关于prometheus:开源signoz实现可观测性的分析

Sginoz是一个开源的APM(Application Performance Management)，它是利用可观测性的一个实际，应用OpenTelemetry协定，将traces/metrics/log交融在一起。

SigNoz is an open-source APM. It helps developers monitor their applications & troubleshoot problems, an open-source alternative to DataDog, NewRelic, etc.

本文重点关注traces和metrics方面的实现。

一. 整体架构

app应用opentelmetry-sdk编写代码(反对java/golang/python等)；
app配置将metrics和tracing发送至otel-collector；
otel-collector定制实现了:
- clickhouse-metrics-exporter: 将metrics发送至clickhouse；
- clickhouse-trace-exporter：将trace发送至clickhouse:
query-server负责从clickhouse中查问metrics和trace信息，并提供前端API；

二. otel-collector

1. app将metrics和trace发送至otel-collector

app中引入opentelemetry:

go.opentelemetry.io/otel

若应用gin作为http框架：

import "go.opentelemetry.io/contrib/instrumentation/github.com/gin-gonic/gin/otelgin"r := gin.Default()r.Use(otelgin.Middleware(serviceName))

app在运行时，指定ServiceName和collector地址：

# SERVICE_NAME=goApp INSECURE_MODE=true OTEL_EXPORTER_OTLP_ENDPOINT=192.168.0.1:4317 go run main.go

2. otel-collector的配置

otel-collector的配置，分为receivers、processors、exporters，
而后由service通过pipeline组织成残缺的性能。

receivers:  opencensus:    endpoint: 0.0.0.0:55678  otlp:    protocols:      grpc:        endpoint: 0.0.0.0:4317      http:        endpoint: 0.0.0.0:4318  jaeger:    protocols:      grpc:        endpoint: 0.0.0.0:14250      thrift_http:        endpoint: 0.0.0.0:14268processors:  batch:    send_batch_size: 10000    send_batch_max_size: 11000    timeout: 10sexporters:  clickhousetraces:    datasource: tcp://clickhouse:9000/?database=signoz_traces  clickhousemetricswrite:    endpoint: tcp://clickhouse:9000/?database=signoz_metrics    resource_to_telemetry_conversion:      enabled: trueservice:  telemetry:    metrics:      address: 0.0.0.0:8888  extensions:    - health_check    - zpages    - pprof  pipelines:    traces:      receivers: [jaeger, otlp]      processors: [signozspanmetrics/prometheus, batch]      exporters: [clickhousetraces]    metrics:      receivers: [otlp]      processors: [batch]      exporters: [clickhousemetricswrite]

3. otel-collector实现clickhouse的exporter

otel-collector将metrics和traces保留至clickhouse，为此，signoz在otel-collector中，实现了：

clickhousemetricsexporter;
clickhousetracesexporter;

存储在clickhouse中的metrics信息：

## time_series信息73048f54a32c :) select * from time_series_v2 limit 3;SELECT *FROM time_series_v2LIMIT 3Query id: 2540c93a-a396-42a7-b4e5-bed772ac00c5┌─metric_name─────────────────────────────────┬──────────fingerprint─┬──timestamp_ms─┬─labels────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┬─labels_object──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐│ otelcol_exporter_enqueue_failed_log_records │   136087390202145591 │ 1659598169051 │ {"__name__":"otelcol_exporter_enqueue_failed_log_records","exporter":"clickhousemetricswrite","instance":"localhost:8888","job":"otel-collector-metrics","service_instance_id":"b3d22c64-5c09-44d7-9367-9716747842db","service_version":"latest"} │ ('otelcol_exporter_enqueue_failed_log_records','','','','','','','clickhousemetricswrite','','','localhost:8888','otel-collector-metrics','','','','','','','','','','b3d22c64-5c09-44d7-9367-9716747842db','','','latest','','','','','') ││ otelcol_exporter_enqueue_failed_log_records │  7551020479515194203 │ 1659598139047 │ {"__name__":"otelcol_exporter_enqueue_failed_log_records","exporter":"prometheus","instance":"otel-collector:8888","job":"otel-collector","service_instance_id":"ed7abeec-d18f-4c4d-8bcb-64a9dea37976","service_version":"latest"}                │ ('otelcol_exporter_enqueue_failed_log_records','','','','','','','prometheus','','','otel-collector:8888','otel-collector','','','','','','','','','','ed7abeec-d18f-4c4d-8bcb-64a9dea37976','','','latest','','','','','')                ││ otelcol_exporter_enqueue_failed_log_records │ 11388370004215594241 │ 1659598139047 │ {"__name__":"otelcol_exporter_enqueue_failed_log_records","exporter":"clickhousetraces","instance":"otel-collector:8888","job":"otel-collector","service_instance_id":"ed7abeec-d18f-4c4d-8bcb-64a9dea37976","service_version":"latest"}          │ ('otelcol_exporter_enqueue_failed_log_records','','','','','','','clickhousetraces','','','otel-collector:8888','otel-collector','','','','','','','','','','ed7abeec-d18f-4c4d-8bcb-64a9dea37976','','','latest','','','','','')          │└─────────────────────────────────────────────┴──────────────────────┴───────────────┴───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

