Sunday, November 9, 2014

Logstash Cookbook


These are the questions that I once had when I was working on Logstash to analyze my application’s logs, I wish there were a compiled list of common questions and answers, which could have saved me many hours of googling and experimenting. This cookbook works for Logstash 1.4.2.

 

No such command "-e"

The HelloWorld example in http://logstash.net/docs/1.4.2/tutorials/getting-started-with-logstash doesn’t work in windows:

c:\logstash-1.4.2\data>..\bin\logstash -e 'input { stdin { } } output { stdout {
} }'
"LS_HOME=" c:\logstash-1.4.2
No such command "-e"
Usage: logstash <command> [command args]
Run a command with the --help flag to see the arguments.

-e doesn’t work in windows, it works in linux. But HelloWorld is not that interesting after all, there is no need crying over it. To try out Logstash in windows, write up your actions in a configuration file, and use logstash agent -f configfile.

Logstash has a simple structure: it reads from loggings from various inputs, pass them down through filters for all kinds of processing, and send them to different outputs. This page http://logstash.net/docs/1.4.2/ lists the inputs, filters and outputs currently supported. 

 

How to drop a logging from the filter chain?


Use drop filter which will drop the logging from the filter chain and from outputs. For example, the following tries to match a logging to a grok pattern %{LOGPATTERN}, if doesn’t match, it drops the logging:

grok {
     match => { "message" => "%{LOGPATTERN}" }  
      tag_on_failure => ["logging not confirm to %{LOGPATTERN}"]    
    }
  
   if "logging not confirm to %{ LOGPATTERN }" in [tags] {   
     drop { }
   }
}

How to skip the rest of filters and only to one output?

If there are multiple drops, I won’t be able to know which drop drops the logging, it would be nice if instead of drop, there is a filter that skips the rest of filters and flow directly to stdout output.

To my knowledge, there is no such filter. And although I have little knowledge about the inner workings of Logstash, it seems not an easy task.

How to debug Logstash itself?

I use three techniques.

  •    Use –vvv to output most verbose Logstash debug information.

bin\logstash agent -f conf/test1.conf –vvv

  • Use stdout to output results:

output { 
  stdout { codec => rubydebug }
}
  • Use -t to test the config file
          bin\logstash agent -t -f conf/test1.conf

Why sometimes Logstash doesn’t seem to be doing anything when processing a static file?

 If you use file input to process a static file, you might notice that the first time you run Logstash, everything works ok, you see results outputted in stdout (if you configured stdout), but the second time you run Logstash, Logstash doesn’t seem to be doing anything, it seems to be stopped and wait silently for something to happen. 

This is because file records the place where it finished processing the file last time and starts from there, it is similar to “tail -0F”. So if the size of the file doesn’t increase after the last time, Logstash will do nothing: it is waiting for new things to be written into the log file. File writes the last pace to a sincedb file

This is quite inconvenient in testing, as in testing, one is usually experimenting different things with the same log file and wants to rerun Logstash to verify. To overcome this, specify the sincedb file and before running Logstash, manually delete the sincedb file.

input {
  file {
    path => "C:/logstash-1.4.2/data/test.log"
    start_position => "beginning"
sincedb_path => "C:/logstash-1.4.2/data/.test.log.sincedb"
   }
}

Logstash can't tail multiple files in windows?

This is a bug in Logstash https://logstash.jira.com/browse/LOGSTASH-2229, which has been solved https://github.com/jordansissel/ruby-filewatch. To patch Logstash 1.4.2 with the solution:

  1.  Download ruby-filewatch-master.zip from https://github.com/jordansissel/ruby-filewatch
  2.  Unzip it
  3. Copy files under ruby-filewatch-master\lib\filewatch to logstash-1.4.2\vendor\bundle\jruby\1.9\gems\filewatch-0.5.1\lib
  4. copy files under ruby-filewatch-master\bin to logstash-1.4.2\vendor\bundle\jruby\1.9\gems\filewatch-0.5.1\bi
  5. copy ruby-filewatch-master\java\JRubyFileExtension.jar into logstash-1.4.2\lib

Logstash mixes loggings from different files in Windows?

This is another bug of Logstash when running in the charming Windows. Some nice person has posted a temporary solution in https://logstash.jira.com/browse/LOGSTASH-2207, two files need to be changed:

  •  logstash-1.4.2\lib\logstash\inputs\file.rb
  • logstash-1.4.2\lib\logstash\codecs\ multiline.rb

How to check if a field exists or not?


#check if exists
if [event]{
}
   #check if not exists
if ![event]{
}

How to check if a tag exists

tag is stored in an array, so to check if a specific tag exists, use in:

if "not a cluster lost event" in [tags] {
     mutate {
        add_field => { "event" => "other events" }
     }
   }

How to do case insensitive comparison?

