Inputs.win_perf_counters.object sends one counter per interval

[bj2001holt]

Bug report

When using an "inputs.win_perf_counters.object" where you are collecting multiple counters you will only see a single counter collected at each interval. For example, the PhysicalDisk Object collects 8 counters, however it only collects one of them per interval. This removes the ability to correlate performance data.

Relevant telegraf.conf:

[agent]
interval = "5s"
round_interval = true
metric_buffer_limit = 1000
flush_buffer_when_full = true
flush_interval = "5s"
flush_jitter = "0s"

[[outputs.influxdb]]
precision = "s"

[[inputs.win_perf_counters.object]]
ObjectName = "PhysicalDisk"
Instances = ["*"]
Counters = [
"Disk Read Bytes/sec",
"Disk Write Bytes/sec",
"Current Disk Queue Length",
"Disk Reads/sec",
"Disk Writes/sec",
"% Disk Time",
"% Disk Read Time",
"% Disk Write Time",
]
Measurement = "win_diskio"

System info:

Influx and Grafana on CentOS 7, telegraf on Windows Server 2012. Tested with telegraf version 1.3.1 and 1.0.0 beta 3.

Steps to reproduce:

1. Install telegraf on windows system using default win_perf_counters
2. From Influx query
   SELECT Current_Disk_Queue_Length as "C: Current Queue Length", Disk_Reads_persec as "C: Read/Sec", Disk_Writes_persec as "C: Writes/Sec" FROM "win_diskio" WHERE "host" = '' AND "instance" = '0 C:'

Expected behavior:

All counters should be collected with every interval

Actual behavior:

One counter is collected in each interval

[cdjc]

The following diff fixes this problem. That is, it collects counters for each instance, rather then one at a time. The description in the issue is a bit ambiguous: Currently all counters are indeed measured, but, for example, measuring 8 counters on 8 cores will give 64 messages, rather than 8 messages (each with 8 fields). This diff fixes that problem. I'm new to go, so feel free to tidy up. @danielnelson Do I need to sign the CLA or is this snippet small enough not to worry?

diff --git a/plugins/inputs/win_perf_counters/win_perf_counters.go b/plugins/inputs/win_perf_counters/win_perf_counters.go
index 7e3991d..9a9e6b4 100644
--- a/plugins/inputs/win_perf_counters/win_perf_counters.go
+++ b/plugins/inputs/win_perf_counters/win_perf_counters.go
@@ -192,6 +192,14 @@ func (m *Win_PerfCounters) Gather(acc telegraf.Accumulator) error {
        var size uint32 = uint32(unsafe.Sizeof(PDH_FMT_COUNTERVALUE_ITEM_DOUBLE{}))
        var emptyBuf [1]PDH_FMT_COUNTERVALUE_ITEM_DOUBLE // need at least 1 addressable null ptr.
 
+       type InstanceGrouping struct {
+               name       string
+               instance   string
+               objectname string
+       }
+
+       var collectFields = make(map[InstanceGrouping]map[string]interface{})
+
        // For iterate over the known metrics and get the samples.
        for _, metric := range m.itemCache {
                // collect
@@ -231,20 +239,22 @@ func (m *Win_PerfCounters) Gather(acc telegraf.Accumulator) error {
                                        }
 
                                        if add {
-                                               fields := make(map[string]interface{})
                                                tags := make(map[string]string)
                                                if s != "" {
                                                        tags["instance"] = s
                                                }
                                                tags["objectname"] = metric.objectName
-                                               fields[sanitizedChars.Replace(metric.counter)] =
-                                                       float32(c.FmtValue.DoubleValue)
 
                                                measurement := sanitizedChars.Replace(metric.measurement)
                                                if measurement == "" {
                                                        measurement = "win_perf_counters"
                                                }
-                                               acc.AddFields(measurement, fields, tags)
+                                               var instance = InstanceGrouping{measurement, s, metric.objectName}
+
+                                               if collectFields[instance] == nil {
+                                                       collectFields[instance] = make(map[string]interface{})
+                                               }
+                                               collectFields[instance][sanitizedChars.Replace(metric.counter)] = float32(c.FmtValue.DoubleValue)
                                        }
                                }
 
@@ -257,6 +267,14 @@ func (m *Win_PerfCounters) Gather(acc telegraf.Accumulator) error {
                }
        }
 
+       for instance, fields := range collectFields {
+               var tags = map[string]string{
+                       "instance":   instance.instance,
+                       "objectname": instance.objectname,
+               }
+               acc.AddFields(instance.name, fields, tags)
+       }
+
        return nil
 }

[danielnelson]

Thanks @cdjc, you will need to sign the CLA still, can you open this as a pull request as well?

[cdjc]

I can sign the CLA, but I've never done a pull request before, so I'll have to figure it out. I did briefly look at what was required for a pull request, but it just seemed so much more complicated than posting a diff.

[danielnelson]

I can create the PR if your not setup for it. If you have git running though can you create the diff using: git format-patch 7e3991d, this way the change is properly attributed.

[cdjc]

I have the patch file. Should I attach it here, or wait until you have set up the pull request?

[danielnelson]

Attach it here please

[cdjc]

0001-Fix-win_perf_counters-to-collect-counters-per-instan.zip

Here it is (zipped, as .patch aren't allowed to be uploaded).