iLogtail

1756
下载
iLogtail 是阿里云日志服务(Log Service)中的一款高性能、可扩展的日志采集工具。它被设计用于实时收集、解析和传输各类日志数据,支持多种数据源和复杂的日志处理场景,广泛应用于分布式系统、Web 服务器、应用程序等环境中的日志管理。

自监控指标接口


iLogtail提供了指标接口,可以方便地为插件增加一些自监控指标,目前支持Counter,Gauge,String,Latency等类型。

接口:

https://github.com/alibaba/ilogtail/blob/main/pkg/pipeline/self_metric.go

实现:

https://github.com/alibaba/ilogtail/blob/main/pkg/helper/self_metrics_vector_imp.go

用户使用时需要引入pkg/helper包:

import (
"github.com/alibaba/ilogtail/pkg/helper"
)

创建指标

指标必须先定义后使用,在插件的结构体内声明具体指标。

type ProcessorRateLimit struct {
// other fields
context pipeline.Context
limitMetric pipeline.CounterMetric // 第一个指标
processedMetric pipeline.CounterMetric // 第二个指标
}

创建指标时,需要将其注册到iLogtail Context 的 MetricRecord 中,以便 iLogtail 能够采集上报数据,在插件的Init方法中,调用context 的 GetMetricRecord()方法来获取MetricRecord,然后调用helper.NewXXXMetricAndRegister函数去注册一个指标,例如:

metricsRecord := p.context.GetMetricRecord()
p.limitMetric = helper.NewCounterMetricAndRegister(metricsRecord, fmt.Sprintf("%v_limited", pluginType))
p.processedMetric = helper.NewCounterMetricAndRegister(metricsRecord, fmt.Sprintf("%v_processed", pluginType))

用户在声明一个Metric时可以还额外注入一些插件级别的静态Label,这是一个可选参数,例如flusher_http就把RemoteURL等配置进行上报:

metricsRecord := f.context.GetMetricRecord()
metricLabels := f.buildLabels()
f.matchedEvents = helper.NewCounterMetricAndRegister(metricsRecord, "http_flusher_matched_events", metricLabels...)

指标打点

不同类型的指标有不同的打点方法,直接调用对应Metric类型的方法即可。
Counter:

p.processedMetric.Add(1)

Latency:

tracker.ProcessLatency.Observe(float64(time.Since(startProcessTime)))

StringMetric:

sc.lastBinLogMetric.Set(string(r.NextLogName))

指标上报

iLogtial会自动采集所有注册的指标,默认采集间隔为60s,然后通过default_flusher上报,数据格式为LogGroup,格式如下:

{"Logs":[{"Time":0,"Contents":[{"Key":"http_flusher_matched_events","Value":"2.0000"},{"Key":"__name__","Value":"http_flusher_matched_events"},{"Key":"RemoteURL","Value":"http://testeof.com/write"},{"Key":"db","Value":"%{metadata.db}"},{"Key":"flusher_http_id","Value":"0"},{"Key":"project","Value":"p"},{"Key":"config_name","Value":"c"},{"Key":"plugins","Value":""},{"Key":"category","Value":"p"},{"Key":"source_ip","Value":"100.80.230.110"}]},{"Time":0,"Contents":[{"Key":"http_flusher_unmatched_events","Value":"0.0000"},{"Key":"__name__","Value":"http_flusher_unmatched_events"},{"Key":"db","Value":"%{metadata.db}"},{"Key":"flusher_http_id","Value":"0"},{"Key":"RemoteURL","Value":"http://testeof.com/write"},{"Key":"project","Value":"p"},{"Key":"config_name","Value":"c"},{"Key":"plugins","Value":""},{"Key":"category","Value":"p"},{"Key":"source_ip","Value":"100.80.230.110"}]},{"Time":0,"Contents":[{"Key":"http_flusher_dropped_events","Value":"0.0000"},{"Key":"__name__","Value":"http_flusher_dropped_events"},{"Key":"RemoteURL","Value":"http://testeof.com/write"},{"Key":"db","Value":"%{metadata.db}"},{"Key":"flusher_http_id","Value":"0"},{"Key":"project","Value":"p"},{"Key":"config_name","Value":"c"},{"Key":"plugins","Value":""},{"Key":"category","Value":"p"},{"Key":"source_ip","Value":"100.80.230.110"}]},{"Time":0,"Contents":[{"Key":"http_flusher_retry_count","Value":"2.0000"},{"Key":"__name__","Value":"http_flusher_retry_count"},{"Key":"RemoteURL","Value":"http://testeof.com/write"},{"Key":"db","Value":"%{metadata.db}"},{"Key":"flusher_http_id","Value":"0"},{"Key":"project","Value":"p"},{"Key":"config_name","Value":"c"},{"Key":"plugins","Value":""},{"Key":"category","Value":"p"},{"Key":"source_ip","Value":"100.80.230.110"}]},{"Time":0,"Contents":[{"Key":"http_flusher_flush_failure_count","Value":"2.0000"},{"Key":"__name__","Value":"http_flusher_flush_failure_count"},{"Key":"db","Value":"%{metadata.db}"},{"Key":"flusher_http_id","Value":"0"},{"Key":"RemoteURL","Value":"http://testeof.com/write"},{"Key":"project","Value":"p"},{"Key":"config_name","Value":"c"},{"Key":"plugins","Value":""},{"Key":"category","Value":"p"},{"Key":"source_ip","Value":"100.80.230.110"}]},{"Time":0,"Contents":[{"Key":"http_flusher_flush_latency_ns","Value":"2504448312.5000"},{"Key":"__name__","Value":"http_flusher_flush_latency_ns"},{"Key":"db","Value":"%{metadata.db}"},{"Key":"flusher_http_id","Value":"0"},{"Key":"RemoteURL","Value":"http://testeof.com/write"},{"Key":"project","Value":"p"},{"Key":"config_name","Value":"c"},{"Key":"plugins","Value":""},{"Key":"category","Value":"p"},{"Key":"source_ip","Value":"100.80.230.110"}]}],"Category":"","Topic":"","Source":"","MachineUUID":""}

一组LogGroup中会有多条Log,每一条Log都对应一条指标,其中 {"Key":"__name__","Value":"http_flusher_matched_events"} 是一个特殊的Label,代表指标的名字。

高级功能

动态Label

和Prometheus SDK类似,iLogtail也允许用户在自监控时上报可变Label,对于这些带可变Label的指标集合,iLogtail称之为MetricVector,
MetricVector同样也支持上述的指标类型,因此把上面的Metric看作是MetricVector不带动态Label的特殊实现。
用例:

type FlusherHTTP struct {
// other fields
context pipeline.Context
statusCodeStatistics pipeline.MetricVector[pipeline.CounterMetric] // 带有动态Label的指标
}

声明并注册MetricVector时,可以使用helper.NewXXXMetricVectorAndRegister方法,
需要将其带有哪些动态Label的Name也进行声明:

f.statusCodeStatistics = helper.NewCounterMetricVectorAndRegister(metricsRecord,
"http_flusher_status_code_count",
map[string]string{"RemoteURL": f.RemoteURL},
[]string{"status_code"},
)

打点时通过WithLabels API传入动态Label的值,拿到一个Metric对象,然后进行打点:

f.statusCodeStatistics.WithLabels(pipeline.Label{Key: "status_code", Value: strconv.Itoa(response.StatusCode)}).Add(1)

示例

可以参考内置的一些插件:

限流插件:
https://github.com/alibaba/ilogtail/blob/main/plugins/processor/ratelimit/processor_rate_limit.go

http flusher插件:
https://github.com/alibaba/ilogtail/blob/main/plugins/flusher/http/flusher_http.go


observability.cn Authors 2024 | Documentation Distributed under CC-BY-4.0
Copyright © 2017-2024, Alibaba. All rights reserved. Alibaba has registered trademarks and uses trademarks.
浙ICP备2021005855号-32