Host variable replaced, but not configured to be

jdmac · September 23, 2016, 4:20pm

Hi All,

TLDR; - I suspect http { } is modifying the received host field value, without being explicitly configured to. I need to stop this happening.

I'm using the following configuration file.

input {
        http {
                codec => json
        }
}

filter {
        if [type] == "load" {
                mutate {
                        add_field => { "1m" => "%{values[0]}" }
                        add_field => { "5m" => "%{values[1]}" }
                        add_field => { "15m" => "%{values[2]}" }
                }
        }
        if [type] == "if_octets" {
                mutate {
                        add_field => { "rx" => "%{values[0]}" }
                        add_field => { "tx" => "%{values[1]}" }
                }
                ruby {
                        code => "event['rx'] = event['rx'] * 8"
                }
                ruby {
                        code => "event['tx'] = event['tx'] * 8"
                }
        }
        date {
                match => [ "time", "UNIX" ]
        }
}

output {
        elasticsearch {}
        stdout { codec => rubydebug }
}

Before I continue on explaining the issue, I will state the first thing I've checked before pointing a finger at Logstash is that I've used tcpdump to examine the raw JSON coming from collectd to logstash and can see the "host" field has been passed and contains a correct value.

However, once logstash has finished with the event, the host field has been changed to what appears to be the source IP of the http traffic (since I don't have resolutions setup yet).

I really want to be able to disable that behavior some how, because I need the host field to remain correctly set from the collectd sender.

I have already tried to see if a mutate filter to add a new field and copy the original %{host} would work but it too ends up being the source IP... Leading me to suspect some behavior of the http input module.

Although I can't see any explanation or toggle in the documentation - Are there any options I can toggle to disable this behavior?

Many thanks,

jdmac

jdmac · September 24, 2016, 11:08pm

Anyone have any thoughts on this or experience of evading the variable being replaced?

magnusbaeck · September 25, 2016, 7:44pm

Unfortunately the http input unconditionally overwrites the host field:

github.com

logstash-plugins/logstash-input-http/blob/v3.0.3/lib/logstash/inputs/http.rb#L141




# proc needs to be defined at this context
# to capture @codecs, @logger and lowercase_keys
p = Proc.new do |req|
  begin
    remote_host = req['puma.socket'].peeraddr[3]
    REJECTED_HEADERS.each {|k| req.delete(k) }
    req = lowercase_keys(req)
    body = req.delete("rack.input")
    @codecs.fetch(req["content_type"], @codec).decode(body.read) do |event|
      event.set("host", remote_host)
      event.set("headers", req)
      decorate(event)
      queue << event
    end
    ['200', @response_headers, ['ok']]
  rescue => e
    @logger.error("unable to process event #{req.inspect}. exception => #{e.inspect}")
    ['500', @response_headers, ['internal error']]
  end
end

This is arguable a bug and I've filed an issue for it:

jdmac · September 25, 2016, 7:56pm

Thanks, I will head over to +1 the bug. Thanks for confirming, I had
started to look at some of source to see what was doing it.

Topic		Replies	Views
Trying to replace 'host' field (created by tcp input) with a hash Logstash	1	1078	January 6, 2019
Trying to replace ‘host’ and port field Logstash	3	2285	January 8, 2020
Logstash Http Input overwriting field Logstash	3	913	December 22, 2017
Dynamic add_field in metrics Logstash	2	1333	September 2, 2017
Not to add 'host', 'path' field Logstash	5	9717	July 6, 2017

Host variable replaced, but not configured to be

Related topics