[
https://issues.apache.org/jira/browse/IGNITE-29236?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ilya Shishkov updated IGNITE-29236:
-----------------------------------
Description:
h3. Problem
The
[Metrics|https://ignite.apache.org/docs/ignite2/latest/monitoring-metrics/new-metrics]
page
({{docs/_docs/monitoring-metrics/new-metrics.adoc}}) mostly matches the metrics
published by a node. It doesn't match
in the following places:
* *Sorted Indexes Operations*
({{index.\{schema_name\}.\{table_name\}.\{index_name\}}}): the table contains
only
the templates {{\{opType\}Count}} and {{\{opType\}Time}}, and the text names
5 of the 11 operations. A node actually
publishes 22 metrics: {{AskNeighbor}}, {{Insert}},
{{LockBackAndRmvFromLeaf}}, {{LockBackAndTail}}, {{LockTail}},
{{LockTailExact}}, {{LockTailForward}}, {{RemoveFromLeaf}},
{{RemoveRangeFromLeaf}}, {{Replace}}, {{Search}}, each
with the {{Count}} and {{Time}} suffixes.
* *Discovery IO* ({{io.discovery}}): only {{TcpDiscoverySpi}} metrics are
described. With {{ZookeeperDiscoverySpi}}
the register contains a different set: {{CommunicationErrors}},
{{Coordinator}}, {{CurrentTopologyVersion}},
{{FailedNodes}}, {{JoinedNodes}}, {{LeftNodes}}. {{CommunicationErrors}}
isn't documented at all, and the counters
are {{long}}, not {{integer}} as for TCP discovery
({{ZookeeperDiscoveryStatistics}},
{{ZookeeperDiscoverySpi#onContextInitialized0}}).
* The page doesn't say which metrics and registers exist on server nodes only:
** {{cacheGroups.\{group_name\}}}: {{InMemoryIndexPages}} and
{{TotalAllocatedPages}} are not registered on client
nodes (no data region, {{CacheGroupMetricsImpl}}).
** {{rest.client}}: on client nodes the register exists only with
{{IGNITE_REST_START_ON_CLIENT=true}}
({{GridRestProcessor}}).
** {{snapshot}}, {{snapshot.incremental}}, {{snapshot-restore}},
{{snapshot-check.\{snapshot_name\}}}: server nodes
only ({{IgniteSnapshotManager#start0}} exits on client nodes).
** {{index.*}}: only on the nodes that store the index data.
h3. Changes
Documentation only, {{new-metrics.adoc}}:
* The {{\{opType\}}} template is replaced by the explicit list of 22 metrics.
The note that the set of operations
depends on the index implementation is kept.
* A "ZooKeeper Discovery" subsection is added to "Discovery IO" with the
metrics of {{ZookeeperDiscoverySpi}}.
* Notes about server-only metrics and registers are added to "Cache Groups",
"Sorted Indexes Operations",
"Ignite REST Client Connector" and "Snapshot".
h3. How it was verified
A cluster of 2 server nodes and 1 client node was started from the current
master with most optional features
enabled, so that conditional metrics are registered:
* persistence with CDC and an in-memory region;
* cache statistics, a near cache and an encrypted cache group with key rotation;
* SSL, a user executor, a service with statistics, H2 and Calcite;
* secondary indexes, a thin client and data structures;
* full and incremental snapshots with check and restore;
* a suppressed failure.
All metric registers of every node, including registers that exist only while
an operation runs, were dumped and
compared with the tables of the page.
Results:
* JMX export ({{JmxMetricExporterSpi}}) matches the registers 1:1 on every node
(srv: 2304/2306 metrics, client: 2177).
* Before the patch: 22 metrics are missing from the page, and 2 template rows
don't exist at runtime.
* After the patch: no discrepancies.
* Not reproducible in this setup and checked against the code:
{{ConflictResolver*}} (needs a conflict resolver; the
page already notes the condition) and the ZooKeeper discovery metrics.
