fix: filter out snapshot-clone-src-*|replica-vol-* (#10)

* fix: filter out `snapshot-clone-src-*|replica-vol-*`

* chore: ignore .txt files

* unfeat: remove 'node_(read|write)_ops_total'

* fix: remove volume info from `solidfire_node_iscsi_sessions`

* remove: volume_last_sample_read_bytes|volume_read_latency_seconds|volume_last_sample_read_ops|volume_last_sample_write_bytes|volume_write_latency_seconds|volume_write_ops_last_sample

Author: MCBrandenburg

.gitignore

@@ -6,3 +6,4 @@ tmp
 ./solidfire-exporter
 archive.tar.gz
 bin/*
+*.txt

README.md

@@ -121,20 +121,15 @@ docker run  --rm -p 9987:9987 ghcr.io/mjavier2k/solidfire-exporter:v0.6.6
 | solidfire_node_interface_utilization_percentage | gauge | Network interface utilization (in percent) of network interface. |
 | solidfire_node_load | histogram | System load histogram |
 | solidfire_node_read_latency_seconds_total | counter | The total time spent performing read operations since the creation of the cluster. |
-| solidfire_node_read_ops_total | counter | Total read operations to a node. |
 | solidfire_node_samples | gauge | Node stat sample count |
 | solidfire_node_total_memory_bytes | gauge | Total node memory in bytes. |
 | solidfire_node_used_memory_bytes | gauge | Total node memory used in bytes. |
 | solidfire_node_write_latency_seconds_total | counter | The total time spent performing write operations since the creation of the cluster. |
-| solidfire_node_write_ops_total | counter | Total write operations to a node. |
 | solidfire_up | gauge | Whether last scrape against Solidfire API was successful |
 | solidfire_volume_actual_iops | gauge | The current actual IOPS to the volume in the last 500 milliseconds |
 | solidfire_volume_average_iop_size_bytes | gauge | The average size in bytes of recent I/O to the volume in the last 500 milliseconds |
 | solidfire_volume_burst_iops_credit | gauge | The total number of IOP credits available to the user. When volumes are not using up to the configured maxIOPS, credits are accrued. |
 | solidfire_volume_client_queue_depth | gauge | The number of outstanding read and write operations to the volume. |
-| solidfire_volume_last_sample_read_bytes | gauge | The total number of bytes read from the volume during the last sample period. |
-| solidfire_volume_last_sample_read_ops | gauge | The total number of read operations during the last sample period |
-| solidfire_volume_last_sample_write_bytes | gauge | The total number of bytes written to the volume during the last sample period. |
 | solidfire_volume_latency_seconds | gauge | The average time, in seconds, to complete operations to the volume in the last 500 milliseconds. A '0' (zero) value means there is no I/O to the volume. |
 | solidfire_volume_non_zero_blocks | gauge | The total number of 4KiB blocks that contain data after the last garbage collection operation has completed. |
 | solidfire_volume_qos_below_min_iops_percentage | histogram | Volume QoS Below minimum IOPS percentage |
@@ -144,7 +139,6 @@ docker run  --rm -p 9987:9987 ghcr.io/mjavier2k/solidfire-exporter:v0.6.6
 | solidfire_volume_qos_throttle_percentage | histogram | Volume QoS throttle percentage |
 | solidfire_volume_qos_write_block_sizes_bytes_bucket | histogram | Volume QoS write block sizes |
 | solidfire_volume_read_bytes_total | counter | The total cumulative bytes read from the volume since the creation of the volume. |
-| solidfire_volume_read_latency_seconds | gauge | The average time, in seconds, to complete read operations to the volume in the last 500 milliseconds. |
 | solidfire_volume_read_latency_seconds_total | counter | The total time spent performing read operations from the volume |
 | solidfire_volume_read_ops_total | counter | The total read operations to the volume since the creation of the volume. |
 | solidfire_volume_size_bytes | gauge | Total provisioned capacity in bytes. |
@@ -153,9 +147,7 @@ docker run  --rm -p 9987:9987 ghcr.io/mjavier2k/solidfire-exporter:v0.6.6
 | solidfire_volume_unaligned_writes_total | counter | The total cumulative unaligned write operations to a volume since the creation of the volume. |
 | solidfire_volume_utilization | gauge | A floating value that describes how much the client is using the volume. Value 0: The client is not using the volume. Value 1: The client is using their maximum. Value 1+: The client is using their burst. |
 | solidfire_volume_write_bytes_total | counter | The total cumulative bytes written to the volume since the creation of the volume. |
-| solidfire_volume_write_latency_seconds | gauge | The average time, in seconds, to complete write operations to a volume in the last 500 milliseconds. |
 | solidfire_volume_write_latency_seconds_total | counter | The total time spent performing write operations to the volume |
-| solidfire_volume_write_ops_last_sample | gauge | The total number of write operations during the last sample period. |
 | solidfire_volume_write_ops_total | counter | The total cumulative write operations to the volume since the creation of the volume. |
 | solidfire_volume_zero_blocks | gauge | The total number of empty 4KiB blocks without data after the last round of garbage collection operation has completed. |
 

