mymachine.local/172.25.4.200:47500,
395/0:0:0:0:0:0:0:1%lo:47500, /127.0.0.1:47500], discPort=47500,
396order=2, intOrder=2, lastExchangeTime=1568187777249, loc=false,
397ver=8.7.5#20190520-sha1:d159cd7a, isClient=false]
398....
399
400| Global
401|===
402
403=== Monitoring Caches
404
405See the new metric system, link:monitoring-metrics/new-metrics#caches[Cache metrics].
406
407==== Monitoring Build and Rebuild Indexes
408
409To get an estimate on how long it takes to rebuild cache indexes, you can use one of the metrics listed below:
410
411. `IsIndexRebuildInProgress` - tells whether indexes are being built or rebuilt at the moment;
412. `IndexBuildCountPartitionsLeft` - gives the remaining number of partitions (by cache group) for indexes to rebuild.
413
414Note that the `IndexBuildCountPartitionsLeft` metric allows to estimate only an approximate number of indexes left to rebuild.
415For a more accurate estimate, use the `IndexRebuildKeyProcessed` cache metric:
416
417* Use `isIndexRebuildInProgress` to know whether the indexes are being rebuilt for the cache.
418
419* Use `IndexRebuildKeysProcessed` to know the number of keys with rebuilt indexes. If the rebuilding is in progress, it gives a number of keys with indexes being rebuilt at the current moment. Otherwise, it gives a total number of the of keys with rebuilt indexes. The values are reset before the start of each rebuilding.
420
421=== Monitoring Transactions
422Note that if a transaction spans multiple nodes (i.e., if the keys that are changed as a result of the transaction execution are located on multiple nodes), the counters will increase on each node. For example, the 'TransactionsCommittedNumber' counter will increase on each node where the keys affected by the transaction are stored.
423
424[{table_opts}]
425|===
426| Attribute | Type | Description | Scope
427| LockedKeysNumber | long | The number of keys locked on the node. | Node
428| TransactionsCommittedNumber |long | The number of transactions that have been committed on the node | Node
429| TransactionsRolledBackNumber | long | The number of transactions that were rolled back. | Node
430| OwnerTransactionsNumber | long | The number of transactions initiated on the node. | Node
431| TransactionsHoldingLockNumber | long | The number of open transactions that hold a lock on at least one key on the node.| Node
432|===
433
434=== Monitoring Snapshots
435
436[{table_opts}]
437|===
438| Attribute | Type | Description | Scope
439| LastSnapshotOperation | | |
440| LastSnapshotStartTime || |
441| SnapshotInProgress | | |
442|===
443
444=== Monitoring Client Connections
445Metrics related to JDBC/ODBC or thin client connections.
446
447[{table_opts}]
448|===
449| Attribute | Type | Description | Scope
450| Connections | java.util.List<String> a| A list of strings, each string containing information about a connection:
451
452....
453JdbcClient [id=4294967297, user=<anonymous>,
454rmtAddr=127.0.0.1:39264, locAddr=127.0.0.1:10800]
455....
456| Node
457|===
458
459
460=== Monitoring Message Queues
461When thread pools queues' are growing, it means that the node cannot keep up with the load, or there was an error while processing messages in the queue.
462Continuous growth of the queue size can lead to OOM errors.
463
464==== Communication Message Queue
465The queue of outgoing communication messages contains communication messages that are waiting to be sent to other nodes.
466If the size is growing, it means there is a problem.
467
468[{table_opts}]
469|===
470| Attribute | Type | Description | Scope
471| OutboundMessagesQueueSize | int | The size of the queue of outgoing communication messages. | Node
472|===
473
474==== Discovery Messages Queue
475
476The queue of discovery messages.
477
478[{table_opts}]
479|===
480| Attribute | Type | Description | Scope
481| MessageWorkerQueueSize | int | The size of the queue of discovery messages that are waiting to be sent to other nodes. | Node
482|AvgMessageProcessingTime|long| Average message processing time. | Node
483|===
484--
485