サポートと今すぐチャット
サポートとのチャット

Foglight for SQL Server (Cartridge) 7.3.0 - User Guide

Introduction to this Guide Using Foglight for SQL Server
Viewing the Databases Dashboard Overview Dashboard Advisories Dashboard Monitoring Data Replication Monitoring SQL Performance Reviewing Memory Usage Reviewing the Instance Activity Reviewing Database Usage Reviewing the Services Using the HADR Drilldown Using the Logs Drilldown Reviewing Configuration Settings Viewing User-defined Performance Counters and Collections
Monitoring Business Intelligence Services Administering Foglight for SQL Server
Configuration Settings Managing Foglight for SQL Server Agent Settings Reviewing Foglight for SQL Server Alarms Generating Reports Monitoring SQL Server instances on VMware servers
Reference
SQL Performance Investigator Metrics
Active Time All SQL Agents CPU Usage All SQL Agents Resident Memory Usage Availability Average Physical I/O Operations Average SQL Response Time Backup Recovery Wait Blocked Lock Requests Checkpoint Pages CLR Wait CPU Usage CPU Wait Cursor Synchronization Wait Database Replication Wait Deferred Task Worker Wait Degree of Parallelism Disk Utilization DTC CPU Usage DTC Resident Memory Usage Distributed Transaction Wait Executions Ended Executions Started External Procedures Wait Full Scans Full Text Search CPU Usage Full Text Search Resident Memory Usage Full Text Search Wait Free Buffer Wait Hosted Components Wait IO Bulk Load Wait IO Completion Wait IO Data Page Wait IO Wait Latch Buffer Wait Latch Wait Latch Savepoint Wait Lazy Writes Lock Wait Lock Bulk Update Wait Lock Exclusive Wait Lock Intent Wait Lock Requests Lock Schema Wait Lock Shared Wait Lock Update Wait Lock Wait Log Buffer Wait Log Flushes Log Other Wait Log Synchronization Wait Log Wait Log Write Wait Memory Wait Network IO Wait Network IPC Wait Network Mirror Wait Network Wait Non SQL Server CPU Usage Non SQL Resident Memory Usage OLAP CPU Usage OLAP Resident Memory Usage OLEDB Provider Full Text Wait Other CPU Usage Other Miscellaneous Wait Other Wait Overall CPU Page Life Expectancy Page Splits Parallel Coordination Wait Physical I/O Physical Memory Used Physical Page Reads Physical Page Writes Probe Scans Plan Cache Hit Rate Range Scans Rec Ended Duration Remote Provider Wait Run Queue Length Samples Service Broker Wait Session Logons Session Logoffs SQL Agent CPU Usage SQL Agent Resident Memory Usage SQL Executions SQL Mail CPU Usage SQL Mail Resident Memory Usage SQL Recompilations SQL Response Time SQL Server Background CPU Usage SQL Server Cache Memory SQL Server Connections Memory SQL Server Connections Summary SQL Server Foreground CPU Usage SQL Server Resident Memory Usage SQL Server Swap Memory Usage Synchronous Task Wait Table Lock Escalation Target Instance Memory Total CPU Usage Total Instance Memory Virtual Memory Used
Rules Collections and Metrics
SQL Server Agent's Default Collections Access Methods Agent Alert List Agent Job List Always On Availability Groups Backup Locations Blocking History Blocking List Buffer Cache List Buffer Manager CLR Assemblies Cluster Summary Configuration Database Index Density Vectors Database Index Details Database Index Fragmentation Info Database Index Histogram Database Index List Database Information Database Properties Database Sessions (Session List) Database Summary Database Tables List Databases Deadlock DTC Information Error Log Error Log List Error Log Scan File Groups File Data Flow Statistics File Groups Files Files Drive Total Files Instance Summary Full Text Catalog InMemory OLTP (XTP) Instance Wait Categories Instance Wait