cmd, core, metrics: always report expensive metrics

This commit is contained in:
Péter Szilágyi 2024-03-07 12:05:05 +02:00
parent aadcb88675
commit 13ff5e1442
8 changed files with 67 additions and 111 deletions

View file

@ -267,7 +267,7 @@ func applyMetricConfig(ctx *cli.Context, cfg *gethConfig) {
cfg.Metrics.Enabled = ctx.Bool(utils.MetricsEnabledFlag.Name) cfg.Metrics.Enabled = ctx.Bool(utils.MetricsEnabledFlag.Name)
} }
if ctx.IsSet(utils.MetricsEnabledExpensiveFlag.Name) { if ctx.IsSet(utils.MetricsEnabledExpensiveFlag.Name) {
cfg.Metrics.EnabledExpensive = ctx.Bool(utils.MetricsEnabledExpensiveFlag.Name) log.Warn("Expensive metrics are collected by default, please remove this flag", "flag", utils.MetricsEnabledExpensiveFlag.Name)
} }
if ctx.IsSet(utils.MetricsHTTPFlag.Name) { if ctx.IsSet(utils.MetricsHTTPFlag.Name) {
cfg.Metrics.HTTP = ctx.String(utils.MetricsHTTPFlag.Name) cfg.Metrics.HTTP = ctx.String(utils.MetricsHTTPFlag.Name)

View file

@ -862,12 +862,6 @@ var (
Usage: "Enable metrics collection and reporting", Usage: "Enable metrics collection and reporting",
Category: flags.MetricsCategory, Category: flags.MetricsCategory,
} }
MetricsEnabledExpensiveFlag = &cli.BoolFlag{
Name: "metrics.expensive",
Usage: "Enable expensive metrics collection and reporting",
Category: flags.MetricsCategory,
}
// MetricsHTTPFlag defines the endpoint for a stand-alone metrics HTTP endpoint. // MetricsHTTPFlag defines the endpoint for a stand-alone metrics HTTP endpoint.
// Since the pprof service enables sensitive/vulnerable behavior, this allows a user // Since the pprof service enables sensitive/vulnerable behavior, this allows a user
// to enable a public-OK metrics endpoint without having to worry about ALSO exposing // to enable a public-OK metrics endpoint without having to worry about ALSO exposing

