lotus/metrics/metrics.go

189 lines
7.5 KiB
Go
Raw Normal View History

package metrics
import (
2020-10-21 08:10:27 +00:00
"context"
"time"
"go.opencensus.io/stats"
"go.opencensus.io/stats/view"
"go.opencensus.io/tag"
2020-05-20 17:43:22 +00:00
rpcmetrics "github.com/filecoin-project/go-jsonrpc/metrics"
)
// Distribution
var defaultMillisecondsDistribution = view.Distribution(0.01, 0.05, 0.1, 0.3, 0.6, 0.8, 1, 2, 3, 4, 5, 6, 8, 10, 13, 16, 20, 25, 30, 40, 50, 65, 80, 100, 130, 160, 200, 250, 300, 400, 500, 650, 800, 1000, 2000, 5000, 10000, 20000, 50000, 100000)
// Global Tags
var (
Version, _ = tag.NewKey("version")
Commit, _ = tag.NewKey("commit")
PeerID, _ = tag.NewKey("peer_id")
FailureType, _ = tag.NewKey("failure_type")
Local, _ = tag.NewKey("local")
MessageFrom, _ = tag.NewKey("message_from")
MessageTo, _ = tag.NewKey("message_to")
MessageNonce, _ = tag.NewKey("message_nonce")
ReceivedFrom, _ = tag.NewKey("received_from")
2020-10-21 08:10:27 +00:00
Endpoint, _ = tag.NewKey("endpoint")
APIInterface, _ = tag.NewKey("api") // to distinguish between gateway api and full node api endpoint calls
)
// Measures
var (
LotusInfo = stats.Int64("info", "Arbitrary counter to tag lotus info to", stats.UnitDimensionless)
ChainNodeHeight = stats.Int64("chain/node_height", "Current Height of the node", stats.UnitDimensionless)
2020-09-06 04:32:05 +00:00
ChainNodeHeightExpected = stats.Int64("chain/node_height_expected", "Expected Height of the node", stats.UnitDimensionless)
ChainNodeWorkerHeight = stats.Int64("chain/node_worker_height", "Current Height of workers on the node", stats.UnitDimensionless)
MessagePublished = stats.Int64("message/published", "Counter for total locally published messages", stats.UnitDimensionless)
MessageReceived = stats.Int64("message/received", "Counter for total received messages", stats.UnitDimensionless)
MessageValidationFailure = stats.Int64("message/failure", "Counter for message validation failures", stats.UnitDimensionless)
MessageValidationSuccess = stats.Int64("message/success", "Counter for message validation successes", stats.UnitDimensionless)
BlockPublished = stats.Int64("block/published", "Counter for total locally published blocks", stats.UnitDimensionless)
BlockReceived = stats.Int64("block/received", "Counter for total received blocks", stats.UnitDimensionless)
BlockValidationFailure = stats.Int64("block/failure", "Counter for block validation failures", stats.UnitDimensionless)
BlockValidationSuccess = stats.Int64("block/success", "Counter for block validation successes", stats.UnitDimensionless)
BlockValidationDurationMilliseconds = stats.Float64("block/validation_ms", "Duration for Block Validation in ms", stats.UnitMilliseconds)
PeerCount = stats.Int64("peer/count", "Current number of FIL peers", stats.UnitDimensionless)
PubsubPublishMessage = stats.Int64("pubsub/published", "Counter for total published messages", stats.UnitDimensionless)
PubsubDeliverMessage = stats.Int64("pubsub/delivered", "Counter for total delivered messages", stats.UnitDimensionless)
PubsubRejectMessage = stats.Int64("pubsub/rejected", "Counter for total rejected messages", stats.UnitDimensionless)
PubsubDuplicateMessage = stats.Int64("pubsub/duplicate", "Counter for total duplicate messages", stats.UnitDimensionless)
2020-08-20 20:14:32 +00:00
PubsubRecvRPC = stats.Int64("pubsub/recv_rpc", "Counter for total received RPCs", stats.UnitDimensionless)
PubsubSendRPC = stats.Int64("pubsub/send_rpc", "Counter for total sent RPCs", stats.UnitDimensionless)
PubsubDropRPC = stats.Int64("pubsub/drop_rpc", "Counter for total dropped RPCs", stats.UnitDimensionless)
2020-10-21 08:10:27 +00:00
APIRequestDuration = stats.Float64("api/request_duration_ms", "Duration of API requests", stats.UnitMilliseconds)
)
var (
InfoView = &view.View{
Name: "info",
Description: "Lotus node information",
Measure: LotusInfo,
Aggregation: view.LastValue(),
TagKeys: []tag.Key{Version, Commit},
}
