Monitoring Swift - OpenStack Summit May 2015, Vancouver

Monitoring your Swift cluster health

Christian SchwedePrincipal Software Engineer, Red HatOpenStack Summit Vancouver, May 2015

All good things come in threes

Swift Architecture MetricsBasic Monitoring

A short Swift overview

Proxy server PUT http://swift.com/v1/account/container/objectname

disk

Se

rve

r

Re

pl ic

ato

r

Au

di to

r

Up

da

ter

disk

disk disk

disk

Se

rve

r

Re

pl ic

ato

r

Au

di to

r

Up

da

ter

disk

disk disk

disk

Se

rve

r

Re

pl ic

ato

r

Au

di to

r

Up

da

ter

disk

disk disk

disk

Se

rve

r

Re

pl ic

ato

r

Au

di to

r

Up

da

ter

disk

disk disk

Basic Monitoring

Basic monitoring

● Services available?

curl http://server:port/healthcheck → “200 OK”

● Drives OK?

swift-drive-audit

● Checking replication, auditors, updaters, async_pending, ...

swift-recon

● Check data availability

swift-dispersion-report

● Audit a speci-c account/container?

swift-account-audit

Metrics

Collecting metrics

[28.381567892711667, 1430596860],

[26.190797487908338, 1430596920],

[28.006374835958336, 1430596980],

[28.425395488741668, 1430597040],

[27.621122305142339, 1430597100],

[30.334730943041667, 1430597160],

[31.013429164883334, 1430597220],

[28.327365745216325, 1430597280],

[27.783294518800002, 1430597340],

[27.764280637108341, 1430597400],

?

Collecting metrics

[28.381567892711667, 1430596860],

[26.190797487908338, 1430596920],

[28.006374835958336, 1430596980],

[28.425395488741668, 1430597040],

[27.621122305142339, 1430597100],

[30.334730943041667, 1430597160],

[31.013429164883334, 1430597220],

[28.327365745216325, 1430597280],

[27.783294518800002, 1430597340],

[27.764280637108341, 1430597400],

Swift, statsd & graphite interaction

object-server object-replicatorcollectd

statsd

carbon-cache

whisperdb

graphite-web

Installation & con-guration

Packages & important con-guration -les

● statsd

● python-carbon

● graphite-web

● graphite-web-selinux

● collectd

/etc/swift/*-server.conf

/etc/collectd.conf

/etc/statsd/con-g.js

/etc/carbon/storage-schemas.conf

/etc/carbon/storage-aggregation.conf

0

2

4

6

8

10

0 1 2 3 4 5 6 7 8 9 10

Sa

mple

valu

e

Time

Retention period & Aggregation method

Working with graphite-web

Selected Metrics

Thank you!

[email protected]

#openstack-swift: cschwede

@cschwede_de

References

● docs.openstack.org/developer/swift/admin_guide.html#cluster-telemetry-and-monitoring

● docs.openstack.org/developer/swift/admin_guide.html#reporting-metrics-to-statsd

● github.com/etsy/statsd/blob/master/docs/graphite.md

● graphite.readthedocs.org/en/latest/

● graphite.readthedocs.org/en/latest/functions.html

● collectd.org/documentation/manpages/collectd.conf.5.shtml#plugin_write_graphite

Used graphite functions

1a groupByNode(stats.counters.*.proxy-server.object.*.2*.xfer.count, 5, "avg")

1b groupByNode(stats.timers.*.proxy-server.object.*.2*.timing.median, 5, "avg")

2a substr(stats.timers.*.proxy-server.object.*.2*.timing.count, 5,6)

2b substr(stats.timers.*.proxy-server.object.*.4*.timing.count, 5,7)

3 substr(avg(*.cpu.*.cpu.wait), 4)

4 substr(lowestCurrent(*.df.*.df_complex.free,5), 0, 1)

5 groupByNode(stats.counters.*.object-replicator.partition.update.count.*.count, 2, "sum")

6 substr(*.counters.*.proxy-server.*.handoff_count.count, 4, 5)

7 groupByNode(*.filecount.*_async_pending.files, 0, "sum")