View file

@ -93,35 +93,35 @@ var (
Name: "light.serve", Name: "light.serve",
Usage: "Maximum percentage of time allowed for serving LES requests (deprecated)", Usage: "Maximum percentage of time allowed for serving LES requests (deprecated)",
Value: ethconfig.Defaults.LightServ, Value: ethconfig.Defaults.LightServ,
Category: flags.LightCategory, Category: flags.DeprecatedCategory,
} }
LightIngressFlag = &cli.IntFlag{ LightIngressFlag = &cli.IntFlag{
Name: "light.ingress", Name: "light.ingress",
Usage: "Incoming bandwidth limit for serving light clients (deprecated)", Usage: "Incoming bandwidth limit for serving light clients (deprecated)",
Value: ethconfig.Defaults.LightIngress, Value: ethconfig.Defaults.LightIngress,
Category: flags.LightCategory, Category: flags.DeprecatedCategory,
} }
LightEgressFlag = &cli.IntFlag{ LightEgressFlag = &cli.IntFlag{
Name: "light.egress", Name: "light.egress",
Usage: "Outgoing bandwidth limit for serving light clients (deprecated)", Usage: "Outgoing bandwidth limit for serving light clients (deprecated)",
Value: ethconfig.Defaults.LightEgress, Value: ethconfig.Defaults.LightEgress,
Category: flags.LightCategory, Category: flags.DeprecatedCategory,
} }
LightMaxPeersFlag = &cli.IntFlag{ LightMaxPeersFlag = &cli.IntFlag{
Name: "light.maxpeers", Name: "light.maxpeers",
Usage: "Maximum number of light clients to serve, or light servers to attach to (deprecated)", Usage: "Maximum number of light clients to serve, or light servers to attach to (deprecated)",
Value: ethconfig.Defaults.LightPeers, Value: ethconfig.Defaults.LightPeers,
Category: flags.LightCategory, Category: flags.DeprecatedCategory,
} }
LightNoPruneFlag = &cli.BoolFlag{ LightNoPruneFlag = &cli.BoolFlag{
Name: "light.nopruning", Name: "light.nopruning",
Usage: "Disable ancient light chain data pruning (deprecated)", Usage: "Disable ancient light chain data pruning (deprecated)",
Category: flags.LightCategory, Category: flags.DeprecatedCategory,
} }
LightNoSyncServeFlag = &cli.BoolFlag{ LightNoSyncServeFlag = &cli.BoolFlag{
Name: "light.nosyncserve", Name: "light.nosyncserve",
Usage: "Enables serving light clients before syncing (deprecated)", Usage: "Enables serving light clients before syncing (deprecated)",
Category: flags.LightCategory, Category: flags.DeprecatedCategory,
} }
// Deprecated November 2023 // Deprecated November 2023
LogBacktraceAtFlag = &cli.StringFlag{ LogBacktraceAtFlag = &cli.StringFlag{
@ -138,19 +138,24 @@ var (
// Deprecated February 2024 // Deprecated February 2024
MinerNewPayloadTimeoutFlag = &cli.DurationFlag{ MinerNewPayloadTimeoutFlag = &cli.DurationFlag{
Name: "miner.newpayload-timeout", Name: "miner.newpayload-timeout",
Usage: "Specify the maximum time allowance for creating a new payload", Usage: "Specify the maximum time allowance for creating a new payload (deprecated)",
Value: ethconfig.Defaults.Miner.Recommit, Value: ethconfig.Defaults.Miner.Recommit,
Category: flags.MinerCategory, Category: flags.DeprecatedCategory,
} }
MinerEtherbaseFlag = &cli.StringFlag{ MinerEtherbaseFlag = &cli.StringFlag{
Name: "miner.etherbase", Name: "miner.etherbase",
Usage: "0x prefixed public address for block mining rewards", Usage: "0x prefixed public address for block mining rewards (deprecated)",
Category: flags.MinerCategory, Category: flags.DeprecatedCategory,
} }
MiningEnabledFlag = &cli.BoolFlag{ MiningEnabledFlag = &cli.BoolFlag{
Name: "mine", Name: "mine",
Usage: "Enable mining", Usage: "Enable mining (deprecated)",
Category: flags.MinerCategory, Category: flags.DeprecatedCategory,
}
MetricsEnabledExpensiveFlag = &cli.BoolFlag{
Name: "metrics.expensive",
Usage: "Enable expensive metrics collection and reporting (deprecated)",
Category: flags.DeprecatedCategory,
} }
) )