ChainNodeHeightView = &view.View{
Measure: ChainNodeHeight,
Aggregation: view.LastValue(),
}
2020-09-06 04:32:05 +00:00
ChainNodeHeightExpectedView = &view.View{
Measure: ChainNodeHeightExpected,
Aggregation: view.LastValue(),
}
ChainNodeWorkerHeightView = &view.View{
Measure: ChainNodeWorkerHeight,
Aggregation: view.LastValue(),
}
BlockReceivedView = &view.View{
Measure: BlockReceived,
Aggregation: view.Count(),
}
BlockValidationFailureView = &view.View{
Measure: BlockValidationFailure,
Aggregation: view.Count(),
TagKeys: []tag.Key{FailureType},
}
BlockValidationSuccessView = &view.View{
Measure: BlockValidationSuccess,
Aggregation: view.Count(),
}
BlockValidationDurationView = &view.View{
Measure: BlockValidationDurationMilliseconds,
Aggregation: defaultMillisecondsDistribution,
}
2020-08-28 06:25:50 +00:00
MessagePublishedView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: MessagePublished,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
MessageReceivedView = &view.View{
Measure: MessageReceived,
Aggregation: view.Count(),
}
MessageValidationFailureView = &view.View{
Measure: MessageValidationFailure,
Aggregation: view.Count(),
TagKeys: []tag.Key{FailureType, Local},
}
MessageValidationSuccessView = &view.View{
Measure: MessageValidationSuccess,
Aggregation: view.Count(),
}
PeerCountView = &view.View{
Measure: PeerCount,
Aggregation: view.LastValue(),
}
2020-08-28 06:25:50 +00:00
PubsubPublishMessageView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubPublishMessage,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
PubsubDeliverMessageView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubDeliverMessage,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
PubsubRejectMessageView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubRejectMessage,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
PubsubDuplicateMessageView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubDuplicateMessage,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
PubsubRecvRPCView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubRecvRPC,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
PubsubSendRPCView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubSendRPC,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
PubsubDropRPCView = &view.View{
2020-08-28 09:51:51 +00:00
Measure: PubsubDropRPC,
2020-08-28 06:25:50 +00:00
Aggregation: view.Count(),
}
2020-10-21 08:10:27 +00:00
APIRequestDurationView = &view.View{
Measure: APIRequestDuration,
Aggregation: defaultMillisecondsDistribution,
TagKeys: []tag.Key{APIInterface, Endpoint},
}
)
// DefaultViews is an array of OpenCensus views for metric gathering purposes
2020-05-20 17:43:22 +00:00
var DefaultViews = append([]*view.View{
InfoView,
ChainNodeHeightView,
2020-09-06 04:32:05 +00:00
ChainNodeHeightExpectedView,
ChainNodeWorkerHeightView,
BlockReceivedView,
BlockValidationFailureView,
BlockValidationSuccessView,
BlockValidationDurationView,
2020-08-28 06:25:50 +00:00
MessagePublishedView,
MessageReceivedView,
MessageValidationFailureView,
MessageValidationSuccessView,
PeerCountView,
2020-08-28 06:25:50 +00:00
PubsubPublishMessageView,
PubsubDeliverMessageView,
PubsubRejectMessageView,
PubsubDuplicateMessageView,
PubsubRecvRPCView,
PubsubSendRPCView,
PubsubDropRPCView,
2020-10-21 08:10:27 +00:00
APIRequestDurationView,
2020-07-28 14:37:29 +00:00
},
rpcmetrics.DefaultViews...)
// SinceInMilliseconds returns the duration of time since the provide time as a float64.
func SinceInMilliseconds(startTime time.Time) float64 {
return float64(time.Since(startTime).Nanoseconds()) / 1e6
2020-07-28 14:37:29 +00:00
}
2020-10-21 08:10:27 +00:00
// Timer is a function stopwatch, calling it starts the timer,
// calling the returned function will record the duration.
func Timer(ctx context.Context, m *stats.Float64Measure) func() {
start := time.Now()
return func() {
stats.Record(ctx, m.M(SinceInMilliseconds(start)))
}
}