zolfa
/
tendermint
1
0
Fork 0
Code Issues 0 Pull Requests 0 Projects 0 Releases 221 Wiki Activity
Browse Source

metrics: add metrics for specific validators (#4294) 

* add metrics

* change == to !=

* update metrics

* cs: check cs.privValidator is not nil in recordMetrics

Co-authored-by: Anton Kaliaev <anton.kalyaev@gmail.com>
rc1/v0.32.12
Marko 4 years ago
committed by Jack Zampolin
parent
be771f225a
commit
51ccaf6fb8
4 changed files with 86 additions and 31 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +8
    -0
      CHANGELOG_PENDING.md
  2. +29
    -0
      consensus/metrics.go
  3. +18
    -3
      consensus/state.go
  4. +31
    -28
      docs/tendermint-core/metrics.md

+ 8
- 0
CHANGELOG_PENDING.md View File

@ -16,10 +16,18 @@ program](https://hackerone.com/tendermint).
- Go API
### FEATURES:
- [rpc/lib] [\#4248](https://github.com/tendermint/tendermint/issues/4248) RPC client basic authentication support (@greg-szabo)
- [metrics] \#4263 Add
- `consensus_validator_power`: track your validators power
- `consensus_validator_last_signed_height`: track at which height the validator last signed
- `consensus_validator_missed_blocks`: total amount of missed blocks for a validator
as gauges in prometheus for validator specific metrics
### IMPROVEMENTS:
### BUG FIXES:
- [rpc/lib] [\#4051](https://github.com/tendermint/tendermint/pull/4131) Fix RPC client, which was previously resolving https protocol to http (@yenkhoon)
- [cs] \#4069 Don't panic when block meta is not found in store (@gregzaitsev)

+ 29
- 0
consensus/metrics.go View File

@ -19,6 +19,9 @@ type Metrics struct {
// Height of the chain.
Height metrics.Gauge
// ValidatorLastSignedHeight of a validator.
ValidatorLastSignedHeight metrics.Gauge
// Number of rounds.
Rounds metrics.Gauge
@ -26,6 +29,10 @@ type Metrics struct {
Validators metrics.Gauge
// Total power of all validators.
ValidatorsPower metrics.Gauge
// Power of a validator.
ValidatorPower metrics.Gauge
// Amount of blocks missed by a validator.
ValidatorMissedBlocks metrics.Gauge
// Number of validators who did not sign.
MissingValidators metrics.Gauge
// Total power of the missing validators.
@ -81,12 +88,30 @@ func PrometheusMetrics(namespace string, labelsAndValues ...string) *Metrics {
Name: "validators",
Help: "Number of validators.",
}, labels).With(labelsAndValues...),
ValidatorLastSignedHeight: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: namespace,
Subsystem: MetricsSubsystem,
Name: "validator_last_signed_height",
Help: "Last signed height for a validator",
}, append(labels, "validator_address")).With(labelsAndValues...),
ValidatorMissedBlocks: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: namespace,
Subsystem: MetricsSubsystem,
Name: "validator_missed_blocks",
Help: "Total missed blocks for a validator",
}, append(labels, "validator_address")).With(labelsAndValues...),
ValidatorsPower: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: namespace,
Subsystem: MetricsSubsystem,
Name: "validators_power",
Help: "Total power of all validators.",
}, labels).With(labelsAndValues...),
ValidatorPower: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: namespace,
Subsystem: MetricsSubsystem,
Name: "validator_power",
Help: "Power of a validator",
}, append(labels, "validator_address")).With(labelsAndValues...),
MissingValidators: prometheus.NewGaugeFrom(stdprometheus.GaugeOpts{
Namespace: namespace,
Subsystem: MetricsSubsystem,
@ -163,10 +188,14 @@ func NopMetrics() *Metrics {
return &Metrics{
Height: discard.NewGauge(),
ValidatorLastSignedHeight: discard.NewGauge(),
Rounds: discard.NewGauge(),
Validators: discard.NewGauge(),
ValidatorsPower: discard.NewGauge(),
ValidatorPower: discard.NewGauge(),
ValidatorMissedBlocks: discard.NewGauge(),
MissingValidators: discard.NewGauge(),
MissingValidatorsPower: discard.NewGauge(),
ByzantineValidators: discard.NewGauge(),