View file

@ -25,7 +25,6 @@ import (
"github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto" "github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/metrics"
"github.com/ethereum/go-ethereum/rlp" "github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode" "github.com/ethereum/go-ethereum/trie/trienode"
"github.com/holiman/uint256" "github.com/holiman/uint256"
@ -197,9 +196,8 @@ func (s *stateObject) GetCommittedState(key common.Hash) common.Hash {
if s.db.snap != nil { if s.db.snap != nil {
start := time.Now() start := time.Now()
enc, err = s.db.snap.Storage(s.addrHash, crypto.Keccak256Hash(key.Bytes())) enc, err = s.db.snap.Storage(s.addrHash, crypto.Keccak256Hash(key.Bytes()))
if metrics.EnabledExpensive { s.db.SnapshotStorageReads += time.Since(start)
s.db.SnapshotStorageReads += time.Since(start)
}
if len(enc) > 0 { if len(enc) > 0 {
_, content, _, err := rlp.Split(enc) _, content, _, err := rlp.Split(enc)
if err != nil { if err != nil {
@ -217,9 +215,8 @@ func (s *stateObject) GetCommittedState(key common.Hash) common.Hash {
return common.Hash{} return common.Hash{}
} }
val, err := tr.GetStorage(s.address, key.Bytes()) val, err := tr.GetStorage(s.address, key.Bytes())
if metrics.EnabledExpensive { s.db.StorageReads += time.Since(start)
s.db.StorageReads += time.Since(start)
}
if err != nil { if err != nil {
s.db.setError(err) s.db.setError(err)
return common.Hash{} return common.Hash{}
@ -283,9 +280,8 @@ func (s *stateObject) updateTrie() (Trie, error) {
return s.trie, nil return s.trie, nil
} }
// Track the amount of time wasted on updating the storage trie // Track the amount of time wasted on updating the storage trie
if metrics.EnabledExpensive { defer func(start time.Time) { s.db.StorageUpdates += time.Since(start) }(time.Now())
defer func(start time.Time) { s.db.StorageUpdates += time.Since(start) }(time.Now())
}
// The snapshot storage map for the object // The snapshot storage map for the object
var ( var (
storage map[common.Hash][]byte storage map[common.Hash][]byte
@ -370,9 +366,8 @@ func (s *stateObject) updateRoot() {
return return
} }
// Track the amount of time wasted on hashing the storage trie // Track the amount of time wasted on hashing the storage trie
if metrics.EnabledExpensive { defer func(start time.Time) { s.db.StorageHashes += time.Since(start) }(time.Now())
defer func(start time.Time) { s.db.StorageHashes += time.Since(start) }(time.Now())
}
s.data.Root = tr.Hash() s.data.Root = tr.Hash()
} }
@ -386,9 +381,8 @@ func (s *stateObject) commit() (*trienode.NodeSet, error) {
return nil, nil return nil, nil
} }
// Track the amount of time wasted on committing the storage trie // Track the amount of time wasted on committing the storage trie
if metrics.EnabledExpensive { defer func(start time.Time) { s.db.StorageCommits += time.Since(start) }(time.Now())
defer func(start time.Time) { s.db.StorageCommits += time.Since(start) }(time.Now())
}
// The trie is currently in an open state and could potentially contain // The trie is currently in an open state and could potentially contain
// cached mutations. Call commit to acquire a set of nodes that have been // cached mutations. Call commit to acquire a set of nodes that have been
// modified, the set can be nil if nothing to commit. // modified, the set can be nil if nothing to commit.

