Monitor Performance for AlwaysOn Availability Groups

 

The performance aspect of AlwaysOn Availability Groups is crucial to maintaining the service-level agreement (SLA) for your mission-critical databases. Understanding how AlwaysOn Availability Groups ship logs to secondary replicas can help you estimate the recovery time objective (RTO) and recovery point objective (RPO) of your AlwaysOn implementation and identify bottlenecks in poorly performing availability groups or replicas. This topic describes the synchronization process, shows you how to calculate some of the key metrics, and gives you the links to some of the common performance troubleshooting scenarios.

The following topics are covered:

  • Data Synchronization Process

  • Flow Control Gates

  • Estimating Failover Time (RTO)

  • Estimating Potential Data Loss (RPO)

  • Monitoring for RTO and RPO

  • Performance Troubleshooting Scenarios

  • Useful Extended Events

Data Synchronization Process

To estimate the time to full synchronization and to identify the bottleneck, you need to understand the synchronization process. Performance bottleneck can be anywhere in the process, and locating the bottleneck can help you dig deeper into the underlying issues. The figure and table below illustrate the data synchronization process:

AlwaysOn Availability Group Data Synchronization

Sequence Step Description Comments Useful Metrics
1 Log Generation Log data is flushed to disk. This log must be replicated to the secondary replicas. The log records enter the send queue. SQL Server:Database > Log bytes flushed\sec
2 Capture Logs for each database is captured and sent to the corresponding partner queue (one per database-replica pair). This capture process runs continuously as long as the availability replica is connected and data movement is not suspended for any reason, and the database-replica pair is shown to be either Synchronizing or Synchronized. If the capture process is not able to scan and enqueue the messages fast enough, the log send queue builds up. SQL Server:Availability Replica > Bytes Sent to Replica\sec, which is an aggregation of the sum of all database messages queued for that availability replica.

 log_send_queue_size (KB) and log_bytes_send_rate (KB/sec) on the primary replica.
3 Send The messages in each database-replica queue is dequeued and sent across the wire to the respective secondary replica. SQL Server:Availability Replica > Bytes sent to transport\sec and SQL Server:Availability Replica > Message Acknowledgement Time (ms)
4 Receive and Cache Each secondary replica receives and caches the message. Performance counter SQL Server:Availabiltiy Replica > Log Bytes Received/sec
5 Harden Log is flushed on the secondary replica for hardening. After the log flush, an acknowledgement is sent back to the primary replica.

Once the log is hardened, data loss is avoided.
Performance counter SQL Server:Database > Log Bytes Flushed/sec

Wait type HADR_LOGCAPTURE_SYNC
6 Redo Redo the flushed pages on the secondary replica. Pages are kept in the redo queue as they wait to be redone. SQL Server:Database Replica > Redone Bytes/sec

 redo_queue_size (KB) and redo_rate.

Wait type REDO_SYNC

Flow Control Gates

AlwaysOn Availability Groups is designed with flow control gates on the primary replica to avoid excessive resource consumption, such as network and memory resources, on all availability replicas. These flow control gates do not affect the synchronization health state of the availability replicas, but they can affect the overall performance of your availability databases, including RPO.

After the logs have been captured on the primary replica, they are subject to two levels of flow controls, as shown in the table below.

Level Number of Gates Number of messages Useful Metrics
Transport 1 per availabiltiy replica 8192 Extended event database_transport_flow_control_action
Database 1 per availability database 11200 (x64)

1600 (x86)
DBMIRROR_SEND

Extended event hadron_database_flow_control_action

Once the message threshold of either gate is reached, log messages are no longer sent to a specific replica or for a specific database. These can be sent once acknowledgement messages are received for the sent messages to bring the number of sent messages below the threshold.

In addition to the flow control gates, there is another factor that can prevent the log messages from being sent. The synchronization of replicas ensures that the messages are sent and applied in the order of the log sequence numbers (LSN). Before a log message is sent, its LSN also checked against the lowest acknowledged LSN number to make sure that it is less than one of thresholds (depending on the message type). If the gap between the two LSN numbers is larger than the threshold, the messages are not sent. Once the gap is below the threshold again, the messages are sent.

