CSV inputs with headers

[kakoni]

I guess currently in_tail/csv parser don't support csv files with headers?
Is this planned?

[repeatedly]

Is this planned?

No plan for now.
Currently, Fluentd's parsers don't have re-initialize configuration mechanizm.
So if we need to support such metadata handling feature, we should re-design parser APIs.

[kakoni]

Ok. Thanks

[repeatedly]

@kakoni Hard to set keys parameter in the configuration?

[kakoni]

@repeatedly No but then I need to filter out headers rows. I wrote my own csvparser where I do something like;

      def parse(text)
        row = CSV.parse_line(text, col_sep: @delimiter)
        if @keys.empty?
          @keys = row
        elsif (@keys - row).empty?
          return
        else
          yield values_map(row)
        end
      end

Obviously this assumes that you read_from_head + input files are "immutable" that is they are written only once, no appends..

[repeatedly]

I see.
We will consider it but we need more time to re-design Parser API because in_tail shares parser instance between target files.
So using your own parser is better for now.

[tagomoris]

In v0.14 parser API design, <parse> section can get arguments for many purposes. For example, it can be used for patterns of filename.

@type tail
path /my/dir/*.csv
<parse> # default pattern
  @type csv
</parse>
<parse myfile.with.header.*.csv>
  @type csv
  csv_with_header true
</parse>

Sharing parsers for all files is from design of in_tail plugin, not parsers.

[Ninir]

@kakoni could you share the whole file (parser) and how to implement it please?
@tagomoris is csv_with_header already implemented? can't find anything... :(

[tagomoris]

I showed just API capability, but it's not implemented yet.

[kakoni]

@Ninir Heres an example
https://gist.github.com/kakoni/b0ef238e630e65e860c83bfe55ffb53a

But obviously this would only work if you always read_from_head (which is exactly the case in my situation)

[Ninir]

@tagomoris got it!
@kakoni Thank you good sir, perfect :)

[John-Lin]

Hi, I'm using fluentd version 0.14.23 and I want to parse csv with header. I found this issue is opened for a long time. The csv2 plugin seems not working due to fluentd upgrade and I modify the csv2 plugin source to

require 'fluent/plugin/parser'

require 'csv'

module Fluent
  module Plugin
    class CSV2Parser < Parser
      Plugin.register_parser('csv2', self)
      
      config_param :keys, :array, value_type: :string
      config_param :delimiter, :string, default: ','

      def parse(text, &block)
        values = CSV.parse_line(text, col_sep: @delimiter)
        if @keys.empty?
          @keys = values
        elsif (@keys - values).empty?
          return
        else
          r = Hash[@keys.zip(values)]
          time, record = convert_values(parse_time(r), r)
          yield time, record
        end
      end
    end
  end
end

it works! The example configuration is like below:

<source>
  @type tail
  path /my/path/to/csv
  tag hello
  format csv2
  keys 
</source>

<match **>
  @type stdout
</match>

note: keys should keep in configuration and the coming parameters should leave empty.

[amitdhawan]

Is this supported now?