Logstash is written in Ruby, and Ruby supports regex insensitive mode in the form of /regex/i, such as:

if "ABC" =~ /abc/i
  puts "true"
end
however, Logstash doesn’t support this form, for example, Logstash will complain the following configuration is illegal:
if [message]=~ /abc/i {
}
Instead, use (?i) form to do case insensitive comparison:
if [message] =~ /(?i)abc/ {
}

How to capture named regex without defining them in grok pattern files?

Sometimes you do not want to define named regex in a grok pattern file, perhaps because it is not convenient to change the grok pattern file every time the named regex changes, you can use (?<name>regex). For example, if you see a logging “fail to hear from k1 for over 3 minutes”, you know an event “cluster failed-lost k1” has happened, to do so:


grok {
    match => { "message" => "fail to hear from (?<nodelost>.*?) for over 3 minutes" }     
    add_field => { "event" => "cluster failed-lost %{nodelost}" }
    tag_on_failure => ["not a cluster lost event"]                             
}

You might be tempted to use this technique directly in if comparison. The comparison will be true, but you won’t get the named capture:


if [message] =~ /fail to hear from (?<nodelost>.*?) for over 3 minutes/ {
   mutate {
          add_field => { "event" => "cluster failed-lost %{nodelost}" }  
  }     
}           

The event field you get is

"event" => "cluster failed-lost %{nodelost}"

This is because the captured %{nodelost} is not passed into the mutate plugin.  In contrast, the named captures from grok becomes fields, which are passed down the filter chain and to the outputs.

How to debug grok patterns?



http://grokdebug.herokuapp.com/ is a great tool to debug grok patterns, however, at first, it is not that easy to figure out how to use it, so I snapped a screenshot which should make it clear. 
Note, sometimes after you input the pattern or the string, nothing happens, and you wonder if the site is freezing or if nothing is captured by the pattern. When in doubt, you can delete the pattern, wait a few seconds, and input the pattern again.

How to add fields?

Many filters can create fields. I usually use grok and mutate.

  • grok:

The named patterns in grok will automatically become fields, which will passed down the filter chain and be used.

  • mutate:

mutate has many operations that can transform a logging, so I use it often.

 

mutate gsub doesn’t evaluate fields?

This is an issue https://github.com/elasticsearch/logstash/issues/1529, which has been solved in https://github.com/elasticsearch/logstash/pull/1552/files. You can patch up logstash-1.4.2\lib\logstash\filters\mutate.rb with the solution. 

In my opinion, the solution is not complete though. With this solution, gsub will evaluate fields in the replacement part, but not the grok patterns in the matching part. For example:


#this doesn’t work: gsub doesn’t evaluate %{JAVAEXCEPTION} – the string matching part
gsub => [ "msg", "%{JAVAEXCEPTION}", "%{javaexception0} %{frommyapp}"]
#this works: gsub evaluates fields in the replacement
gsub => [ "msg", "(?:[a-zA-Z0-9-]+\.)+[A-Za-z0-9$_]+Exception", "%{javaexception0} %{frommyapp}"]

How to reference captured groups in gsub?

Use \1, \2.

For example, to change a timestamp from 2016-05-03-06:22:36.641Z to 2016-05-03T06:22:36.641Z, use:



gsub => [“timestamp","-(\d\d):","T\1:"]

How to deal with multiline loggings?

A logging containing multiple lines is quite often, think of a stacktrace. You can use multiline codec to deal with multiline loggings. For example, typically a logging is logged with a timestamp, which follows a specific pattern. If a timestamp following the pattern doesn’t exist in a line, then the line is not the beginning of a new logging: it belongs to the previous logging. 


#define the logging timestamp pattern
YEARMONDAY %{YEAR}[/-](%{MONTHNUM})[/-](%{MONTHDAY})
TZ0 (?:[a-zA-Z]{3,4})
LOGDATE %{YEARMONDAY}-%{TIME} %{TZ0}

#within input, use multiline codec
input {
  file {
    path => "C:/logstash-1.4.2-0/data/test/**/serverLog.txt"
    start_position => "beginning"
    sincedb_path => "C:/logstash-1.4.2-0/data/.test.log.sincedb"
      
    codec => multiline {    
      pattern => "%{LOGDATE}:"
      negate => true
      what => previous
    }
  }
}

Why does Logstash ElasticSearch output create indices with wrong dates?

