Edit

Performance metrics

  • Article

Metrics should be collected to understand the performance of your cluster as well as the applications running in it. For Service Fabric clusters, we recommend collecting the following performance counters.

Nodes

For the machines in your cluster, consider collecting the following performance counters to better understand the load on each machine and make appropriate cluster scaling decisions.

Counter Category Counter Name
Logical Disk Logical Disk Free Space
PhysicalDisk(per Disk) Avg. Disk Read Queue Length
PhysicalDisk(per Disk) Avg. Disk Write Queue Length
PhysicalDisk(per Disk) Avg. Disk sec/Read
PhysicalDisk(per Disk) Avg. Disk sec/Write
PhysicalDisk(per Disk) Disk Reads/sec
PhysicalDisk(per Disk) Disk Read Bytes/sec
PhysicalDisk(per Disk) Disk Writes/sec
PhysicalDisk(per Disk) Disk Write Bytes/sec
Memory Available MBytes
PagingFile % Usage
Processor(Total) % Processor Time
Process (per service) % Processor Time
Process (per service) ID Process
Process (per service) Private Bytes
Process (per service) Thread Count
Process (per service) Virtual Bytes
Process (per service) Working Set
Process (per service) Working Set - Private
Network Interface(all-instances) Bytes recd
Network Interface(all-instances) Bytes sent
Network Interface(all-instances) Bytes total
Network Interface(all-instances) Output Queue Length
Network Interface(all-instances) Packets Outbound Discarded
Network Interface(all-instances) Packets Received Discarded
Network Interface(all-instances) Packets Outbound Errors
Network Interface(all-instances) Packets Received Errors

.NET applications and services

Collect the following counters if you are deploying .NET services to your cluster.

Counter Category Counter Name
.NET CLR Memory (per service) Process ID
.NET CLR Memory (per service) # Total committed Bytes
.NET CLR Memory (per service) # Total reserved Bytes
.NET CLR Memory (per service) # Bytes in all Heaps
.NET CLR Memory (per service) Large Object Heap size
.NET CLR Memory (per service) # GC Handles
.NET CLR Memory (per service) # Gen 0 Collections
.NET CLR Memory (per service) # Gen 1 Collections
.NET CLR Memory (per service) # Gen 2 Collections
.NET CLR Memory (per service) % Time in GC

Service Fabric's custom performance counters

Service Fabric generates a substantial amount of custom performance counters. If you have the SDK installed, you can see the comprehensive list on your Windows machine in your Performance Monitor application (Start > Performance Monitor).

In the applications you are deploying to your cluster, if you are using Reliable Actors, add counters from Service Fabric Actor and Service Fabric Actor Method categories (see Service Fabric Reliable Actors Diagnostics).

If you use Reliable Services or Service Remoting, we similarly have Service Fabric Service and Service Fabric Service Method counter categories that you should collect counters from, see monitoring with service remoting and reliable services performance counters.

If you use Reliable Collections, we recommend adding the Avg. Transaction ms/Commit from the Service Fabric Transactional Replicator to collect the average commit latency per transaction metric.

Next steps