Events Job Messages Latches and Locks Lock Statistics Locks List Log Shipping Log Shipping Error Logical Disks Long Running Session Memory Manager Mirroring Mirroring Performance Counters Missing Indexes Plan Cache Distribution Plan Cache List Replication Agents Replication Agent Session Actions Replication Agent Session Merge Articles Replication Agent Sessions Replication Agent Sessions by Type Replication Available Replication Publications Replication Subscriptions Reporting Services Resource Pool Session Data Session Trace SQL PI Instance Statistics SQL Server Connections Summary SQL Server Global Variables SQL Server Host SQL Server Load SQL Server Services SQL Server Throughput SQL Server Version Info SSIS OS Statistics SSIS Summary Statistics Top SQLs Top SQL Batch Text Top SQL Long Text Top SQL Plan Top SQL Short Text Top SQL Summary Traced SQL PA Usability User-defined Performance Counters User-defined Queries Virtualization XTP Session Transactions Statistics
Glossary
A
Access methods Adhoc SQL Plans Alarm Alert Allow updates Anonymous subscription Authentication AutoClose AutoGrow Automatic Discovery AutoShrink B Batch BCP (Bulk Copy Program) Blocking Books Online Bound trees Buffer Buffer cache Buffer pool Bulk copy Bulkinsert Bulk load C Cache CAL Calibration Cardinality Cartridge Chart Checkpoint process Client network utility CLR Compile Connect Connection Connectivity software CPU Usage Cursors D Data access components Data file Data storage engine Database Database object DataFlow DBCC DBID DBO Deadlock Disk queue length Disk transfer time Disk utilization DiskPerf Distributing instance Distributor DMO Drilldown DTC DTS E Error log Event alert Execution contexts Extended stored procedures Extent External procedures F File Filegroup File cache Foglight Agent Manager Foglight Management Server Foreign key Forwarded records Free buffers Free list Free pages Free potential Free space Freespace scans Full text search G GAM Ghosted records Growth increment GUID H Hash buckets Hashing Heap Hit rate Hit ratio Host name Host process I I/O Index Indid Integrated security Intent Locks J Job K Kernel memory Kernel mode Kill L Latch Lazy writer Licensing Lightweight pooling Lock Lock area Lock escalation Lock mode Log Log cache Log writer Logical I/O LRU LSN M Master MaxSize MDAC Metric Misc.normalized trees Model Monitor page file N Named pipes Net library NIC Null O OBID Object plans OLAP OLAP service OLTP Optimizer Optimizer cache osql P Page life expectancy Paging Panel Parse Parser Per seat licensing Per server licensing Performance alert Physical I/O Physical read Physical write PID Pinned Plan Plan cache Potential growth Prepared SQL plans Primary key Privileged mode Procedure cache Procedure plans Process Profiler Publication database Publisher Publisher databases Publishing server Pull subscription Pulse Push subscription Q Query plan R RAID Random I/O Read ahead Recompile Referential integrity Relational data engine Replication procedure plans Role Rollback S sa Schema locks Sequential I/O Session Severity SGAM Shared locks Show advanced options SMP Sort, Hash, Index Area SPID Spike Spinner SQL Agent Mail SQL Mail SQL Plans SQL Server Agent SQL Server authentication SQL Server books online Standard deviation Stolen pages Stored procedure Support service SYSADMIN role T TDS TempDB Temporary tables and table variables Threshold Torn page detection Transaction Trigger Trigger plans Truncate Trusted U UMS Unused space User connection area User mode V Virtual log file VLF W Waitfor Windows authentication mode Working set
Technical support resources