Two useful performance counters, SQL Server:Availability Replica > Flow control/sec and SQL Server:Availability Replica > Flow Control Time (ms/sec), show you, within the last second, how many times flow control was activated and how much time was spent waiting on flow control. Higher wait time on the flow control translate to higher RPO. For more information on the types of issues that can cause a high wait time on the flow control, see Troubleshoot: Availability Group Exceeded RPO.

Estimating Failover Time (RTO)

The RTO in your SLA depends on the failover time of your AlwaysOn implementation at any given time, which can be expressed in the following formula.

AlwaysOn Availability Groups RTO Calculation

Important


If an availability group contains more than one availability database, then the availability database with the highest Tfailover becomes the limiting value for RTO compliance.

The failure detection time, Tdetection, is the time it takes for the system to detect the failure. This time depends on cluster-level settings and not on the individual availability replicas. Depending on the automatic failover condition that is configured, a failover can be triggered as an instant response to a critical SQL Server internal error, such as orphaned spinlocks. In this case, detection can be as fast as the sp_server_diagnostics (Transact-SQL) error report is sent to the WSFC cluster (the default interval is 1/3 of the health check timeout). A failover can also be triggered because of a timeout, such as the cluster health check timeout has expired (30 seconds by default) or the lease between the resource DLL and the SQL Server instance has expired (20 seconds by default). In this case, detection time is as long as the timeout interval. For more information, see Flexible Failover Policy for Automatic Failover of an Availability Group (SQL Server).

The only thing that the secondary replica needs to do to become ready for a failover is for the redo to catch up to the end of log. The redo time, Tredo, is calculated using the following formula:

AlwaysOn Availability Groups Redo Time Calculation

where redo_queue is the value in redo_queue_size and redo_rate is the value in redo_rate.

The failover overhead time, Toverhead, includes the time it takes to fail over the WSFC cluster and to bring the databases online. This time is usually short and constant.

Estimating Potential Data Loss (RPO)

The RPO in your SLA depends on the possible data loss of your AlwaysOn implementation at any given time. This possible data loss can be expressed in the following formula:

AlwaysOn Availability Groups RPO Calculation

where log_send_queue is the value of log_send_queue_size and log generation rate is the value of SQL Server:Database > Log Bytes Flushed/sec.

Warning


If an availability group contains more than one availability database, then the availability database with the highest Tdata_loss becomes the limiting value for RPO compliance.

The log send queue represents all the data that can be lost from a catastrophic failure. At first glance, it is curious that the log generation rate is used instead of the log send rate (see log_send_rate). However, remember that using the log send rate only gives you the time to synchronize, while RPO measures data loss based on how fast it is generated, not on how fast it is synchronized.

A simpler way to estimate Tdata_loss is to use last_commit_time. The DMV on the primary replica reports this value for all replicas. You can calculate the difference between the value for the primary replica and the value for the secondary replica to estimate how fast the log on the secondary replica is catching up to the primary replica. As stated previously, this calculation does not tell you the potential data loss based on how fast the log is generated, but it should be a close approximation.

Monitoring for RTO and RPO

This section demonstrates how to monitor your availability groups for RTO and RPO metrics. This demonstration is similar to the GUI tutorial given in The AlwaysOn Health Model Part 2 -- Extending the Health Model.

Elements of the failover time and potential data loss calculations in Estimating Failover Time (RTO) and Estimating Potential Data Loss (RPO) are conveniently provided as performance metrics in the policy management facet Database Replica State (see View the Policy-Based Management Facets on a SQL Server Object). You can monitor these two metrics on a schedule and be alerted when the metrics exceed your RTO and RPO, respectively.

The demonstrated scripts create two system policies that are run on their respective schedules, with the following characteristics:

  • An RTO policy that fails when estimated failover time exceeds 10 minutes, evaluated every 5 minutes

  • An RPO policy that fails when estimated data loss exceeds 1 hour, evaluated every 30 minutes

  • The two policies have identical configuration on all availability replicas

  • Policies are evaluated on all servers, but only on the availability groups for which the local availability replica is the primary replica. If the local availability replica is not the primary replica, the policies are not evaluated.

  • Policy failures are conveniently displayed in the AlwaysOn Dashboard when you view it on the primary replica.

