Skip to content

Commit

Permalink
Add metrics for rocksdb bloom filter and lsm tree (#1710) (#1728)
Browse files Browse the repository at this point in the history 
* Add metrics for rocksdb bloom filter

* Add metrics for rocksdb lsm tree

* Add metrics for rocksdb lsm tree

* Add metrics for rocksdb detailed cache

* Fix tests

* Add help for metrics

* Add help for metrics

(cherry picked from commit 9aefbac)

Co-authored-by: Evgeniy Scherbina <evgeniy.shcherbina.es@gmail.com>
  • Loading branch information
@mergify @evgeniy-scherbina @pirtleshell
2 people authored and pirtleshell committed Oct 25, 2023
1 parent a6f771e commit f8ca056
Show file tree
Hide file tree
Showing 3 changed files with 208 additions and 1 deletion.
181 changes: 180 additions & 1 deletion cmd/kava/opendb/metrics.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,19 @@ type Metrics struct {
BlockCacheAdd metrics.Gauge
BlockCacheAddFailures metrics.Gauge

// Detailed Cache
BlockCacheIndexMiss metrics.Gauge
BlockCacheIndexHit metrics.Gauge
BlockCacheIndexBytesInsert metrics.Gauge

BlockCacheFilterMiss metrics.Gauge
BlockCacheFilterHit metrics.Gauge
BlockCacheFilterBytesInsert metrics.Gauge

BlockCacheDataMiss metrics.Gauge
BlockCacheDataHit metrics.Gauge
BlockCacheDataBytesInsert metrics.Gauge

// Latency
DBGetMicrosP50 metrics.Gauge
DBGetMicrosP95 metrics.Gauge
Expand All @@ -58,6 +71,21 @@ type Metrics struct {
DBWriteStallP100 metrics.Gauge
DBWriteStallCount metrics.Gauge
DBWriteStallSum metrics.Gauge

// Bloom Filter
BloomFilterUseful metrics.Gauge
BloomFilterFullPositive metrics.Gauge
BloomFilterFullTruePositive metrics.Gauge

// LSM Tree Stats
LastLevelReadBytes metrics.Gauge
LastLevelReadCount metrics.Gauge
NonLastLevelReadBytes metrics.Gauge
NonLastLevelReadCount metrics.Gauge

GetHitL0 metrics.Gauge
GetHitL1 metrics.Gauge
GetHitL2AndUp metrics.Gauge
}

// registerMetrics registers metrics in prometheus and initializes rocksdbMetrics variable
Expand Down Expand Up @@ -161,6 +189,64 @@ func registerMetrics() {
Help: "number of failures when adding blocks to block cache",
}, labels),

// Detailed Cache
BlockCacheIndexMiss: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_index_miss",
Help: "",
}, labels),
BlockCacheIndexHit: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_index_hit",
Help: "",
}, labels),
BlockCacheIndexBytesInsert: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_index_bytes_insert",
Help: "",
}, labels),

BlockCacheFilterMiss: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_filter_miss",
Help: "",
}, labels),
BlockCacheFilterHit: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_filter_hit",
Help: "",
}, labels),
BlockCacheFilterBytesInsert: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_filter_bytes_insert",
Help: "",
}, labels),

BlockCacheDataMiss: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_data_miss",
Help: "",
}, labels),
BlockCacheDataHit: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_data_hit",
Help: "",
}, labels),
BlockCacheDataBytesInsert: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "detailed_cache",
Name: "block_cache_data_bytes_insert",
Help: "",
}, labels),

// Latency
DBGetMicrosP50: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Expand Down Expand Up @@ -229,7 +315,7 @@ func registerMetrics() {
Namespace: "rocksdb",
Subsystem: "stall",
Name: "stall_micros",
Help: "",
Help: "Writer has to wait for compaction or flush to finish.",
}, labels),