Reviewing Foglight for SQL Server Alarms

Alarms are the warnings that Foglight for SQL Server raises when a metric falls outside its “normal” range of values, which is defined by setting thresholds and severities for the metric within the Metric editor. A new alarm is raised whenever the severity for a metric changes. When the severity returns to normal, the alarm is canceled.

Several alarms can be investigated using the home page’s Sessions pane and network packet flows, as follows:

The Response Time alarm becomes active when the execution time of the Response Time SQL exceeds a threshold.

Response time is the full time (in milliseconds) it has taken a query (select 1, by default) to get from the application to SQL Server and back. Every time a real-time sampled interval starts (by default: 20 seconds), a query is sent and its response time value is displayed. Any value higher than 20 ms may indicate a performance issue, which should then be investigated to detect its source, identify the possible bottleneck, and take correcting measures.

The Response Time SQL is a user-defined Transact-SQL batch that can be used for indicating application response time.

To change the query used for determining response time, edit the Response time section in the Agent properties.

The Packet Errors alarm becomes active when the rate at which SQL Server is encountering network packet errors exceeds a threshold.

When this alarm is fired, investigate what is causing the packet errors on the network.

Alarms Displayed in the SQL Processes Panel

Several alarms can be investigated using the SQL Processes panel, as follows:

The Blocking alarm is raised when at least one SQL Server session is waiting on a lock held by another session. The waiting user is said to be “blocked” by the one holding the lock, and waits until one of the following scenarios realizes:

Following any of these scenarios, the blocked command is cancelled.

Excessive blocking can be a major cause of poor application performance, as users of an application often do not realize they are waiting on a lock held by another user. From their point of view, it often seems like their application has stopped responding.

When this alarm occurs, look at:

The Blocking panel on the SQL Activity drilldown, to see who is blocking whom, and what resources are involved (for example, database and table names). In the case of multiple blocks, where blocked sessions are also blocking others, this panel displays the sessions at the top of the tree (those that do not have a “parent” in the tree). These sessions, by being at the head of the blocking chain, are the root blockers. This panel also shows how many sessions were blocked over time.
The Sessions panel on the SQL Activity drilldown, to view the most recent SQL for the sessions involved in the blocking. This can help track down sub-optimal SQL that may contribute to the locking problem.
The Locks panel on the SQL Activity drilldown, to view all locks in the system. This panel displays all SQL Server locks currently granted or requested.

The Deadlocks alarm becomes active when the number of deadlocks per second exceeds a threshold.

A Deadlock occurs when multiple SQL Server sessions request conflicting locks in such a way that two locks are blocked by one another.

In this scenario, Session 1 is waiting on a lock held by Session 2, and Session 2 is waiting on one held by Session 1. In this example, to prevent both sessions from waiting forever, SQL Server enables only one session to continue and aborts the other session, by killing its connections and rolling back its updates. The user whose session is rolled back often receives an error message.

This scenario is relatively simple to resolve. However, deadlocks can involve many more sessions, which make it extremely difficult to track down the root cause of all the trouble. When this alarm occurs, review the Locks panel on the SQL Activity drilldown, to view the Deadlocks/sec counter on the Lock Types chart. This counter displays the frequency of deadlocks in the currently monitored system.

The Recompiles alarm becomes active when the ratio between the number of recompiles and the total number of compiles exceeds a threshold. This highlights when more than a certain percentage of compilations are due to run-time recompiles.

A compile can be a relatively time-consuming, CPU-intensive process, especially when the query or stored procedure is large or complex. For this reason, SQL Server stores execution plans in the Plan Cache, so that they would satisfy future I/O requests, thereby saving the need for physical reads from the disk.

A Recompile takes place when SQL Server Optimizer determines that the execution plan for a stored procedure that is currently executing may no longer be the optimal plan. SQL Server pauses the query execution and compiles the stored procedure again. This not only slows down the process that is executing the procedure, but adds extra CPU load on the server.

When many recompiles take place, the SQL Server’s CPU can become overloaded, thereby slowing down everything running on that computer. Therefore, it is advisable to keep the number of recompiles as low as possible.

A stored procedure can be recompiled by the SQL Server for various reasons, the most common of which are as follows:

The use of the WITH RECOMPILE clause in the CREATE PROCEDURE or EXECUTE statement.
Running sp_recompile against any table referenced in the stored procedure.
On the SQL Activity drilldown, click SQL Instance Summary and view the Call Rates graph to determine if this is a persistent problem. Consistently high percentage of the Re-Compiles rate within the total Compiles rate requires further investigation.