By default,  for each logging passing from inputs, through filters and out to outputs, Logstash creates a default @timestamp field, which is the time that Logstash processes the logging. Logstash will also use the date part of the @timestamp to create an ElasticSearch index. For example, the test.log file contains two loggings:


ERROR server:ClusterMonitor:2013/10/19-23:51:09.637 CDT: fail to hear from k2 for over 3 minutes
ERROR server:ClusterMonitor:2013/10/19-23:51:39.792 CDT: fail to hear from k3 for over 3 minutes


Without any filters, Logstash will create these two documents:
{
     "message" => "ERROR server:ClusterMonitor:2013/10/19-23:51:09.637 CDT: fail to hear from k2 for over 3 minutes\r",
     "@version" => "1",
     "@timestamp" => "2014-11-08T10:17:52.605Z",
     "host" => "Apple",
      "path" => "C:/logstash-1.4.2-0/data/test.log"
}
{
       "message" => "ERROR server:ClusterMonitor:2013/10/19-23:51:39.792 CDT: fail to hear from k3 for over 3 minutes\r",
      "@version" => "1",
       "@timestamp" => "2014-11-08T10:17:52.605Z",
       "host" => "Apple",
       "path" => "C:/logstash-1.4.2-0/data/test.log"
}
Notice @timestamp field is the time Logstash processes the logging. Logstash will also create ElasticSearch index named logstash-2014.11.08.  This is most likely not what you want, you want to see index “logstash-2013.10.19” containing the two loggings logged on 2013/10/19. 

To accomplish this, you need to capture the timestamp from the loggings (this can be done via a grok pattern, or via the technique described in “how to capture named regex”), and use date filter to make the timestamp as the timestamp for the document and for the index name. 
  •   Capture the timestamp from the loggings
Use the same example, using the existing grok pattern, it is easy to create a pattern to match logging timestamps such as “2013/10/19-23:51:09.637 CDT”:

#define the pattern
YEARMONDAY %{YEAR}[/-](%{MONTHNUM})[/-](%{MONTHDAY})
TZ0 (?:[a-zA-Z]{3,4})
LOGDATE %{YEARMONDAY}-%{TIME} %{TZ0}
   
#use grok to capture timestamp from loggings
      grok {
          match => { "message" => "%{LOGDATE:timestamp}" }
         tag_on_failure => ["logging not confirm to %{PPMLOG} pattern"]     
     }
  •   Use date filter
date{          
       match => [ "timestamp" , "yyyy/MM/dd-HH:mm:ss.SSS z" ]
}


date match takes the “timestamp” field, which follows the format “yyyy/MM/dd-HH:mm:ss.SSS z  -- this is very inconvenient: you just had to figure out LOGDATE regex format using a regex pattern, now you had to further figure out a date format following a different standards in joda.time.format.DateTimeFormat

It seems Logstash is only able to recognize some North American timezones and fails to deal with Asia and European timezones. 

For example, “CST” can be China Standard Time or Central Standard Time, Logstash will treat is Central Standard Time. To properly interpret a timestamp string with a timezone, you need to:


  •   First capture the timezone part using Grok pattern, for example:

LOGDATE %{YEARMONDAY}-%{TIME} %{TZ0:tz}

  • Remove the timezone part from the timestamp string

mutate{
      gsub => ["timestamp","\s*[A-Za-z]+",""]
}



  

if [tz]=~ /CST/{
             date{                    
                    timezone=>"Asia/Shanghai"
                    match => [ "timestamp" , "yyyy-MM-dd-HH:mm:ss.SSS" ]          
             }
             mutate{
                    gsub => ["timestamp","$"," %{tz}"]                 
             }
}

Now the output becomes:
{
       "message" => "ERROR server:ClusterMonitor:2013/10/19-23:51:09.637 CDT: fail to hear from k2 for over 3 minutes\r",
      "@version" => "1",
       "@timestamp" => "2013-10-20T04:51:09.637Z",
          "host" => "Apple",
          "path" => "C:/logstash-1.4.2-0/data/test.log",
          "timestamp" => "2013/10/19-23:51:09.637 CDT
}
{
       "message" => "ERROR server:ClusterMonitor:2013/10/19-23:51:39.792 CDT: fail to hear from k3 for over 3 minutes\r",
     "@version" => "1",
    "@timestamp" => "2013-10-20T04:51:39.792Z",
      "host" => "CHENPING4",
      "path" => "C:/logstash-1.4.2-0/data/test.log",
     "timestamp" => "2013/10/19-23:51:39.792 CDT"
}
Notice @timestamp field now has the same value as "timestamp", but they are of different timezones.
@timestamp is in UTC timezone. 
 
The index created is named logstash-2013.10.20.

Note ElasticSearch seems only able to recognize North American timezones as well, and worse still,  ElasticSearch creates indices but fails to create documents silently, without any errors in ElasticSearch log files or error responses returned to Logstash. I used to have a field timestamp mapped to date type, if the timestamp is, for example, of EDT (an European timezone), no documents will be created even though indices were created. In the end, I gave up on timestamp  field being date type and return it to String type.

