• Stars
    star
    614
  • Rank 73,061 (Top 2 %)
  • Language
    Go
  • License
    Apache License 2.0
  • Created about 4 years ago
  • Updated 7 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Kvass is a Prometheus horizontal auto-scaling solution , which uses Sidecar to generate special config file only containes part of targets assigned from Coordinator for every Prometheus shard.

中文版

Kvass is a Prometheus horizontal auto-scaling solution , which uses Sidecar to generate special config file only containes part of targets assigned from Coordinator for every Prometheus shard.

Coordinator do service discovery, Prometheus shards management and assign targets to each of shard. Thanos (or other storage solution) is used for global data view.

Go Report Card Build


Table of Contents

Overview

Kvass is a Prometheus horizontal auto-scaling solution with following features.

  • Easy to use
  • Tens of millions series supported (thousands of k8s nodes)
  • One prometheus configuration file
  • Auto scaling
  • Sharding according to the actual target load instead of label hash
  • Multiple replicas supported

Design

image-20201126031456582

Components

Coordinator

See flags of Coordinator code

  • Coordinator loads origin config file and do all prometheus service discovery
  • For every active target, Coordinator do all "relabel_configs" and explore target series scale
  • Coordinaotr periodly try assgin explored targets to Sidecar according to Head Block Series of Prometheus.

image-20201126031409284

Sidecar

See flags of Sidecar code

  • Sidecar receive targets from Coordinator.Labels result of target after relabel process will also be send to Sidecar.

  • Sidecar generate a new Prometheus config file only use "static_configs" service discovery, and delete all "relabel_configs".

  • All Prometheus scraping request will be proxied to Sidecar for target series statistics.

    image-20201126032909776

Kvass + Thanos

Since the data of Prometheus now distribut on shards, we need a way to get global data view.

Thanos is a good choice. What we need to do is adding Kvass sidecar beside Thanos sidecar, and setting up a Kvass coordinator.

image-20201126035103180

Kvass + Remote storage

If you want to use remote storage like influxdb, just set "remote write" in origin Prometheus config.

Multiple replicas

Coordinator uses the label selector to select StatefulSets. Each StatefulSet is a replica. Management of targets between replicas is independent.

--shard.selector=app.kubernetes.io/name=prometheus

Targets transfer

There are scenarios where we need to move an assigned Target from one shard to another (for example, to de-pressure a shard).

In order to keep scraping normally, the Target transfer is divided into the following steps.

  • Mark the state of the Target in the original shard as IN_TRANSFER, and assign the Target to another shard with the state as Normal.
  • Wait for the Target to be scraped by both shards for at least 3 times.
  • Delete Target from the original shard.

Shard de-pressure

The series of Target products may increase over time, and the head series of shard may exceeding the threshold, such as the newly added K8S node, whose cadvisor data size may increase as POD is scheduled.

When the head series of a shard exceeds a certain proportion of the threshold, the Coordinator will attempt to de-pressurize the shard. That is, according to the proportion of the shard exceeding the threshold, the Coordinator will transfer some targets from the shard to other free shards. The higher the proportion of the shard exceeding the threshold, the more targets will be transferred.

Shard scaling down

Scaling down will only start at the largest shard.

When all targets on the highest-numbered shard can be migrated to other shards, a transfer attempt is made, which empties the highest-numbered shard.

When the shard is emptied, the shard becomes idle, and wiil be deleted after a certain period of time (waiting for the shard data to be deleted or uploaded to object storage).

You can set the idle time of Coordinaor using the following parameters and turn off shrinkage by set it to 0.

- --shard.max-idle-time=3h
- --shard.max-idle-time=0 // default

If StatefulSet is used to manage shards, you can add a parameter that will allow the Coordinator to automatically remove the PVC when the shard is removed

- --shard.delete-pvc=true // default

Limit shards number

The maximum and minimum number of shards for the Coordinator can be limited by setting the following flags. Note that if the minimum number of sharding is set, then you will only start Coordinate if the number of sharding available is at least the minimum number.