It is advisable to review the code of each of the stored procedures. Consider changing the stored procedures to remove coding practices that can cause recompiles.

The Error Log alarm becomes active when Foglight for SQL Server, which scans the error logs at the configured frequency, detects messages that could indicate potential problems in the SQL Sever error log or SQL Agent error log.

When this alarm occurs, look at the Error Log drilldown to view the errors.

To select which error logs generated by the SQL Server database are displayed in the Error Log drilldown, use the Error Log Scanning view in the Databases Administration dashboard. For details, see Defining Error Log Filtering .

Alarms Displayed in the SQL Memory Panel

Several alarms can be investigated using the SQL Memory panel, as follows:

The Buffer Cache Hit Ratio alarm becomes active when the ratio of physical reads to logical reads falls below a threshold.

SQL Server holds recently accessed database pages in a memory area called the Buffer Cache. If an SQL process needs to access a database page, finding this page in the buffer cache saves SQL Server the need to read the page from disk, thereby significantly reducing the amount of disk I/O and, in most probability, speeding up queries.

Buffer Cache Hit Ratio is the ratio of logical reads to physical reads. It indicates the percentage of database page I/O requests that were satisfied from the Buffer Cache and therefore did not have to perform disk reads. This ratio measures how efficiently SQL Server is using the memory allocated to its buffer cache.

A low Buffer Cache hit rate indicates that SQL Server is finding fewer pages already in memory, and therefore has to perform more disk reads. This is often caused by either lack of SQL Server memory or use of inefficient SQL queries, which are accessing a very large number of pages in a non-sequential manner. The best figure varies from one application to another, but ideally this ratio should be above 90%.

View the Buffer Cache panel on the Memory drilldown to see the largest objects in the Buffer Cache.
Use the Sessions panel on the SQL Activity drilldown to identify inefficient SQL queries. Look for currently active sessions that are generating a lot of I/O. Such sessions can be traced also by using the Session Trace pane on the Sessions panel.

The Free Buffers alarm becomes active when the amount of SQL Server memory available for immediate reuse drops below a threshold.

The Lazy Writer Process periodically scans all SQL Server caches, and maintains a list of “free” pages that are available for immediate reuse.

When SQL Server needs a free memory page (for example, when reading a database page from disk into the buffer cache), and no free pages are immediately available, the connection needing the free page must wait while SQL Server makes buffers available. This results in slower performance. In the worst case, the connection has to wait while SQL Server writes a modified page out to disk, in order to make a free buffer.

This alarm does not always indicate a problem with SQL Server, especially if the alarm is not active for more than 10-20 seconds.

On the Memory drilldown, select the Summary panel. Check the Memory Areas chart to determine the amount of time for which the Free List has been very low. This alarm normally only indicates a problem if the Free List has been very low for more than a few minutes.
Check the Configuration drilldown to view the currently set recovery interval parameter. Setting this too high can cause the Checkpoint process to run infrequently, which can in turn cause the Lazy Writer process to perform the majority of the I/O that the Checkpoint process normally does. This can keep the Lazy Writer so busy that it does not maintain the Free List efficiently.

The Page Life Expectancy alarm becomes active when the page life expectancy falls below a threshold.

Page life expectancy is the length of time in seconds that a database page will stay in the buffer cache without being accessed, before it is flushed out. Microsoft recommends keeping this value greater than five minutes (300 seconds).

Values smaller than 300 indicate that pages are being flushed out of the cache within a small period of time. The resulting lack of pages in the buffer cache requires SQL Server to carry out more disk reads, thereby degrading its performance.

This alarm is often invoked by memory shortage (either memory on the system or memory configured for SQL Server’s use) or use of inefficient SQL queries, which are accessing a very large number of pages in a non-sequential manner.