## samples信息73048f54a32c :) select * from samples_v2 limit 3SELECT *FROM samples_v2LIMIT 3Query id: 40dfc9f5-e76e-4ecc-8b50-c791fa9e041c┌─metric_name─────────────────────────────────┬────────fingerprint─┬──timestamp_ms─┬─value─┐│ otelcol_exporter_enqueue_failed_log_records │ 136087390202145591 │ 1659598160222 │     0 ││ otelcol_exporter_enqueue_failed_log_records │ 136087390202145591 │ 1659598220222 │     0 ││ otelcol_exporter_enqueue_failed_log_records │ 136087390202145591 │ 1659598280222 │     0 │└─────────────────────────────────────────────┴────────────────────┴───────────────┴───────┘3 rows in set. Elapsed: 0.016 sec.

存储在clickhouse中的traces信息：

73048f54a32c :) select  * from signoz_spans limit 3;SELECT *FROM signoz_spansLIMIT 3Query id: d9d84316-c6f0-4089-87a1-187132a295b1Connecting to database signoz_traces at localhost:9000 as user default.Connected to ClickHouse server version 22.4.5 revision 54455.┌─────────────────────timestamp─┬─traceID──────────────────────────┬─model──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐│ 2022-08-04 14:42:21.920273000 │ 00000000000000000000976d1a8457bf │ {"traceId":"00000000000000000000976d1a8457bf","spanId":"0c8946ee620b3b72","name":"HTTP GET /route","durationNano":35578000,"startTimeUnixNano":1659624141920273000,"serviceName":"route","kind":2,"references":[{"traceId":"00000000000000000000976d1a8457bf","spanId":"75509681018b9e65","refType":"CHILD_OF"}],"tagMap":{"client-uuid":"1f460e28dcd19136","component":"net/http","host.name":"93a9af66f59d","http.method":"GET","http.status_code":"200","http.url":"/route?dropoff=728%2C326\u0026pickup=232%2C495","ip":"172.18.0.3","opencensus.exporterversion":"Jaeger-Go-2.30.0","service.name":"route"},"event":["{\"timeUnixNano\":1659624141920308000,\"attributeMap\":{\"event\":\"HTTP request received\",\"level\":\"info\",\"method\":\"GET\",\"url\":\"/route?dropoff=728%2C326\\u0026pickup=232%2C495\"}}"]} ││ 2022-08-04 14:42:21.920759000 │ 00000000000000000000976d1a8457bf │ {"traceId":"00000000000000000000976d1a8457bf","spanId":"3a48059caa1e422a","name":"HTTP GET /route","durationNano":50933000,"startTimeUnixNano":1659624141920759000,"serviceName":"route","kind":2,"references":[{"traceId":"00000000000000000000976d1a8457bf","spanId":"37835dd357b1d10c","refType":"CHILD_OF"}],"tagMap":{"client-uuid":"1f460e28dcd19136","component":"net/http","host.name":"93a9af66f59d","http.method":"GET","http.status_code":"200","http.url":"/route?dropoff=728%2C326\u0026pickup=745%2C522","ip":"172.18.0.3","opencensus.exporterversion":"Jaeger-Go-2.30.0","service.name":"route"},"event":["{\"timeUnixNano\":1659624141920796000,\"attributeMap\":{\"event\":\"HTTP request received\",\"level\":\"info\",\"method\":\"GET\",\"url\":\"/route?dropoff=728%2C326\\u0026pickup=745%2C522\"}}"]} ││ 2022-08-04 14:42:21.920505000 │ 00000000000000000000976d1a8457bf │ {"traceId":"00000000000000000000976d1a8457bf","spanId":"2754f5b55262e27b","name":"HTTP GET /route","durationNano":71846000,"startTimeUnixNano":1659624141920505000,"serviceName":"route","kind":2,"references":[{"traceId":"00000000000000000000976d1a8457bf","spanId":"44b47488d0e49c98","refType":"CHILD_OF"}],"tagMap":{"client-uuid":"1f460e28dcd19136","component":"net/http","host.name":"93a9af66f59d","http.method":"GET","http.status_code":"200","http.url":"/route?dropoff=728%2C326\u0026pickup=462%2C723","ip":"172.18.0.3","opencensus.exporterversion":"Jaeger-Go-2.30.0","service.name":"route"},"event":["{\"timeUnixNano\":1659624141920525000,\"attributeMap\":{\"event\":\"HTTP request received\",\"level\":\"info\",\"method\":\"GET\",\"url\":\"/route?dropoff=728%2C326\\u0026pickup=462%2C723\"}}"]} │└───────────────────────────────┴──────────────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