Follow the instructions below on all server instances that participate in the availability group:

  1. Start the SQL Server Agent service if it is not already started.

  2. In SQL Server Management Studio, from the Tools menu, click Options.

  3. In the SQL Server AlwaysOn tab, select Enable user-defined AlwaysOn policy and click OK.

    This setting enables you to display properly configured custom policies in the AlwaysOn Dashboard.

  4. Create a Policy-Based Management condition using the following specifications:

    • Name: RTO

    • Facet: Database Replica State

    • Field: Add(@EstimatedRecoveryTime, 60)

    • Operator: <=

    • Value: 600

    This condition fails when potential failover time exceeds 10 minutes, including a 60 second overhead for both failure detection and failover.

  5. Create a second Policy-Based Management condition using the following specifications:

    • Name: RPO

    • Facet: Database Replica State

    • Field: @EstimatedDataLoss

    • Operator: <=

    • Value: 3600

    This condition fails when potential data loss exceeds 1 hour.

  6. Create a third Policy-Based Management condition using the following specifications:

    • Name: IsPrimaryReplica

    • Facet: Availability Group

    • Field: @LocalReplicaRole

    • Operator: =

    • Value: Primary

    This condition checks whether the local availability replica for a given availability group is the primary replica.

  7. Create a Create a Policy-Based Management policy using the following specifications:

    • General page:

      • Name: CustomSecondaryDatabaseRTO

      • Check condition: RTO

      • Against targets: Every DatabaseReplicaState in IsPrimaryReplica AvailabilityGroup

        This setting ensures that the policy is evaluated only on availability groups for which the local availability replica is the primary replica.

      • Evaluation mode: On schedule

      • Schedule: CollectorSchedule_Every_5min

      • Enabled: selected

    • Description page:

      • Category: Availability database warnings

        This setting enables the policy evaluation results to be displayed in the AlwaysOn Dashboard.

      • Description: The current replica has an RTO that exceeds 10 minutes, assuming an overhead of 1 minute for discovery and failover. You should investigate performance issues on the respective server instance immediately.

      • Text to display: RTO Exceeded!

  8. Create a second Create a Policy-Based Management policy using the following specifications:

    • General page:

      • Name: CustomAvailabilityDatabaseRPO

      • Check condition: RPO

      • Against targets: Every DatabaseReplicaState in IsPrimaryReplica AvailabilityGroup

      • Evaluation mode: On schedule

      • Schedule: CollectorSchedule_Every_30min

      • Enabled: selected

    • Description page:

      • Category: Availability database warnings

      • Description: The availability database has exceeded your RPO of 1 hour. You should investigate performance issues on the availability replicas immediately.

      • Text to display: RPO Exceeded!

When you are done, two new SQL Server Agent jobs are created, one for each of the policy evaluation schedule. These jobs should have names that begin with syspolicy_check_schedule.

You can view the job history to inspect evaluation results. Evaluation failures will also be recorded in the Windows application log (in the Event Viewer) with the Event ID 34052. You can also configure SQL Server Agent to send alerts on the policy failures. For more information, see Configure Alerts to Notify Policy Administrators of Policy Failures.

Performance Troubleshooting Scenarios

The table below lists the common performance-related troubleshooting scenarios.

Scenario Description
Troubleshoot: Availability Group Exceeded RTO After an automatic failover or a planned manual failover without data loss, the failover time exceeds your RTO. Or, when you estimate the failover time of a synchronous-commit secondary replica (such as an automatic failover partner), you find that it exceeds your RTO.
Troubleshoot: Availability Group Exceeded RPO After you perform a forced manual failover, your data loss is more than your RPO. Or, when you calculate the potential data loss of an asynchronous-commit secondary replica, you find that it exceeds your RPO.
Troubleshoot: Changes on the Primary Replica are not Reflected on the Secondary Replica The client application completes an update on the primary replica successfully, but querying the secondary replica shows that the change is not reflected.

Useful Extended Events

The following extended events are useful when troubleshooting replicas in the Synchronizing state.

Event Name Category Channel Availability Replica
redo_caught_up transactions Debug Secondary
redo_worker_entry transactions Debug Secondary
hadr_transport_dump_message alwayson Debug Primary
hadr_worker_pool_task alwayson Debug Primary
hadr_dump_primary_progress alwayson Debug Primary
hadr_dump_log_progress alwayson Debug Primary
hadr_undo_of_redo_log_scan alwayson Analytic Secondary

See Also