Identify inefficient SQL queries, by using the Sessions panel on the SQL Activity drilldown. Look for sessions that are currently active and generating a large number of I/O operations. This task can be carried out by using the Session Trace pane on the Sessions panel. For details, see Reviewing Session Trace details .

The Procedure Cache Hit Ratio alarm is raised when the ratio between the number of times SQL Server looks for a plan in the plan Cache, and the number of times it finds the requested plan in the plan Cache, falls below a threshold.

A low plan cache hit rate indicates that SQL Server is finding fewer of the query execution plans it needs already in memory, and therefore has to perform more compiles. These extra compilations degrade SQL Server performance by causing extra CPU load.

To prevent this alarm being caused by adhoc SQL requests (which often produce non-reusable execution plans), Foglight for SQL Server removes adhoc plan statistics from this metric.

Check the Call Rates chart on the SQL Instance Summary panel of the SQL Activity drilldown for a high number of Re-Compiles. Follow the suggestions listed under the Recompiles alarm.
On the Memory drilldown, select the Plan Cache panel and then use the Hit Rate and Use Rate counters on the Object Types chart to identify which types of objects are causing the problem.

Alarms Displayed in the Background Processes Panel

Several alarms can be investigated using the Background Processes panel, as detailed in the following sections:

The Cluster Server Down alarm is raised when Foglight for SQL Server detects that at least one cluster node (server) is not currently running as part of the cluster.

When this alarm is active, take these measures:

Check the Cluster Services panel of the Support Services drilldown to determine which cluster node is unavailable.

The log shipping alarm is invoked when the out-of-sync threshold has been exceeded for any log-shipping pairs, that is, when the time between the last backup of the source database and the restore in the target database has exceeded the allowed length specified. Because the log shipping operation comprises copy, backup and restore phases, the log shipping alarm can be invoked as a result of a failure to:

When the alarm is raised, it is accompanied by a prompt to go to the Log Shipping panel of the Support Services drilldown. For details, see Tracking the Status of the Mirroring Operation .

The Non-preferred Cluster Node alarm is raised when Foglight for SQL Server detects that SQL Server is not running on its preferred cluster node.

This alarm can be raised only when the currently connected SQL Server is running as part of a Microsoft Cluster Server (MSCS).

In a Windows cluster, each SQL Server instance belongs to a single cluster group. Preferred cluster nodes are allocated to each group. Normally, the group should run on these preferred cluster nodes.

The SQL Agent Alerts alarm is activated when Foglight for SQL Server detects that at least one SQL Agent alert has occurred in the last few minutes.

On the Support Services drilldown, view the SQL Agent Alerts page to determine which alerts have occurred recently. This page displays the last occurrence time for each alert, and the alert history for the specified time range.

The SQL Agent Job Failure alarm is activated when Foglight for SQL Server detects that at least one SQL Agent job has failed in the last few minutes.

View the SQL Agent Jobs panel, on the Support Services drilldown, to determine which jobs have failed recently. Double-click any job to view the messages that it logged during its last run. This page displays the last run time and completion status of each job, as well as a graph showing which jobs ran recently, and the completion status for each run.

The SQL Server I/O Errors Per Second alarm is raised when I/O errors are encountered by SQL Server.

The majority of I/O errors reported by SQL Server are caused by hardware failures, such as disk or controller failures.

The Support Services alarm becomes active when any of SQL Server's supporting services are installed, but not active.

The services currently monitored are detailed in the following list.

The Table Lock Escalation alarm is raised when the number of times page locks are escalated to table locks per second exceeds a threshold.

Lock escalation is not necessarily a problem by itself; however, it can cause concurrency issues and lock conflicts and can be a major contributor to blocking problems.

In certain cases, this alarm may also indicate use of inefficient SQL queries, which leads to a large number of page locks instead of enforcing one table lock, and therefore forces the SQL Server to escalate the lock to a table lock.

関連ドキュメント

The document was helpful.

評価を選択

I easily found the information I needed.

評価を選択