* The {{cdc}} register described in {{persistence/change-data-capture.adoc}}
matches the code (13/13).
was:
h3. Problem
The [System
Views|https://ignite.apache.org/docs/ignite2/latest/monitoring-metrics/system-views]
page
({{docs/_docs/monitoring-metrics/system-views.adoc}}) mostly matches the views
published by a node. It has the
following gaps:
* The {{METRICS}} view ({{SYS.METRICS}}, created by the SQL view metric
exporter, enabled by default) is published
through SQL, JMX and {{control.sh --system-view}}, but the page doesn't
describe it. It is mentioned only on the
Metrics System page.
* {{METASTORAGE}} exists only on server nodes with persistence enabled, and
{{SNAPSHOT}} exists only on server nodes.
On other nodes a query fails with "table not found". The page doesn't mention
this.
* Wrong column types:
** {{CACHES.REBALANCE_BATCHES_PREFETCH_COUNT}} is {{long}}, not {{int}}.
** 13 columns are documented as {{UUID}}, but they are {{IgniteUuid}}: their
value isn't a UUID string, and
{{SYS.VIEW_COLUMNS}} also reports {{IgniteUuid}}:
*** {{TASKS.ID}}, {{TASKS.SESSION_ID}}, {{TASKS.JOB_ID}};
*** {{JOBS.ID}}, {{JOBS.SESSION_ID}};
*** {{SERVICES.SERVICE_ID}};
*** {{TRANSACTIONS.XID}}, {{TRANSACTIONS.ORIGINATING_XID}};
*** {{CACHE_EXPLICIT_LOCKS.XID}};
*** {{CACHE_LOCKS.XID}}, {{CACHE_LOCKS.ORIGINATING_XID}};
*** {{DS_QUEUES.ID}}, {{DS_SETS.ID}}.
** {{SERVICES.TOPOLOGY_SNAPSHOT}} is {{Map<UUID, Integer>}}, not {{string}}.
* Empty descriptions: {{CACHES.IS_MANAGEMENT_ENABLED}},
{{CACHES.IS_STATISTICS_ENABLED}},
{{SCAN_QUERIES.TASK_NAME}}.
* Javadoc markup in the text: {{CACHES.IS_STORE_KEEP_BINARY}} contains
{{\{@link CacheStore\}}}.
* The {{CACHES}}, {{SERVICES}}, {{SQL_QUERIES_HISTORY}},
{{CACHE_GROUP_PAGE_LISTS}} and {{DATA_REGION_PAGE_LISTS}}
sections have no introductory sentence.
h3. Changes
Documentation only, {{system-views.adoc}}:
* The {{METRICS}} section is added, with links to the SQL View and Histograms
sections of the Metrics System page.
* Availability notes are added to {{METASTORAGE}} and {{SNAPSHOT}}.
* The column types listed above are fixed.
* The empty descriptions are filled in and the javadoc markup is removed.
* Introductory sentences are added to the 5 sections.
* A typo is fixed: "Ttransaction" in {{TRANSACTIONS.ORIGINATING_XID}}.
The column order and the mixed notation of types ({{string}}/{{long}} vs
{{VARCHAR}}/{{BIGINT}}) are unchanged and can
be unified separately.
h3. How it was verified
A cluster of 2 server nodes and 1 client node was started from the current
master with most optional features
enabled: persistence, encryption, SSL, services, H2 and Calcite, indexes, a
thin client, data structures, continuous
and scan queries, transactions, locks and snapshots.
On every node the following were dumped and compared with the tables of the
page:
* all system views, with the attributes of their walkers;
* {{SYS.VIEWS}} and {{SYS.VIEW_COLUMNS}} through both SQL engines;
* the JMX MBeans of the views.
Results:
* JMX export ({{JmxSystemViewExporterSpi}}) matches the registered views 1:1
(server: 53, client: 51).
* H2 and Calcite expose the same views and columns.
* Before the patch: 1 view is missing from the page, plus 19 column-level
discrepancies (types, empty descriptions,
markup).
* After the patch: no discrepancies. All 53 views and their columns match the
page.