+ 18
- 3
consensus/state.go View File

@ -1459,8 +1459,11 @@ func (cs *ConsensusState) finalizeCommit(height int64) {
func (cs *ConsensusState) recordMetrics(height int64, block *types.Block) {
cs.metrics.Validators.Set(float64(cs.Validators.Size()))
cs.metrics.ValidatorsPower.Set(float64(cs.Validators.TotalVotingPower()))
missingValidators := 0
missingValidatorsPower := int64(0)
var (
missingValidators = 0
missingValidatorsPower int64
)
for i, val := range cs.Validators.Validators {
var vote *types.CommitSig
if i < len(block.LastCommit.Precommits) {
@ -1470,9 +1473,22 @@ func (cs *ConsensusState) recordMetrics(height int64, block *types.Block) {
missingValidators++
missingValidatorsPower += val.VotingPower
}
if cs.privValidator != nil && bytes.Equal(val.Address, cs.privValidator.GetPubKey().Address()) {
label := []string{
"validator_address", val.Address.String(),
}
cs.metrics.ValidatorPower.With(label...).Set(float64(val.VotingPower))
if vote != nil {
cs.metrics.ValidatorLastSignedHeight.With(label...).Set(float64(height))
} else {
cs.metrics.ValidatorMissedBlocks.With(label...).Add(float64(1))
}
}
}
cs.metrics.MissingValidators.Set(float64(missingValidators))
cs.metrics.MissingValidatorsPower.Set(float64(missingValidatorsPower))
cs.metrics.ByzantineValidators.Set(float64(len(block.Evidence.Evidence)))
byzantineValidatorsPower := int64(0)
for _, ev := range block.Evidence.Evidence {
@ -1493,7 +1509,6 @@ func (cs *ConsensusState) recordMetrics(height int64, block *types.Block) {
cs.metrics.BlockSizeBytes.Set(float64(block.Size()))
cs.metrics.TotalTxs.Set(float64(block.TotalTxs))
cs.metrics.CommittedHeight.Set(float64(block.Height))
}
//-----------------------------------------------------------------------------