--shard.max-shard=99999 //default
--shard.min-shard=0 //default

Target scheduling strategy

If --shard.max-idle-time!=0, both new and migrated targets will be assigned to lower-numbered shards in preference.

If --shard.max-idle-time=0, it will be randomly allocated to the shard with space, which is especially useful with the --shard.min-shard flag.

Demo

There is a example to show how Kvass work.

git clone https://github.com/tkestack/kvass

cd kvass/example

kubectl create -f ./examples

you can found a Deployment named "metrics" with 6 Pod, each Pod will generate 10045 series (45 series from golang default metrics) metircs。

we will scrape metrics from them。

image-20200916185943754

the max series each Prometheus Shard can scrape is a flag of Coordinator Pod.

in the example case we set to 30000.

--shard.max-series=30000

now we have 6 target with 60000+ series and each Shard can scrape 30000 series,so need 3 Shard to cover all targets.

Coordinator automaticly change replicate of Prometheus Statefulset to 3 and assign targets to them.

image-20210618184014599

only 20000+ series in prometheus_tsdb_head of one Shard

image-20200917112924277

but we can get global data view use thanos-query

image-20210618184029590

we can also get currect record rule result.

image-20210618184146772

Best practice

Flag values suggestion

The memory useage of every Prometheus is associated with the max head series.

The recommended "max series" is 750000, set Coordinator flag

--shard.max-series=750000

The memory request of Prometheu with 750000 max series is 8G.

The flag --store.path of Sidecar must be set to a persistent stores since Sidecar will save scraping targets to disk and reload when it launchs.

License

Apache License 2.0, see LICENSE.

More Repositories

1

tke

Native Kubernetes container management platform supporting multi-tenant and multi-cluster
Go
1,474
star
2

gpu-manager

Go
823
star
3

kstone

Kstone is an etcd management platform, providing cluster management, monitoring, backup, inspection, data migration, visual viewing of etcd data, and intelligent diagnosis.
Go
526
star
4

vcuda-controller

C
494
star
5

gpu-admission

Go
128
star
6

tapp

Another great app kind for Kubernetes!
Go
113
star
7

galaxy

Providing high-performance network for Kubernetes
Go
109
star
8

image-transfer

Go
74
star
9

cron-hpa

Cron Horizontal Pod Autoscaler(CronHPA)
Go
60
star
10

lb-controlling-framework

A controller that helps you manipulate arbitrary load balancers
Go
56
star
11

charts

Mustache
40
star
12

docs

https://tkestack.github.io/docs/
37
star
13

go-nvml

Go
31
star
14

kube-jarvis

Go
28
star
15

tke-k8s-distro

Repository of Tencent TKE Kubernetes Distribution
Shell
28
star
16

yarn-opterator

An kubernetes opterator to manager the yarn nodemanager, which can be used to deploy yarn on kubernetes.
Go
22
star
17

volume-decorator

Go
14
star
18

csi-operator

Go
14
star
19

kstone-dashboard

Kstone Dashboard is a web-based UI for managing Kstone. It can help users to manage etcd based kstone and enjoy monitoring, inspection, visualization, backup and other features.
TypeScript
14
star
20

kstone-etcd-operator

kstone-etcd-operator is a subproject of etcd cluster management platform kstone. It's inspired by etcd-operator. And has more complete support for persistent storage and better disaster tolerance.
Go
11
star
21

toc

7
star
22

knitnet-operator

Operator for support direct networking between Pods and Services in different Kubernetes clusters
Go
7
star
23

gpu-exporter

C
5
star
24

ronqir

CRI runtime in Rust
5
star
25

gc-controller

Go
4
star
26

kube-event-watcher

Go
4
star
27

aia-ip-controller

Aia-ip-controller is a Kubernetes application, which make it possible for Tencent Kubernetes Engine(TKE) customers to leverage Tencent Cloud's AIa Service to accelerate network with global coverage.
Go
4
star
28

tke-extend-network-controller

A Network Controller for TKE
Go
3
star
29

web

TKEStack website
HTML
1
star