> Docs: describe all index operation metrics, ZooKeeper discovery metrics and
> server-only metric registers
> --------------------------------------------------------------------------------------------------------
>
> Key: IGNITE-29236
> URL: https://issues.apache.org/jira/browse/IGNITE-29236
> Project: Ignite
> Issue Type: Task
> Components: documentation
> Reporter: Ilya Shishkov
> Assignee: Ilya Shishkov
> Priority: Minor
>
> h3. Problem
> The
> [Metrics|https://ignite.apache.org/docs/ignite2/latest/monitoring-metrics/new-metrics]
> page
> ({{docs/_docs/monitoring-metrics/new-metrics.adoc}}) mostly matches the
> metrics published by a node. It doesn't match
> in the following places:
> * *Sorted Indexes Operations*
> ({{index.\{schema_name\}.\{table_name\}.\{index_name\}}}): the table contains
> only
> the templates {{\{opType\}Count}} and {{\{opType\}Time}}, and the text
> names 5 of the 11 operations. A node actually
> publishes 22 metrics: {{AskNeighbor}}, {{Insert}},
> {{LockBackAndRmvFromLeaf}}, {{LockBackAndTail}}, {{LockTail}},
> {{LockTailExact}}, {{LockTailForward}}, {{RemoveFromLeaf}},
> {{RemoveRangeFromLeaf}}, {{Replace}}, {{Search}}, each
> with the {{Count}} and {{Time}} suffixes.
> * *Discovery IO* ({{io.discovery}}): only {{TcpDiscoverySpi}} metrics are
> described. With {{ZookeeperDiscoverySpi}}
> the register contains a different set: {{CommunicationErrors}},
> {{Coordinator}}, {{CurrentTopologyVersion}},
> {{FailedNodes}}, {{JoinedNodes}}, {{LeftNodes}}. {{CommunicationErrors}}
> isn't documented at all, and the counters
> are {{long}}, not {{integer}} as for TCP discovery
> ({{ZookeeperDiscoveryStatistics}},
> {{ZookeeperDiscoverySpi#onContextInitialized0}}).
> * The page doesn't say which metrics and registers exist on server nodes only:
> ** {{cacheGroups.\{group_name\}}}: {{InMemoryIndexPages}} and
> {{TotalAllocatedPages}} are not registered on client
> nodes (no data region, {{CacheGroupMetricsImpl}}).
> ** {{rest.client}}: on client nodes the register exists only with
> {{IGNITE_REST_START_ON_CLIENT=true}}
> ({{GridRestProcessor}}).
> ** {{snapshot}}, {{snapshot.incremental}}, {{snapshot-restore}},
> {{snapshot-check.\{snapshot_name\}}}: server nodes
> only ({{IgniteSnapshotManager#start0}} exits on client nodes).
> ** {{index.*}}: only on the nodes that store the index data.
> h3. Changes
> Documentation only, {{new-metrics.adoc}}:
> * The {{\{opType\}}} template is replaced by the explicit list of 22 metrics.
> The note that the set of operations
> depends on the index implementation is kept.
> * A "ZooKeeper Discovery" subsection is added to "Discovery IO" with the
> metrics of {{ZookeeperDiscoverySpi}}.
> * Notes about server-only metrics and registers are added to "Cache Groups",
> "Sorted Indexes Operations",
> "Ignite REST Client Connector" and "Snapshot".
> h3. How it was verified
> A cluster of 2 server nodes and 1 client node was started from the current
> master with most optional features
> enabled, so that conditional metrics are registered:
> * persistence with CDC and an in-memory region;
> * cache statistics, a near cache and an encrypted cache group with key
> rotation;
> * SSL, a user executor, a service with statistics, H2 and Calcite;
> * secondary indexes, a thin client and data structures;
> * full and incremental snapshots with check and restore;
> * a suppressed failure.
> All metric registers of every node, including registers that exist only while
> an operation runs, were dumped and
> compared with the tables of the page.
> Results:
> * JMX export ({{JmxMetricExporterSpi}}) matches the registers 1:1 on every
> node (srv: 2304/2306 metrics, client: 2177).
> * Before the patch: 22 metrics are missing from the page, and 2 template rows
> don't exist at runtime.
> * After the patch: no discrepancies.
> * Not reproducible in this setup and checked against the code:
> {{ConflictResolver*}} (needs a conflict resolver; the
> page already notes the condition) and the ZooKeeper discovery metrics.
> * The {{cdc}} register described in {{persistence/change-data-capture.adoc}}
> matches the code (13/13).
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
