Wildcard and quoting handling are not correct in Windows

[tagomoris]

Backslashes are handled as quoting character in path of in_tail.
But backslashes are also path separator in Windows environment, and it is not determined automatically whether a backslash is separator or quoting character.

We should have a way to handle it correctly in Windows environment.
Options:

• add a configuration parameter not to extract globs and handle backslashes as path separators always

[daipom]

This problem also exists in v1.
We should need to be careful to use Slash / for the path separator for Windows.

I found the file buffer uses Dir.glob when resuming, so it can't find the existing chunk files if we use Backslash \ to the buffer path.

fluentd/lib/fluent/plugin/buf_file.rb

Lines 139 to 146 in 77d7229

	def resume
	stage = {}
	queue = []
	exist_broken_file = false
	patterns = [@path]
	patterns.unshift @additional_resume_path if @additional_resume_path
	Dir.glob(escaped_patterns(patterns)) do |path|

Dir.glob doesn't work with Backslash as the path separator for Windows, even if escaping it like \\.

Dir.glob("C:\\test\\fluentd\\buffer\\buffer.*.log")
=> []

Dir.glob("C:/test/fluentd/buffer/buffer.*.log")
=>  ["C:/test/fluentd/buffer/buffer.b5ff1d72f0a4b33c509f95a77bab5bf89.log", "C:/test/fluentd/buffer/buffer.b5ff1d74b9eb7a2e23696bf59405d3fce.log", "C:/test/fluentd/buffer/buffer.b5ff1d8af5a361ee8bc9a81ed39e877da.log"]

How to Reproduce

env

• Windows 10 Home
• Fluentd v1.16.1
• ruby 3.2.0 (2022-12-25 revision a528908271) [x64-mingw-ucrt]

config

<source>
  @type sample
  tag test
</source>

<match test.**>
  @type stdout
  <buffer>
    @type file
    path C:\test\fluentd\buffer
    flush_at_shutdown false
    flush_interval 10
  </buffer>
</match>

steps

• Run Fluentd: $ bundle exec fluentd -c /path/to/config
• Stop Fluentd so that some chunk files remain.
• Run Fluentd again.
• We can find the existing chunk files not being loaded and remaining forever.