How to process only a part of the log files?



Sometimes I got a huge log file spanning many months, while I am only interested in getting the most recent month’s loggings. I can do so by filtering loggings that are logged before a certain time. 


mutate {
             add_field => { "startdate" => "2014-07-01T00:00:00 GMT" }
       }
date {
       //convert startdate from a string to a date
             match => [ "startdate" , "yyyy-MM-dd'T'HH:mm:ss z" ]
             target => "startdate"
       }
if [@timestamp] < [startdate] {
//drop loggings if they are logged before the startdate
           drop{}
       }
 

How to create ElasticSearch indices with different name patterns?

By default, Logstash creates indices with the name pattern logstash-%{+YYYY.MM.dd}, such as logstash-2013.10.20, you can change it. For example, I analyze log files from different customers, every customer has multiple nodes. I change the name pattern to be:
output {
  elasticsearch {
             host => localhost        
             cluster => AppleElastic1 
             index => "logstash-%{customer}-%{node}-%{+YYYY.MM.dd}"       
  }
}
Note the index name has to be lowercase, so the fields customer and node has to be lowercased. You can use mutate to do lowercasing:
mutate {                   
                    lowercase => [ "customer" ,”node”]
}
                   

How to overwrite ElasticSearch documents? 

By default, ElasticSearch uses a random string as the id for a document, so if you rerun Logstash to process the same log files, the original documents won’t be overwritten.
To overcome this, you can specify the id, typically, you should use the timestamp of the logging, for example:
 elasticsearch {
             host => localhost        
             cluster => AppleElastic1 
             index => "logstash-%{customer}-%{node}-%{+YYYY.MM.dd}"
             document_id=>"%{timestamp}"     
  }

How to change the default ElastiSearch mapping?  

Logstash ElasticSearch uses dynamic mapping which is defined in logstash-1.4.2\lib\logstash\outputs\elasticsearch\elasticsearch-template.json, this dynamic mapping often is not ideal, for example, although in “Why does Logstash ElasticSearch output create indices with wrong dates?”, we’ve managed to get logging timestamps that look like they are dates, they are actually stored as strings. Checking the mapping of the created indices, you will see:
 "@timestamp": {
        "format": "dateOptionalTime",
        "type": "date"
    },
    "timestamp": {
        "norms": {
            "enabled": false
        },
        "type": "string",
        "fields": {
            "raw": {
                "index": "not_analyzed",
                "ignore_above": 256,
                "type": "string"
            }
        }
    }
You will see the default built-in @timestamp is of date type, while the timestamp is of string type. 

Another thing I want to change is that for some fields, I do not want them to be analyzed, for example, event and subevent fields, I often need to do aggregation searches on them, if they are analyzed, the result will be wrong. Although I can use event.raw and subevent.raw fields (which are not analyzed, as the above example shows) as an alternative, it is not convenient, and it is a waste of resources to have .raw fields. 

To overcome, copy the default mapping file logstash-1.4.2\lib\logstash\outputs\elasticsearch\elasticsearch-template.json and make changes to it, in my case:
{
  "template" : "logstash-*",
  "settings" : {
    "index.refresh_interval" : "5s"
  },
  "mappings" : {
    "_default_" : {
       "_all" : {"enabled" : true},             
       "dynamic_templates" : [ {
         "string_fields" : {
           "match" : "*",
           "match_mapping_type" : "string",
           "mapping" : {
             "type" : "string", "index" : "analyzed", "omit_norms" : true,
               "fields" : {
                 "raw" : {"type": "string", "index" : "not_analyzed", "ignore_above" : 256}
               }
           }
         }
       }],         
       "properties" : {
         "@version": { "type": "string", "index": "not_analyzed" },
         "geoip"  : {
           "type" : "object",
             "dynamic": true,
             "path": "full",
             "properties" : {
               "location" : { "type" : "geo_point" }
             }
         },
              "timestamp" : {
                    "format" : "yyyy/MM/dd-HH:mm:ss.SSS z",
                    "type" : "date"
              },
             "event": {"type": "string", "index": "not_analyzed" },
              "subevent": {"type": "string", "index": "not_analyzed" }
       }
    }
  }
}
And use this customized mapping in ElasticSearch output:
  elasticsearch {
             host => localhost        
             cluster => AppleElastic1
             document_id=>"%{timestamp}"
             template=>"../data/conf/logstash-template.json"           
  }
Before you run Logstash, make sure that you’ve deleted the existing mapping.
#check the existing mapping:
curl -XGET http://localhost:9200/_template/logstash?pretty
#delete the existing mapping
curl -XDELETE http://localhost:9200/_template/logstash?pretty

No comments:

Post a Comment