View file

@ -28,7 +28,6 @@ import (
"github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto" "github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/log" "github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/metrics"
"github.com/ethereum/go-ethereum/params" "github.com/ethereum/go-ethereum/params"
"github.com/ethereum/go-ethereum/trie" "github.com/ethereum/go-ethereum/trie"
"github.com/ethereum/go-ethereum/trie/trienode" "github.com/ethereum/go-ethereum/trie/trienode"
@ -495,9 +494,8 @@ func (s *StateDB) GetTransientState(addr common.Address, key common.Hash) common
// updateStateObject writes the given object to the trie. // updateStateObject writes the given object to the trie.
func (s *StateDB) updateStateObject(obj *stateObject) { func (s *StateDB) updateStateObject(obj *stateObject) {
// Track the amount of time wasted on updating the account from the trie // Track the amount of time wasted on updating the account from the trie
if metrics.EnabledExpensive { defer func(start time.Time) { s.AccountUpdates += time.Since(start) }(time.Now())
defer func(start time.Time) { s.AccountUpdates += time.Since(start) }(time.Now())
}
// Encode the account and update the account trie // Encode the account and update the account trie
addr := obj.Address() addr := obj.Address()
if err := s.trie.UpdateAccount(addr, &obj.data); err != nil { if err := s.trie.UpdateAccount(addr, &obj.data); err != nil {
@ -527,9 +525,8 @@ func (s *StateDB) updateStateObject(obj *stateObject) {
// deleteStateObject removes the given object from the state trie. // deleteStateObject removes the given object from the state trie.
func (s *StateDB) deleteStateObject(obj *stateObject) { func (s *StateDB) deleteStateObject(obj *stateObject) {
// Track the amount of time wasted on deleting the account from the trie // Track the amount of time wasted on deleting the account from the trie
if metrics.EnabledExpensive { defer func(start time.Time) { s.AccountUpdates += time.Since(start) }(time.Now())
defer func(start time.Time) { s.AccountUpdates += time.Since(start) }(time.Now())
}
// Delete the account from the trie // Delete the account from the trie
addr := obj.Address() addr := obj.Address()
if err := s.trie.DeleteAccount(addr); err != nil { if err := s.trie.DeleteAccount(addr); err != nil {
@ -561,9 +558,8 @@ func (s *StateDB) getDeletedStateObject(addr common.Address) *stateObject {
if s.snap != nil { if s.snap != nil {
start := time.Now() start := time.Now()
acc, err := s.snap.Account(crypto.HashData(s.hasher, addr.Bytes())) acc, err := s.snap.Account(crypto.HashData(s.hasher, addr.Bytes()))
if metrics.EnabledExpensive { s.SnapshotAccountReads += time.Since(start)
s.SnapshotAccountReads += time.Since(start)
}
if err == nil { if err == nil {
if acc == nil { if acc == nil {
return nil return nil
@ -587,9 +583,8 @@ func (s *StateDB) getDeletedStateObject(addr common.Address) *stateObject {
start := time.Now() start := time.Now()
var err error var err error
data, err = s.trie.GetAccount(addr) data, err = s.trie.GetAccount(addr)
if metrics.EnabledExpensive { s.AccountReads += time.Since(start)
s.AccountReads += time.Since(start)
}
if err != nil { if err != nil {
s.setError(fmt.Errorf("getDeleteStateObject (%x) error: %w", addr.Bytes(), err)) s.setError(fmt.Errorf("getDeleteStateObject (%x) error: %w", addr.Bytes(), err))
return nil return nil
@ -917,9 +912,8 @@ func (s *StateDB) IntermediateRoot(deleteEmptyObjects bool) common.Hash {
s.stateObjectsPending = make(map[common.Address]struct{}) s.stateObjectsPending = make(map[common.Address]struct{})
} }
// Track the amount of time wasted on hashing the account trie // Track the amount of time wasted on hashing the account trie
if metrics.EnabledExpensive { defer func(start time.Time) { s.AccountHashes += time.Since(start) }(time.Now())
defer func(start time.Time) { s.AccountHashes += time.Since(start) }(time.Now())
}
return s.trie.Hash() return s.trie.Hash()
} }
@ -1042,16 +1036,16 @@ func (s *StateDB) deleteStorage(addr common.Address, addrHash common.Hash, root
if err != nil { if err != nil {
return nil, nil, err return nil, nil, err
} }
if metrics.EnabledExpensive { // Report the metrics
n := int64(len(slots)) n := int64(len(slots))
slotDeletionMaxCount.UpdateIfGt(int64(len(slots))) slotDeletionMaxCount.UpdateIfGt(int64(len(slots)))