pkg/prom/collector.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"math"
+	"regexp"
 	"strconv"
 	"strings"
 	"sync"
@@ -58,22 +59,16 @@ func (c *SolidfireCollector) Describe(ch chan<- *prometheus.Desc) {
 	ch <- MetricDescriptions.VolumeLatencySeconds
 	ch <- MetricDescriptions.VolumeNonZeroBlocks
 	ch <- MetricDescriptions.VolumeReadBytesTotal
-	ch <- MetricDescriptions.VolumeLastSampleReadBytes
-	ch <- MetricDescriptions.VolumeReadLatencySeconds
 	ch <- MetricDescriptions.VolumeReadLatencySecondsTotal
 	ch <- MetricDescriptions.VolumeReadOpsTotal
-	ch <- MetricDescriptions.VolumeLastSampleReadOps
 	ch <- MetricDescriptions.VolumeThrottle
 	ch <- MetricDescriptions.VolumeUnalignedReadsTotal
 	ch <- MetricDescriptions.VolumeUnalignedWritesTotal
 	ch <- MetricDescriptions.VolumeSizeBytes
 	ch <- MetricDescriptions.VolumeUtilization
 	ch <- MetricDescriptions.VolumeWriteBytesTotal
-	ch <- MetricDescriptions.VolumeLastSampleWriteBytes
-	ch <- MetricDescriptions.VolumeWriteLatencySeconds
 	ch <- MetricDescriptions.VolumeWriteLatencyTotal
 	ch <- MetricDescriptions.VolumeWriteOpsTotal
-	ch <- MetricDescriptions.VolumeWriteOpsLastSample
 	ch <- MetricDescriptions.VolumeStatsZeroBlocks
 
 	ch <- MetricDescriptions.ClusterActiveBlockSpaceBytes
@@ -112,10 +107,8 @@ func (c *SolidfireCollector) Describe(ch chan<- *prometheus.Desc) {
 	ch <- MetricDescriptions.NodeInterfaceOutBytesTotal
 	ch <- MetricDescriptions.NodeInterfaceUtilizationPercentage
 	ch <- MetricDescriptions.NodeReadLatencyTotal
-	ch <- MetricDescriptions.NodeReadOpsTotal
 	ch <- MetricDescriptions.NodeUsedMemoryBytes
 	ch <- MetricDescriptions.NodeWriteLatencyTotal
-	ch <- MetricDescriptions.NodeWriteOpsTotal
 	ch <- MetricDescriptions.NodeLoadHistogram
 
 	ch <- MetricDescriptions.NodeInfo
@@ -249,6 +242,9 @@ func (c *SolidfireCollector) collectVolumeStats(ctx context.Context, ch chan<- p
 	mu.Lock()
 	defer mu.Unlock()
 	for _, vol := range volumeStats.Result.VolumeStats {
+		if ok, _ := regexp.MatchString(`snapshot-clone-src-*|replica-vol-*`, c.volumeNamesByID[vol.VolumeID]); ok {
+			continue
+		}
 		ch <- prometheus.MustNewConstMetric(
 			MetricDescriptions.VolumeActualIOPS,
 			prometheus.GaugeValue,
@@ -298,20 +294,6 @@ func (c *SolidfireCollector) collectVolumeStats(ctx context.Context, ch chan<- p
 			strconv.Itoa(vol.VolumeID),
 			c.volumeNamesByID[vol.VolumeID])
 
