I am struggling with High CPU on logstash cluster . I have 4 nodes with 8 CPUs and 8 GB memory . Memory seems to used half of it. but CPU is always above 90% .
Below is my logstash config
input {
beats {
client_inactivity_timeout => 86400
port => 5044
}
}
filter {
mutate {
gsub => [
# replace all forward slashes with underscore
#"fieldname", "/", "_",
# replace backslashes, question marks, hashes, and minuses
# with a dot "."
#"fieldname2", "[\\?#-]", "."
"message", "\t", " ",
"message", "\n", " "
]
}
grok {
match => { "message" => "\[%{TIMESTAMP_ISO8601:timestamp_match}\]%{SPACE}\:\|\:%{SPACE}%{WORD:level}%{SPACE}\:\|\:%{SPACE}%{USERNAME:hostname}%{SPACE}\:\|\:%{SPACE}%{GREEDYDATA:coidkey}%{SPACE}\:\|\:%{SPACE}%{GREEDYDATA:clientinfo}%{SPACE}\:\|\:%{SPACE}(%{IP:clientip})?%{SPACE}\:\|\:%{SPACE}%{GREEDYDATA:Url}%{SPACE}\:\|\:%{SPACE}%{JAVACLASS:class}%{SPACE}\:\|\:%{SPACE}%{USER:ident}%{SPACE}%{GREEDYDATA:msg}"}
}
}
output {
stdout { codec => rubydebug }
if "_grokparsefailure" in [tags] {
# write events that didn't match to a file
file { "path" => "/tmp/grok_failures.txt" }
} else{
elasticsearch {
hosts => "dfsyselastic.df.jabodo.com:9200"
user => "UN"
password => "PW"
index => "vicinio-%{+YYYY.MM.dd}"
document_type => "log"
}
}
}
Having multiple GREEDYDATA or DATA patterns in the middle of a grok expression can be very inefficient and use a lot of CPU. I would recommend trying to replace the ones you can with more targeted patterns, e.g. NOTSPACE.
Thanks Cristian for advise . I have tried few different patterns but for most of them it fails. My grok is working 100% at the moment . Can you please advise which field i can change in grok , my input looks like
[2017-06-16 11:25:02,111] :|: INFO :|: lvprdsndfe1.lv.jabodo.com :|: fd5dd89c4a434df9acd480c51fcbe :|: [BT:OT
HER, CC:99] :|:
10.90.53.181 :|: http://free.couponxplorer.com/index.jhtml :|: c.m.w.d.m.UnifiedLoggerW
rapper :|: - [ET: SplashPageServed, IP: 10.90.53.181]
[2017-06-16 11:25:02,184] :|: INFO :|: lvprdsndfe1.lv.jabodo.com :|: 59a0d92a2a404e467b972f555b046 :|: [BT:OT
HER, CC:99] :|:
10.90.53.181 :|: http://www.translationbuddy.com/index.jhtml :|: c.m.w.d.m.UnifiedLogge
rWrapper :|: - [ET: SplashPageServed, IP: 10.90.53.181]
[2017-06-16 11:25:02,186] :|: INFO :|: lvprdsndfe1.lv.jabodo.com :|: 12301ef5d74a4911971dddfdb72c :|: [BT:OT
HER, CC:99] :|:
10.90.53.181 :|: http://www.radiorage.com/index.jhtml :|: c.m.w.d.i.LocaleInterceptor
:|: - Error setting the locale based on the model:
[2017-06-16 11:25:02,269] :|: INFO :|: lvprdsndfe1.lv.jabodo.com :|: b4cfe445a551436f9ed86d5b724bf :|: [BT:OT
HER, CC:99] :|:
10.90.53.181 :|: http://free.fromdoctopdf.com/index.jhtml :|: c.m.w.d.i.LocaleIntercept
or :|: - Error setting the locale based on the model:
As Christian said, try NOTSPACE instead of DATA or GREEDYDATA. If that doesn't work, debug it systematically by using the simplest possible expression (\[%{TIMESTAMP_ISO8601:timestamp_match}\]) and building incrementally from there.
Replaced - %{GREEDYDATA:coidkey} with %{DATA:coidkey} for inputs which are coming into format fd5dd89c4a434df9acd480c51fcbe solved the issue . Replacing other GREEDYDATA with NOTSPACE are causing issues in grok filter and make it fail . Load has dropped 90% to ~ 15% .
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.