slotDeletionMaxSize.UpdateIfGt(int64(size)) slotDeletionMaxSize.UpdateIfGt(int64(size))
slotDeletionTimer.UpdateSince(start)
slotDeletionCount.Mark(n)
slotDeletionSize.Mark(int64(size))
slotDeletionTimer.UpdateSince(start)
slotDeletionCount.Mark(n)
slotDeletionSize.Mark(int64(size))
}
return slots, nodes, nil return slots, nodes, nil
} }
@ -1190,10 +1184,8 @@ func (s *StateDB) Commit(block uint64, deleteEmptyObjects bool) (common.Hash, er
} }
} }
// Write the account trie changes, measuring the amount of wasted time // Write the account trie changes, measuring the amount of wasted time
var start time.Time start := time.Now()
if metrics.EnabledExpensive {
start = time.Now()
}
root, set, err := s.trie.Commit(true) root, set, err := s.trie.Commit(true)
if err != nil { if err != nil {
return common.Hash{}, err return common.Hash{}, err
@ -1205,23 +1197,23 @@ func (s *StateDB) Commit(block uint64, deleteEmptyObjects bool) (common.Hash, er
} }
accountTrieNodesUpdated, accountTrieNodesDeleted = set.Size() accountTrieNodesUpdated, accountTrieNodesDeleted = set.Size()
} }
if metrics.EnabledExpensive { // Report the commit metrics
s.AccountCommits += time.Since(start) s.AccountCommits += time.Since(start)
accountUpdatedMeter.Mark(int64(s.AccountUpdated))
storageUpdatedMeter.Mark(int64(s.StorageUpdated))
accountDeletedMeter.Mark(int64(s.AccountDeleted))
storageDeletedMeter.Mark(int64(s.StorageDeleted))
accountTrieUpdatedMeter.Mark(int64(accountTrieNodesUpdated))
accountTrieDeletedMeter.Mark(int64(accountTrieNodesDeleted))
storageTriesUpdatedMeter.Mark(int64(storageTrieNodesUpdated))
storageTriesDeletedMeter.Mark(int64(storageTrieNodesDeleted))
s.AccountUpdated, s.AccountDeleted = 0, 0
s.StorageUpdated, s.StorageDeleted = 0, 0
accountUpdatedMeter.Mark(int64(s.AccountUpdated))
storageUpdatedMeter.Mark(int64(s.StorageUpdated))
accountDeletedMeter.Mark(int64(s.AccountDeleted))
storageDeletedMeter.Mark(int64(s.StorageDeleted))
accountTrieUpdatedMeter.Mark(int64(accountTrieNodesUpdated))
accountTrieDeletedMeter.Mark(int64(accountTrieNodesDeleted))
storageTriesUpdatedMeter.Mark(int64(storageTrieNodesUpdated))
storageTriesDeletedMeter.Mark(int64(storageTrieNodesDeleted))
s.AccountUpdated, s.AccountDeleted = 0, 0
s.StorageUpdated, s.StorageDeleted = 0, 0
}
// If snapshotting is enabled, update the snapshot tree with this new version // If snapshotting is enabled, update the snapshot tree with this new version
if s.snap != nil { if s.snap != nil {
start := time.Now() start = time.Now()
// Only update if there's a state transition (skip empty Clique blocks) // Only update if there's a state transition (skip empty Clique blocks)
if parent := s.snap.Root(); parent != root { if parent := s.snap.Root(); parent != root {
if err := s.snaps.Update(root, parent, s.convertAccountSet(s.stateObjectsDestruct), s.accounts, s.storages); err != nil { if err := s.snaps.Update(root, parent, s.convertAccountSet(s.stateObjectsDestruct), s.accounts, s.storages); err != nil {
@ -1235,9 +1227,7 @@ func (s *StateDB) Commit(block uint64, deleteEmptyObjects bool) (common.Hash, er
log.Warn("Failed to cap snapshot tree", "root", root, "layers", 128, "err", err) log.Warn("Failed to cap snapshot tree", "root", root, "layers", 128, "err", err)
} }
} }
if metrics.EnabledExpensive { s.SnapshotCommits += time.Since(start)
s.SnapshotCommits += time.Since(start)
}
s.snap = nil s.snap = nil
} }
if root == (common.Hash{}) { if root == (common.Hash{}) {
@ -1248,15 +1238,14 @@ func (s *StateDB) Commit(block uint64, deleteEmptyObjects bool) (common.Hash, er
origin = types.EmptyRootHash origin = types.EmptyRootHash
} }
if root != origin { if root != origin {
start := time.Now() start = time.Now()
set := triestate.New(s.accountsOrigin, s.storagesOrigin) set := triestate.New(s.accountsOrigin, s.storagesOrigin)
if err := s.db.TrieDB().Update(root, origin, block, nodes, set); err != nil { if err := s.db.TrieDB().Update(root, origin, block, nodes, set); err != nil {
return common.Hash{}, err return common.Hash{}, err
} }
s.originalRoot = root s.originalRoot = root
if metrics.EnabledExpensive { s.TrieDBCommits += time.Since(start)
s.TrieDBCommits += time.Since(start)
}
if s.onCommit != nil { if s.onCommit != nil {
s.onCommit(set) s.onCommit(set)
} }