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.VolumeLastSampleReadBytes,
-			prometheus.GaugeValue,
-			vol.ReadBytesLastSample,
-			strconv.Itoa(vol.VolumeID),
-			c.volumeNamesByID[vol.VolumeID])
-
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.VolumeReadLatencySeconds,
-			prometheus.GaugeValue,
-			MicrosecondsToSeconds(vol.ReadLatencyUSec),
-			strconv.Itoa(vol.VolumeID),
-			c.volumeNamesByID[vol.VolumeID])
-
 		ch <- prometheus.MustNewConstMetric(
 			MetricDescriptions.VolumeReadLatencySecondsTotal,
 			prometheus.CounterValue,
@@ -326,13 +308,6 @@ func (c *SolidfireCollector) collectVolumeStats(ctx context.Context, ch chan<- p
 			strconv.Itoa(vol.VolumeID),
 			c.volumeNamesByID[vol.VolumeID])
 
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.VolumeLastSampleReadOps,
-			prometheus.GaugeValue,
-			vol.ReadOpsLastSample,
-			strconv.Itoa(vol.VolumeID),
-			c.volumeNamesByID[vol.VolumeID])
-
 		ch <- prometheus.MustNewConstMetric(
 			MetricDescriptions.VolumeThrottle,
 			prometheus.GaugeValue,
@@ -375,20 +350,6 @@ func (c *SolidfireCollector) collectVolumeStats(ctx context.Context, ch chan<- p
 			strconv.Itoa(vol.VolumeID),
 			c.volumeNamesByID[vol.VolumeID])
 
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.VolumeLastSampleWriteBytes,
-			prometheus.GaugeValue,
-			vol.WriteBytesLastSample,
-			strconv.Itoa(vol.VolumeID),
-			c.volumeNamesByID[vol.VolumeID])
-
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.VolumeWriteLatencySeconds,
-			prometheus.GaugeValue,
-			MicrosecondsToSeconds(vol.WriteLatencyUSec),
-			strconv.Itoa(vol.VolumeID),
-			c.volumeNamesByID[vol.VolumeID])
-
 		ch <- prometheus.MustNewConstMetric(
 			MetricDescriptions.VolumeWriteLatencyTotal,
 			prometheus.CounterValue,
@@ -403,13 +364,6 @@ func (c *SolidfireCollector) collectVolumeStats(ctx context.Context, ch chan<- p
 			strconv.Itoa(vol.VolumeID),
 			c.volumeNamesByID[vol.VolumeID])
 
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.VolumeWriteOpsLastSample,
-			prometheus.GaugeValue,
-			vol.WriteOpsLastSample,
-			strconv.Itoa(vol.VolumeID),
-			c.volumeNamesByID[vol.VolumeID])
-
 		ch <- prometheus.MustNewConstMetric(
 			MetricDescriptions.VolumeStatsZeroBlocks,
 			prometheus.GaugeValue,
@@ -712,14 +666,6 @@ func (c *SolidfireCollector) collectClusterNodeStats(ctx context.Context, ch cha
 			c.nodesNamesByID[stats.NodeID],
 		)
 
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.NodeReadOpsTotal,
-			prometheus.CounterValue,
-			stats.ReadOps,
-			strconv.Itoa(stats.NodeID),
-			c.nodesNamesByID[stats.NodeID],
-		)
-
 		ch <- prometheus.MustNewConstMetric(
 			MetricDescriptions.NodeInterfaceInBytesTotal,
 			prometheus.CounterValue,
@@ -754,13 +700,6 @@ func (c *SolidfireCollector) collectClusterNodeStats(ctx context.Context, ch cha
 			c.nodesNamesByID[stats.NodeID],
 		)
 