三. Query Service

query-service服务中读取clickhouse的metrics和traces，给前端提供了HTTP接口。

另外，query-service还将prometheus引入作为sdk，实现了告警规定的创立及告警性能。

1. 查问metrics

API入口：/api/v1/query_range

// app/http_handler.gofunc (aH *APIHandler) RegisterRoutes(router *mux.Router) {    router.HandleFunc("/api/v1/query_range", ViewAccess(aH.queryRangeMetrics)).Methods(http.MethodGet)    ...}func (aH *APIHandler) queryRangeMetrics(w http.ResponseWriter, r *http.Request) {    query, apiErrorObj := parseQueryRangeRequest(r)    ...    ctx := r.Context()    ...    res, qs, apiError := (*aH.reader).GetQueryRangeResult(ctx, query)    ...    response_data := &model.QueryData{        ResultType: res.Value.Type(),        Result:     res.Value,        Stats:      qs,    }    aH.respond(w, response_data)}

为clickhouse实现了remote-read，故查问跟prometheus的查问代码差不多：

结构查问对象；
调用engine执行查问；

// app/clickhouseReader/reader.gofunc (r *ClickHouseReader) GetQueryRangeResult(ctx context.Context, query *model.QueryRangeParams) (*promql.Result, *stats.QueryStats, *model.ApiError) {    qry, err := r.queryEngine.NewRangeQuery(r.remoteStorage, query.Query, query.Start, query.End, query.Step)    ...    res := qry.Exec(ctx)    // Optional stats field in response if parameter "stats" is not empty.    var qs *stats.QueryStats    if query.Stats != "" {        qs = stats.NewQueryStats(qry.Stats())    }    qry.Close()    return res, qs, nil}

2. 查问traces

查问的url: /api/v1/traces/00000000000000001348070b30cd6aec

入口代码：/api/v1/traces

// app/http_handler.gofunc (aH *APIHandler) RegisterRoutes(router *mux.Router) {    ...    router.HandleFunc("/api/v1/traces/{traceId}", ViewAccess(aH.searchTraces)).Methods(http.MethodGet)    ...}func (aH *APIHandler) searchTraces(w http.ResponseWriter, r *http.Request) {    vars := mux.Vars(r)    traceId := vars["traceId"]    result, err := (*aH.reader).SearchTraces(r.Context(), traceId)    if aH.handleError(w, err, http.StatusBadRequest) {        return    }    aH.writeJSON(w, r, result)}