View file

@ -21,7 +21,6 @@ import "github.com/urfave/cli/v2"
const ( const (
EthCategory = "ETHEREUM" EthCategory = "ETHEREUM"
BeaconCategory = "BEACON CHAIN" BeaconCategory = "BEACON CHAIN"
LightCategory = "LIGHT CLIENT"
DevCategory = "DEVELOPER CHAIN" DevCategory = "DEVELOPER CHAIN"
StateCategory = "STATE HISTORY MANAGEMENT" StateCategory = "STATE HISTORY MANAGEMENT"
TxPoolCategory = "TRANSACTION POOL (EVM)" TxPoolCategory = "TRANSACTION POOL (EVM)"

View file

@ -19,7 +19,7 @@ package metrics
// Config contains the configuration for the metric collection. // Config contains the configuration for the metric collection.
type Config struct { type Config struct {
Enabled bool `toml:",omitempty"` Enabled bool `toml:",omitempty"`
EnabledExpensive bool `toml:",omitempty"` EnabledExpensive bool `toml:"-"`
HTTP string `toml:",omitempty"` HTTP string `toml:",omitempty"`
Port int `toml:",omitempty"` Port int `toml:",omitempty"`
EnableInfluxDB bool `toml:",omitempty"` EnableInfluxDB bool `toml:",omitempty"`

View file

@ -24,23 +24,12 @@ import (
// for less cluttered pprof profiles. // for less cluttered pprof profiles.
var Enabled = false var Enabled = false
// EnabledExpensive is a soft-flag meant for external packages to check if costly
// metrics gathering is allowed or not. The goal is to separate standard metrics
// for health monitoring and debug metrics that might impact runtime performance.
var EnabledExpensive = false
// enablerFlags is the CLI flag names to use to enable metrics collections. // enablerFlags is the CLI flag names to use to enable metrics collections.
var enablerFlags = []string{"metrics"} var enablerFlags = []string{"metrics"}
// enablerEnvVars is the env var names to use to enable metrics collections. // enablerEnvVars is the env var names to use to enable metrics collections.
var enablerEnvVars = []string{"GETH_METRICS"} var enablerEnvVars = []string{"GETH_METRICS"}
// expensiveEnablerFlags is the CLI flag names to use to enable metrics collections.
var expensiveEnablerFlags = []string{"metrics.expensive"}
// expensiveEnablerEnvVars is the env var names to use to enable metrics collections.
var expensiveEnablerEnvVars = []string{"GETH_METRICS_EXPENSIVE"}
// Init enables or disables the metrics system. Since we need this to run before // Init enables or disables the metrics system. Since we need this to run before
// any other code gets to create meters and timers, we'll actually do an ugly hack // any other code gets to create meters and timers, we'll actually do an ugly hack
// and peek into the command line args for the metrics flag. // and peek into the command line args for the metrics flag.
@ -53,14 +42,6 @@ func init() {
} }
} }
} }
for _, enabler := range expensiveEnablerEnvVars {
if val, found := syscall.Getenv(enabler); found && !EnabledExpensive {
if enable, _ := strconv.ParseBool(val); enable { // ignore error, flag parser will choke on it later
log.Info("Enabling expensive metrics collection")
EnabledExpensive = true
}
}
}
for _, arg := range os.Args { for _, arg := range os.Args {
flag := strings.TrimLeft(arg, "-") flag := strings.TrimLeft(arg, "-")
@ -70,12 +51,6 @@ func init() {
Enabled = true Enabled = true
} }
} }
for _, enabler := range expensiveEnablerFlags {
if !EnabledExpensive && flag == enabler {
log.Info("Enabling expensive metrics collection")
EnabledExpensive = true
}
}
} }
} }