-		ch <- prometheus.MustNewConstMetric(
-			MetricDescriptions.NodeWriteOpsTotal,
-			prometheus.CounterValue,
-			stats.WriteOps,
-			strconv.Itoa(stats.NodeID),
-			c.nodesNamesByID[stats.NodeID],
-		)
 	}
 	return nil
 }
@@ -773,6 +712,9 @@ func (c *SolidfireCollector) collectVolumeQosHistograms(ctx context.Context, ch
 	mu.Lock()
 	defer mu.Unlock()
 	for _, h := range VolumeQoSHistograms.Result.QosHistograms {
+		if ok, _ := regexp.MatchString(`snapshot-clone-src-*|replica-vol-*`, c.volumeNamesByID[h.VolumeID]); ok {
+			continue
+		}
 		// Below Min IOPS Percentage
 		BelowMinIopsPercentages := map[float64]uint64{
 			19:  h.Histograms.BelowMinIopsPercentages.Bucket1To19,
@@ -1256,27 +1198,20 @@ func (c *SolidfireCollector) collectISCSISessions(ctx context.Context, ch chan<-
 	}
 	mu.Lock()
 	defer mu.Unlock()
-	sessions := make(map[int]map[int]float64)
+	sessions := make(map[int]float64)
 
 	for _, session := range ListISCSISessions.Result.Sessions {
-		if sessions[session.NodeID] == nil {
-			sessions[session.NodeID] = make(map[int]float64)
-		}
-		sessions[session.NodeID][session.VolumeID]++
+		sessions[session.NodeID]++
 	}
 
-	for node, v := range sessions {
-		for vol, val := range v {
-			ch <- prometheus.MustNewConstMetric(
-				MetricDescriptions.NodeISCSISessions,
-				prometheus.GaugeValue,
-				val,
-				strconv.Itoa(node),
-				c.nodesNamesByID[node],
-				strconv.Itoa(vol),
-				c.volumeNamesByID[vol],
-			)
-		}
+	for node, val := range sessions {
+		ch <- prometheus.MustNewConstMetric(
+			MetricDescriptions.NodeISCSISessions,
+			prometheus.GaugeValue,
+			val,
+			strconv.Itoa(node),
+			c.nodesNamesByID[node],
+		)
 	}
 	return nil
 }

pkg/prom/metrics.go

@@ -13,13 +13,9 @@ type Descriptions struct {
 	VolumeAverageIOPSizeBytes     *prometheus.Desc
 	VolumeBurstIOPSCredit         *prometheus.Desc
 	VolumeClientQueueDepth        *prometheus.Desc
-	VolumeLastSampleReadBytes     *prometheus.Desc
-	VolumeLastSampleReadOps       *prometheus.Desc
-	VolumeLastSampleWriteBytes    *prometheus.Desc
 	VolumeLatencySeconds          *prometheus.Desc
 	VolumeNonZeroBlocks           *prometheus.Desc
 	VolumeReadBytesTotal          *prometheus.Desc
-	VolumeReadLatencySeconds      *prometheus.Desc
 	VolumeReadLatencySecondsTotal *prometheus.Desc
 	VolumeReadOpsTotal            *prometheus.Desc
 	VolumeThrottle                *prometheus.Desc
@@ -28,9 +24,7 @@ type Descriptions struct {
 	VolumeSizeBytes               *prometheus.Desc
 	VolumeUtilization             *prometheus.Desc
 	VolumeWriteBytesTotal         *prometheus.Desc
-	VolumeWriteLatencySeconds     *prometheus.Desc
 	VolumeWriteLatencyTotal       *prometheus.Desc
-	VolumeWriteOpsLastSample      *prometheus.Desc
 	VolumeWriteOpsTotal           *prometheus.Desc
 	VolumeStatsZeroBlocks         *prometheus.Desc
 
@@ -82,12 +76,10 @@ type Descriptions struct {
 	NodeInterfaceUtilizationPercentage *prometheus.Desc
 	NodeLoadHistogram                  *prometheus.Desc
 	NodeReadLatencyTotal               *prometheus.Desc
-	NodeReadOpsTotal                   *prometheus.Desc
 	NodeSamples                        *prometheus.Desc
 	NodeTotalMemoryBytes               *prometheus.Desc
 	NodeUsedMemoryBytes                *prometheus.Desc
 	NodeWriteLatencyTotal              *prometheus.Desc
-	NodeWriteOpsTotal                  *prometheus.Desc
 