执行查问，通过clickhouse的查问语法实现：

// app/clickhouseReader/reader.gofunc (r *ClickHouseReader) SearchTraces(ctx context.Context, traceId string) (*[]model.SearchSpansResult, error) {    var searchScanReponses []model.SearchSpanDBReponseItem    query := fmt.Sprintf("SELECT timestamp, traceID, model FROM %s.%s WHERE traceID=$1", r.traceDB, r.spansTable)    err := r.db.Select(ctx, &searchScanReponses, query, traceId)    ...    searchSpansResult := []model.SearchSpansResult{        {            Columns: []string{"__time", "SpanId", "TraceId", "ServiceName", "Name", "Kind", "DurationNano", "TagsKeys", "TagsValues", "References", "Events", "HasError"},            Events:  make([][]interface{}, len(searchScanReponses)),        },    }    for i, item := range searchScanReponses {        var jsonItem model.SearchSpanReponseItem        json.Unmarshal([]byte(item.Model), &jsonItem)        jsonItem.TimeUnixNano = uint64(item.Timestamp.UnixNano() / 1000000)        spanEvents := jsonItem.GetValues()        searchSpansResult[0].Events[i] = spanEvents    }    return &searchSpansResult, nil}

3. 创立告警规定

页面上创立告警规定后，保留在sqlite中：

sqlite> select * from rules;1|2022-08-05 07:18:31.989405383+00:00|0|{"condition":{"compositeMetricQuery":{"builderQueries":{"A":{"queryName":"A","name":"A","formulaOnly":false,"metricName":"up","tagFilters":{"op":"AND","items":[]},"groupBy":["job"],"aggregateOperator":1,"expression":"A","disabled":false,"toggleDisable":false,"toggleDelete":false}},"promQueries":{"A":{"query":"up ","stats":"","name":"A","legend":"","disabled":false}},"queryType":3},"op":"3","matchType":"1"},"labels":{"severity":"warning"},"annotations":{"description":"A new alert"},"evalWindow":"5m0s","alert":"service_down","source":"http://192.168.0.1:3301/alerts/new","ruleType":"promql_rule"}sqlite>

创立告警规定的入口：

// app/http_handler.gofunc (aH *APIHandler) RegisterRoutes(router *mux.Router) {    ...    router.HandleFunc("/api/v1/rules", EditAccess(aH.createRule)).Methods(http.MethodPost)    ...}func (aH *APIHandler) createRule(w http.ResponseWriter, r *http.Request) {    decoder := json.NewDecoder(r.Body)    var postData map[string]string    err := decoder.Decode(&postData)    ...    apiErrorObj := (*aH.reader).CreateRule(postData["data"])    ...    aH.respond(w, "rule successfully added")}

创立规定的具体实现：

首先，向db插入一条rules记录；
而后，给prometheus减少一条alertRule；
最初，事务性的提交；

// app/clickhouseReader/reader.gofunc (r *ClickHouseReader) CreateRule(rule string) *model.ApiError {    tx, err := r.localDB.Begin()    ...    var lastInsertId int64    {        // 向db插入一条记录        stmt, err := tx.Prepare(`INSERT into rules (updated_at, data) VALUES($1,$2);`)        ...        defer stmt.Close()        result, err := stmt.Exec(time.Now(), rule)        ...        lastInsertId, _ = result.LastInsertId()        groupName := fmt.Sprintf("%d-groupname", lastInsertId)        // 让prometheus新增一条AlertRule        err = r.ruleManager.AddGroup(time.Duration(r.promConfig.GlobalConfig.EvaluationInterval), rule, groupName)        if err != nil {            tx.Rollback()            return &model.ApiError{Typ: model.ErrorInternal, Err: err}        }    }    err = tx.Commit()    if err != nil {        zap.S().Errorf("Error in committing transaction for INSERT to rules\n", err)        return &model.ApiError{Typ: model.ErrorInternal, Err: err}    }    return nil}

signoz还部署了alertmanager实例，当query-service中的prometheus有告警触发时，将告警发送到alertmanager。

signoz前端查问告警时，通过查问alertmanager实现：

HTTP GET /api/alertmanager/alerts?active=true&inhibited=true&silenced=false

四. 特点

对立存储和读取metrics和traces，均存储在clickhouse；
- 解决传统上metrics存prometheus，traces存Jager的痛点；
集成了prometheus的告警规定性能:
- 通过prometheus sdk引入，并未引入prometheus过程；
prometheus告警的去重、聚合、发送，通过alertmanager实例实现；

参考：

1.https://www.cnblogs.com/rongf...
2.signoz docs: https://signoz.io/docs/archit...
3.opentelmetry docs: https://opentelemetry.io/docs...