+ 31
- 28
docs/tendermint-core/metrics.md View File

@ -18,34 +18,37 @@ Listen address can be changed in the config file (see
The following metrics are available:
| **Name** | **Type** | **Since** | **Tags** | **Description** |
| --------------------------------------- | --------- | --------- | -------------- | --------------------------------------------------------------- |
| consensus\_height | Gauge | 0.21.0 | | Height of the chain |
| consensus\_validators | Gauge | 0.21.0 | | Number of validators |
| consensus\_validators\_power | Gauge | 0.21.0 | | Total voting power of all validators |
| consensus\_missing\_validators | Gauge | 0.21.0 | | Number of validators who did not sign |
| consensus\_missing\_validators\_power | Gauge | 0.21.0 | | Total voting power of the missing validators |
| consensus\_byzantine\_validators | Gauge | 0.21.0 | | Number of validators who tried to double sign |
| consensus\_byzantine\_validators\_power | Gauge | 0.21.0 | | Total voting power of the byzantine validators |
| consensus\_block\_interval\_seconds | Histogram | 0.21.0 | | Time between this and last block (Block.Header.Time) in seconds |
| consensus\_rounds | Gauge | 0.21.0 | | Number of rounds |
| consensus\_num\_txs | Gauge | 0.21.0 | | Number of transactions |
| consensus\_block\_parts | counter | on dev | peer\_id | number of blockparts transmitted by peer |
| consensus\_latest\_block\_height | gauge | on dev | | /status sync\_info number |
| consensus\_fast\_syncing | gauge | on dev | | either 0 (not fast syncing) or 1 (syncing) |
| consensus\_total\_txs | Gauge | 0.21.0 | | Total number of transactions committed |
| consensus\_block\_size\_bytes | Gauge | 0.21.0 | | Block size in bytes |
| p2p\_peers | Gauge | 0.21.0 | | Number of peers node's connected to |
| p2p\_peer\_receive\_bytes\_total | counter | on dev | peer\_id, chID | number of bytes per channel received from a given peer |
| p2p\_peer\_send\_bytes\_total | counter | on dev | peer\_id, chID | number of bytes per channel sent to a given peer |
| p2p\_peer\_pending\_send\_bytes | gauge | on dev | peer\_id | number of pending bytes to be sent to a given peer |
| p2p\_num\_txs | gauge | on dev | peer\_id | number of transactions submitted by each peer\_id |
| p2p\_pending\_send\_bytes | gauge | on dev | peer\_id | amount of data pending to be sent to peer |
| mempool\_size | Gauge | 0.21.0 | | Number of uncommitted transactions |
| mempool\_tx\_size\_bytes | histogram | on dev | | transaction sizes in bytes |
| mempool\_failed\_txs | counter | on dev | | number of failed transactions |
| mempool\_recheck\_times | counter | on dev | | number of transactions rechecked in the mempool |
| state\_block\_processing\_time | histogram | on dev | | time between BeginBlock and EndBlock in ms |
| **Name** | **Type** | **Since** | **Tags** | **Description** |
| -------------------------------------- | --------- | --------- | ------------- | ---------------------------------------------------------------------- |
| consensus_height | Gauge | 0.21.0 | | Height of the chain |
| consensus_validators | Gauge | 0.21.0 | | Number of validators |
| consensus_validators_power | Gauge | 0.21.0 | | Total voting power of all validators |
| consensus_validator_power | Gauge | 0.33.0 | | Voting power of the node if in the validator set |
| consensus_validator_last_signed_height | Gauge | 0.33.0 | | Last height the node signed a block, if the node is a validator |
| consensus_validator_missed_blocks | Gauge | 0.33.0 | | Total amount of blocks missed for the node, if the node is a validator |
| consensus_missing_validators | Gauge | 0.21.0 | | Number of validators who did not sign |
| consensus_missing_validators_power | Gauge | 0.21.0 | | Total voting power of the missing validators |
| consensus_byzantine_validators | Gauge | 0.21.0 | | Number of validators who tried to double sign |
| consensus_byzantine_validators_power | Gauge | 0.21.0 | | Total voting power of the byzantine validators |
| consensus_block_interval_seconds | Histogram | 0.21.0 | | Time between this and last block (Block.Header.Time) in seconds |
| consensus_rounds | Gauge | 0.21.0 | | Number of rounds |
| consensus_num_txs | Gauge | 0.21.0 | | Number of transactions |
| consensus_total_txs | Gauge | 0.21.0 | | Total number of transactions committed |
| consensus_block_parts | counter | on dev | peer_id | number of blockparts transmitted by peer |
| consensus_latest_block_height | gauge | on dev | | /status sync_info number |
| consensus_fast_syncing | gauge | on dev | | either 0 (not fast syncing) or 1 (syncing) |
| consensus_block_size_bytes | Gauge | 0.21.0 | | Block size in bytes |
| p2p_peers | Gauge | 0.21.0 | | Number of peers node's connected to |
| p2p_peer_receive_bytes_total | counter | on dev | peer_id, chID | number of bytes per channel received from a given peer |
| p2p_peer_send_bytes_total | counter | on dev | peer_id, chID | number of bytes per channel sent to a given peer |
| p2p_peer_pending_send_bytes | gauge | on dev | peer_id | number of pending bytes to be sent to a given peer |
| p2p_num_txs | gauge | on dev | peer_id | number of transactions submitted by each peer_id |
| p2p_pending_send_bytes | gauge | on dev | peer_id | amount of data pending to be sent to peer |
| mempool_size | Gauge | 0.21.0 | | Number of uncommitted transactions |
| mempool_tx_size_bytes | histogram | on dev | | transaction sizes in bytes |
| mempool_failed_txs | counter | on dev | | number of failed transactions |
| mempool_recheck_times | counter | on dev | | number of transactions rechecked in the mempool |
| state_block_processing_time | histogram | on dev | | time between BeginBlock and EndBlock in ms |
## Useful queries


Write Preview
Loading…
Cancel
Save