 	// ListAllNodes
 	NodeInfo *prometheus.Desc
@@ -211,20 +203,6 @@ func NewMetricDescriptions(namespace string) *Descriptions {
 		nil,
 	)
 
-	d.VolumeLastSampleReadBytes = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "volume_last_sample_read_bytes"),
-		"The total number of bytes read from the volume during the last sample period.",
-		[]string{"volume_id", "volume_name"},
-		nil,
-	)
-
-	d.VolumeReadLatencySeconds = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "volume_read_latency_seconds"),
-		"The average time, in seconds, to complete read operations to the volume in the last 500 milliseconds.",
-		[]string{"volume_id", "volume_name"},
-		nil,
-	)
-
 	d.VolumeReadLatencySecondsTotal = prometheus.NewDesc(
 		prometheus.BuildFQName(namespace, "", "volume_read_latency_seconds_total"),
 		"The total time spent performing read operations from the volume",
@@ -239,13 +217,6 @@ func NewMetricDescriptions(namespace string) *Descriptions {
 		nil,
 	)
 
-	d.VolumeLastSampleReadOps = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "volume_last_sample_read_ops"),
-		"The total number of read operations during the last sample period",
-		[]string{"volume_id", "volume_name"},
-		nil,
-	)
-
 	d.VolumeThrottle = prometheus.NewDesc(
 		prometheus.BuildFQName(namespace, "", "volume_throttle"),
 		"A floating value between 0 and 1 that represents how much the system is throttling clients below their maxIOPS because of rereplication of data, transient errors, and snapshots taken.",
@@ -288,20 +259,6 @@ func NewMetricDescriptions(namespace string) *Descriptions {
 		nil,
 	)
 
-	d.VolumeLastSampleWriteBytes = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "volume_last_sample_write_bytes"),
-		"The total number of bytes written to the volume during the last sample period.",
-		[]string{"volume_id", "volume_name"},
-		nil,
-	)
-
-	d.VolumeWriteLatencySeconds = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "volume_write_latency_seconds"),
-		"The average time, in seconds, to complete write operations to a volume in the last 500 milliseconds.",
-		[]string{"volume_id", "volume_name"},
-		nil,
-	)
-
 	d.VolumeWriteLatencyTotal = prometheus.NewDesc(
 		prometheus.BuildFQName(namespace, "", "volume_write_latency_seconds_total"),
 		"The total time spent performing write operations to the volume",
@@ -316,13 +273,6 @@ func NewMetricDescriptions(namespace string) *Descriptions {
 		nil,
 	)
 
-	d.VolumeWriteOpsLastSample = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "volume_write_ops_last_sample"),
-		"The total number of write operations during the last sample period.",
-		[]string{"volume_id", "volume_name"},
-		nil,
-	)
-
 	d.VolumeStatsZeroBlocks = prometheus.NewDesc(
 		prometheus.BuildFQName(namespace, "", "volume_zero_blocks"),
 		"The total number of empty 4KiB blocks without data after the last round of garbage collection operation has completed.",
@@ -544,20 +494,6 @@ func NewMetricDescriptions(namespace string) *Descriptions {
 		nil,
 	)
 
-	d.NodeReadOpsTotal = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "node_read_ops_total"),
-		"Total read operations to a node.", // undocumented metric
-		[]string{"node_id", "node_name"},
-		nil,
-	)
-
-	d.NodeWriteOpsTotal = prometheus.NewDesc(
-		prometheus.BuildFQName(namespace, "", "node_write_ops_total"),
-		"Total write operations to a node", // undocumented metric
-		[]string{"node_id", "node_name"},
-		nil,
-	)
-
 	d.NodeTotalMemoryBytes = prometheus.NewDesc(
 		prometheus.BuildFQName(namespace, "", "node_total_memory_bytes"),
 		"Total node memory in bytes.",
@@ -932,7 +868,7 @@ func NewMetricDescriptions(namespace string) *Descriptions {
 	d.NodeISCSISessions = prometheus.NewDesc(
 		prometheus.BuildFQName(namespace, "", "node_iscsi_sessions"),
 		"The total number of iscsi sessions per node and volume",
-		[]string{"node_id", "node_name", "volume_id", "volume_name"},
+		[]string{"node_id", "node_name"},
 		nil,
 	)