DBWriteStallP50: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Expand Down Expand Up @@ -268,6 +354,71 @@ func registerMetrics() {
Name: "db_write_stall_sum",
Help: "",
}, labels),

// Bloom Filter
BloomFilterUseful: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "filter",
Name: "bloom_filter_useful",
Help: "number of times bloom filter has avoided file reads, i.e., negatives.",
}, labels),
BloomFilterFullPositive: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "filter",
Name: "bloom_filter_full_positive",
Help: "number of times bloom FullFilter has not avoided the reads.",
}, labels),
BloomFilterFullTruePositive: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "filter",
Name: "bloom_filter_full_true_positive",
Help: "number of times bloom FullFilter has not avoided the reads and data actually exist.",
}, labels),

// LSM Tree Stats
LastLevelReadBytes: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "last_level_read_bytes",
Help: "",
}, labels),
LastLevelReadCount: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "last_level_read_count",
Help: "",
}, labels),
NonLastLevelReadBytes: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "non_last_level_read_bytes",
Help: "",
}, labels),
NonLastLevelReadCount: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "non_last_level_read_count",
Help: "",
}, labels),

GetHitL0: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "get_hit_l0",
Help: "number of Get() queries served by L0",
}, labels),
GetHitL1: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "get_hit_l1",
Help: "number of Get() queries served by L1",
}, labels),
GetHitL2AndUp: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: "rocksdb",
Subsystem: "lsm",
Name: "get_hit_l2_and_up",
Help: "number of Get() queries served by L2 and up",
}, labels),
}
}

Expand Down Expand Up @@ -295,6 +446,19 @@ func (m *Metrics) report(props *properties, stats *stats) {
m.BlockCacheAdd.Set(float64(stats.BlockCacheAdd))
m.BlockCacheAddFailures.Set(float64(stats.BlockCacheAddFailures))

// Detailed Cache
m.BlockCacheIndexMiss.Set(float64(stats.BlockCacheIndexMiss))
m.BlockCacheIndexHit.Set(float64(stats.BlockCacheIndexHit))
m.BlockCacheIndexBytesInsert.Set(float64(stats.BlockCacheIndexBytesInsert))

m.BlockCacheFilterMiss.Set(float64(stats.BlockCacheFilterMiss))
m.BlockCacheFilterHit.Set(float64(stats.BlockCacheFilterHit))
m.BlockCacheFilterBytesInsert.Set(float64(stats.BlockCacheFilterBytesInsert))

m.BlockCacheDataMiss.Set(float64(stats.BlockCacheDataMiss))
m.BlockCacheDataHit.Set(float64(stats.BlockCacheDataHit))
m.BlockCacheDataBytesInsert.Set(float64(stats.BlockCacheDataBytesInsert))

// Latency
m.DBGetMicrosP50.Set(stats.DBGetMicros.P50)
m.DBGetMicrosP95.Set(stats.DBGetMicros.P95)
Expand All @@ -317,4 +481,19 @@ func (m *Metrics) report(props *properties, stats *stats) {
m.DBWriteStallP100.Set(stats.DBWriteStallHistogram.P100)
m.DBWriteStallCount.Set(stats.DBWriteStallHistogram.Count)
m.DBWriteStallSum.Set(stats.DBWriteStallHistogram.Sum)

// Bloom Filter
m.BloomFilterUseful.Set(float64(stats.BloomFilterUseful))
m.BloomFilterFullPositive.Set(float64(stats.BloomFilterFullPositive))
m.BloomFilterFullTruePositive.Set(float64(stats.BloomFilterFullTruePositive))

