How to get accurate Counter increase on Prometheus

0x00: Ultimate query

(
sum(max_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="service"}[24h]
                                    )
        ) 
- 
sum(min_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="service"}[24h]
                                    ) unless (min_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="service"}[24h]
                                                                                    ) unless kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="eservice"} offset 24h
                                                        )
        )
)
image.png

0x01 Why the increase is inaccurate?

the max value for last 24 hours:

sum(max_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="event-emitter-master"}[24h]))
image.png

the min value for last 24 hours:

we can see there are some data are inaccurate with gaps.

kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="event-emitter-master"}
image.png

An inaccurate instance:

kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}
image.png
increase(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png
max_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h]) - min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png
max_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png

An accurate instance:

kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}
image.png
max_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h]) - min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png

A consistent example

kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}
image.png
increase(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png
max_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h]) - min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
image.png

0x03 How to revise this?

We have 3 kinds of data:

  1. consistent instance whose min value starts from a normal positive value
  2. inconsistent instance but with 0 as min value
  3. inconsistent instance whose min value starts from a positive value - this one is actual we need to handle with.

Basically, we don’t have to do anything for 1 and 2.

But for the 3rd, we have to amend the min value as 0 instead, and must be careful, we also need to keep the normal one as normal.

I tried several approaches for this, including:

  1. clamp_min()
  2. OR vector(0)
  3. unless

Eventually, unless helps me out of there accurately.

vector1 unless vector2 operator means exclude vector2 from vector1 .

For the 3 kinds of data, we intent to:

  1. consistent instance whose min value starts from a normal positive value
    1. exclude nothing
  2. inconsistent instance but with 0 as min value
    1. exclude nothing
  3. inconsistent instance whose min value starts from a positive value - this one is actual we need to handle with.
    1. exclude itself

we do this to get nothing or itself:

min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id" ...}[24h]) 
unless 
kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id" ...} offset 24h

results for this:

  1. consistent instance
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
kafka_consumer_fetch_manager_records_consumed_total{kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"} offset 24h
image.png
  1. inconsistent instance but with 0 as min value
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"} offset 24h
image.png
  1. inconsistent instance whose min value starts from a positive value
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"} offset 24h
image.png

then do unless again with above results to exclude nothing or itself to get the accurate min value.

results for this:

  1. consistent instance
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
(min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"} offset 24h)
image.png
  1. inconsistent instance but with 0 as min value
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
(min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"} offset 24h)
image.png
  1. inconsistent instance whose min value starts from a positive value
min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
(min_over_time(kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"}[24h])
unless
kafka_consumer_fetch_manager_records_consumed_total{client_id="client_id", container="app", endpoint="http", instance="instance", job="job", kafka_version="3.7.0", kubernetes_namespace="engineering", kubernetes_pod_name="kubernetes_pod_name", namespace="engineering", pod="kubernetes_pod_name", service="service", topic="topic"} offset 24h)
image.png

last, we just need to get the difference with max and min sum them for all instance values.

Thus, we get the ultimate query as the very above shows:

(
sum(max_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="service"}[24h]
                                    )
        ) 
- 
sum(min_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="service"}[24h]
                                    ) unless (min_over_time(kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="service"}[24h]
                                                                                    ) unless kafka_consumer_fetch_manager_records_consumed_total{namespace="engineering", topic="topic", service="eservice"} offset 24h
                                                        )
        )
)

0x04 Discussions online

https://github.com/prometheus/prometheus/issues/6779

https://github.com/skaes/logjam-tools/pull/31

increase() in Prometheus sometimes doubles values: how to avoid?

Alternative tool:

https://github.com/VictoriaMetrics/VictoriaMetrics

最后编辑于 ：
©著作权归作者所有,转载或内容合作请联系作者
  • 人面猴
    序言：七十年代末，一起剥皮案震惊了整个滨河市，随后出现的几起案子，更是在滨河造成了极大的恐慌，老刑警刘岩，带你破解...
    沈念sama阅读 200,302评论 5赞 470
  • 死咒
    序言：滨河连续发生了三起死亡事件，死亡现场离奇诡异，居然都是意外死亡，警方通过查阅死者的电脑和手机，发现死者居然都...
    沈念sama阅读 84,232评论 2赞 377
  • 救了他两次的神仙让他今天三更去死
    文/潘晓璐 我一进店门，熙熙楼的掌柜王于贵愁眉苦脸地迎上来，“玉大人，你说我怎么就摊上这事。” “怎么了？”我有些...
    开封第一讲书人阅读 147,337评论 0赞 332
  • 道士缉凶录：失踪的卖姜人
    文/不坏的土叔 我叫张陵，是天一观的道长。 经常有香客问我，道长，这世上最难降的妖魔是什么？ 我笑而不...
    开封第一讲书人阅读 53,977评论 1赞 272
  • 港岛之恋（遗憾婚礼）
    正文 为了忘掉前任，我火速办了婚礼，结果婚礼上，老公的妹妹穿的比我还像新娘。我一直安慰自己，他们只是感情好，可当我...
    茶点故事阅读 62,920评论 5赞 360
  • 恶毒庶女顶嫁案：这布局不是一般人想出来的
    文/花漫 我一把揭开白布。 她就那样静静地躺着，像睡着了一般。 火红的嫁衣衬着肌肤如雪。 梳的纹丝不乱的头发上，一...
    开封第一讲书人阅读 48,194评论 1赞 277
  • 城市分裂传说
    那天，我揣着相机与录音，去河边找鬼。 笑死，一个胖子当着我的面吹牛，可吹牛的内容都是我干的。 我是一名探鬼主播，决...
    沈念sama阅读 37,638评论 3赞 390
  • 双鸳鸯连环套：你想象不到人心有多黑
    文/苍兰香墨 我猛地睁开眼，长吁一口气：“原来是场噩梦啊……” “哼！你这毒妇竟也来了？” 一声冷哼从身侧响起，我...
    开封第一讲书人阅读 36,319评论 0赞 254
  • 万荣杀人案实录
    序言：老挝万荣一对情侣失踪，失踪者是张志新（化名）和其女友刘颖，没想到半个月后，有当地人在树林里发现了一具尸体，经...
    沈念sama阅读 40,455评论 1赞 294
  • 护林员之死
    正文 独居荒郊野岭守林人离奇死亡，尸身上长有42处带血的脓包…… 初始之章·张勋 以下内容为张勋视角 年9月15日...
    茶点故事阅读 35,379评论 2赞 317
  • 白月光启示录
    正文 我和宋清朗相恋三年，在试婚纱的时候发现自己被绿了。 大学时的朋友给我发了我未婚夫和他白月光在一起吃饭的照片。...
    茶点故事阅读 37,426评论 1赞 329
  • 活死人
    序言：一个原本活蹦乱跳的男人离奇死亡，死状恐怖，灵堂内的尸体忽然破棺而出，到底是诈尸还是另有隐情，我是刑警宁泽，带...
    沈念sama阅读 33,106评论 3赞 315
  • 日本核电站爆炸内幕
    正文 年R本政府宣布，位于F岛的核电站，受9级特大地震影响，放射性物质发生泄漏。R本人自食恶果不足惜，却给世界环境...
    茶点故事阅读 38,696评论 3赞 303
  • 男人毒药：我在死后第九天来索命
    文/蒙蒙 一、第九天 我趴在偏房一处隐蔽的房顶上张望。 院中可真热闹，春花似锦、人声如沸。这庄子的主人今日做“春日...
    开封第一讲书人阅读 29,786评论 0赞 19
  • 一桩弑父案，背后竟有这般阴谋
    文/苍兰香墨 我抬头看了看天上的太阳。三九已至，却和暖如春，着一层夹袄步出监牢的瞬间，已是汗流浃背。 一阵脚步声响...
    开封第一讲书人阅读 30,996评论 1赞 255
  • 情欲美人皮
    我被黑心中介骗来泰国打工， 没想到刚下飞机就差点儿被人妖公主榨干…… 1. 我叫王不留，地道东北人。 一个月前我还...
    沈念sama阅读 42,467评论 2赞 346
  • 代替公主和亲
    正文 我出身青楼，却偏偏与公主长得像，于是被迫代替她去往敌国和亲。 传闻我的和亲对象是个残疾皇子，可洞房花烛夜当晚...
    茶点故事阅读 42,043评论 2赞 341