GoogleCloudPlatform/k8s-metadata-proxy

Name: k8s-metadata-proxy

Owner: Google Cloud Platform

Description: null

Created: 2017-10-12 19:34:24.0

Updated: 2018-05-20 23:08:17.0

Pushed: 2018-03-02 00:17:26.0

Homepage: null

Size: 234

Language: Go

GitHub Committers

User	Most Recent Commit	# Commits

Other Committers

User	Email	Most Recent Commit	# Commits

README

k8s-metadata-proxy

This repo contains a simple proxy for serving concealed metadata to container workloads running in kubernetes/kubernetes on a GCE VM.

Performance

This proxy has been benchmarked at requiring no more than 25Mi memory. With such a constraint and effectively no cpu constraint, it can serve 200 concurrent requests indefinitely at around 700 qps:

bectl describe pod metadata-proxy-v0.1-xxxxx
]
ainers:
tadata-proxy:
[...]
Limits:
  cpu: 500m
  memory: 25Mi
Requests:
  cpu: 500m
  memory: 25Mi
]

 -n 200000 -c 200 -H 'Metadata-Flavor:Google' http://127.0.0.1:988/computeMetadata/v1/instance/service-accounts/default/token
 is ApacheBench, Version 2.3 <$Revision: 1604373 $>
right 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
nsed to The Apache Software Foundation, http://www.apache.org/

]

er Software:        Metadata
er Hostname:        127.0.0.1
er Port:            988

ment Path:          /computeMetadata/v1/instance/service-accounts/default/token
ment Length:        202 bytes

urrency Level:      200
 taken for tests:   251.792 seconds
lete requests:      200000
ed requests:        0
l transferred:      86000000 bytes
 transferred:       40400000 bytes
ests per second:    794.31 [#/sec] (mean)
 per request:       251.792 [ms] (mean)
 per request:       1.259 [ms] (mean, across all concurrent requests)
sfer rate:          333.55 [Kbytes/sec] received

ection Times (ms)
          min  mean[+/-sd] median   max
ect:        0    0  17.5      0    1003
essing:    42  251  60.5    247     959
ing:       42  251  60.6    246     958
l:         42  252  62.5    247    1212

entage of the requests served within a certain time (ms)
%    247
%    271
%    288
%    298
%    329
%    357
%    396
%    423
%   1212 (longest request)

Under cpu constraint, the qps goes down to about 50, but the pod serves all requests successfully:

bectl describe pod metadata-proxy-v0.1-xxxxx
]
ainers:
tadata-proxy:
[...]
Limits:
  cpu: 30m
  memory: 25Mi
Requests:
  cpu: 30m
  memory: 25Mi
]

 -n 200000 -c 200 -H 'Metadata-Flavor:Google' http://127.0.0.1:988/computeMetadata/v1/instance/service-accounts/default/token
 is ApacheBench, Version 2.3 <$Revision: 1604373 $>
right 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
nsed to The Apache Software Foundation, http://www.apache.org/

]

er Software:        Metadata
er Hostname:        127.0.0.1
er Port:            988

ment Path:          /computeMetadata/v1/instance/service-accounts/default/token
ment Length:        202 bytes

urrency Level:      200
 taken for tests:   3592.015 seconds
lete requests:      200000
ed requests:        0
l transferred:      86000000 bytes
 transferred:       40400000 bytes
ests per second:    55.68 [#/sec] (mean)
 per request:       3592.015 [ms] (mean)
 per request:       17.960 [ms] (mean, across all concurrent requests)
sfer rate:          23.38 [Kbytes/sec] received

ection Times (ms)
          min  mean[+/-sd] median   max
ect:        0    1  32.4      0    1004
essing:   892 3590 633.2   3593    8102
ing:      892 3581 630.2   3504    7999
l:        899 3591 634.3   3595    8102

entage of the requests served within a certain time (ms)
%   3595
%   3798
%   3901
%   4000
%   4300
%   4602
%   5200
%   5699
%   8102 (longest request)

Above 200 concurrent requests, it starts resetting connections, but does not go above 25MiB memory.