// LSM Tree Stats
m.LastLevelReadBytes.Set(float64(stats.LastLevelReadBytes))
m.LastLevelReadCount.Set(float64(stats.LastLevelReadCount))
m.NonLastLevelReadBytes.Set(float64(stats.NonLastLevelReadBytes))
m.NonLastLevelReadCount.Set(float64(stats.NonLastLevelReadCount))

m.GetHitL0.Set(float64(stats.GetHitL0))
m.GetHitL1.Set(float64(stats.GetHitL1))
m.GetHitL2AndUp.Set(float64(stats.GetHitL2AndUp))
}
19 changes: 19 additions & 0 deletions cmd/kava/opendb/stats_loader.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,16 @@ type stats struct {
// # of failures when adding blocks to block cache.
BlockCacheAddFailures int64

BlockCacheIndexMiss int64
BlockCacheIndexHit int64
BlockCacheIndexBytesInsert int64
BlockCacheFilterMiss int64
BlockCacheFilterHit int64
BlockCacheFilterBytesInsert int64
BlockCacheDataMiss int64
BlockCacheDataHit int64
BlockCacheDataBytesInsert int64

CompactReadBytes int64 // Bytes read during compaction
CompactWriteBytes int64 // Bytes written during compaction

Expand Down Expand Up @@ -161,6 +171,15 @@ func (l *statLoader) load() (*stats, error) {
BlockCacheHit: l.getInt64StatValue("rocksdb.block.cache.hit", count),
BlockCacheAdd: l.getInt64StatValue("rocksdb.block.cache.add", count),
BlockCacheAddFailures: l.getInt64StatValue("rocksdb.block.cache.add.failures", count),
BlockCacheIndexMiss: l.getInt64StatValue("rocksdb.block.cache.index.miss", count),
BlockCacheIndexHit: l.getInt64StatValue("rocksdb.block.cache.index.hit", count),
BlockCacheIndexBytesInsert: l.getInt64StatValue("rocksdb.block.cache.index.bytes.insert", count),
BlockCacheFilterMiss: l.getInt64StatValue("rocksdb.block.cache.filter.miss", count),
BlockCacheFilterHit: l.getInt64StatValue("rocksdb.block.cache.filter.hit", count),
BlockCacheFilterBytesInsert: l.getInt64StatValue("rocksdb.block.cache.filter.bytes.insert", count),
BlockCacheDataMiss: l.getInt64StatValue("rocksdb.block.cache.data.miss", count),
BlockCacheDataHit: l.getInt64StatValue("rocksdb.block.cache.data.hit", count),
BlockCacheDataBytesInsert: l.getInt64StatValue("rocksdb.block.cache.data.bytes.insert", count),
CompactReadBytes: l.getInt64StatValue("rocksdb.compact.read.bytes", count),
CompactWriteBytes: l.getInt64StatValue("rocksdb.compact.write.bytes", count),
CompactionTimesMicros: l.getFloat64HistogramStatValue("rocksdb.compaction.times.micros"),
Expand Down
9 changes: 9 additions & 0 deletions cmd/kava/opendb/stats_loader_test.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,15 @@ func TestStatsLoader(t *testing.T) {
"rocksdb.block.cache.hit": &defaultStat,
"rocksdb.block.cache.add": &defaultStat,
"rocksdb.block.cache.add.failures": &defaultStat,
"rocksdb.block.cache.index.miss": &defaultStat,
"rocksdb.block.cache.index.hit": &defaultStat,
"rocksdb.block.cache.index.bytes.insert": &defaultStat,
"rocksdb.block.cache.filter.miss": &defaultStat,
"rocksdb.block.cache.filter.hit": &defaultStat,
"rocksdb.block.cache.filter.bytes.insert": &defaultStat,
"rocksdb.block.cache.data.miss": &defaultStat,
"rocksdb.block.cache.data.hit": &defaultStat,
"rocksdb.block.cache.data.bytes.insert": &defaultStat,
"rocksdb.compact.read.bytes": &defaultStat,
"rocksdb.compact.write.bytes": &defaultStat,
"rocksdb.compaction.times.micros": &defaultHistogramStat,
Expand Down

0 comments on commit f8ca056

Please sign in to comment.