Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

in_udp: add test of message_length_limit #4117

Merged
merged 1 commit into from
Apr 11, 2023

Conversation

daipom
Copy link
Contributor

@daipom daipom commented Mar 29, 2023

Which issue(s) this PR fixes:
None.

What this PR does / why we need it:
Add test of message_length_limit for in_udp because there was no test for this option.

Docs Changes:
Not needed.

Release Note:
Not needed.

@daipom daipom requested review from ashie and removed request for ashie March 29, 2023 12:57
@daipom daipom marked this pull request as draft March 29, 2023 13:45
@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

The new test is failing on Windows.
Looks like the specification is different only on Windows.

I'm looking into it in detail.

Error: test: message_length_limit(UdpInputTest::hoge): Errno::EMSGSIZE: A message sent on a datagram socket was larger than the internal message buffer or some other network limit, or the buffer used to receive a datagram into was smaller than the datagram itself. - recvfrom(2)
C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/server.rb:558:in `recvfrom'
C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/server.rb:558:in `on_readable_with_sock'
C:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/cool.io-1.7.1/lib/cool.io/io.rb:186:in `on_readable'
C:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/cool.io-1.7.1/lib/cool.io/loop.rb:88:in `run_once'
C:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/cool.io-1.7.1/lib/cool.io/loop.rb:88:in `run'
C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/event_loop.rb:93:in `block in start'
C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/thread.rb:78:in `block in thread_create'

@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

At least on non-Windows, the data over message_length_limit are discarded.

https://pubs.opengroup.org/onlinepubs/009695399/functions/recvfrom.html

If a message is too long to fit in the supplied buffer, and MSG_PEEK is not set in the flags argument, the excess bytes shall be discarded.

And there is no error EMSGSIZE.

@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

On Windows, sending data over message_length_limit causes an unexpected error and the data is completely lost.

$ bundle exec fluentd -c .\fluent\fluent.conf -v
2023-03-29 23:14:07 +0900 [info]: fluent/log.rb:360:info: init supervisor logger path=nil rotate_age=nil rotate_size=nil
2023-03-29 23:14:07 +0900 [info]: fluent/log.rb:360:info: parsing config file is succeeded path=".\\fluent\\fluent.conf"
2023-03-29 23:14:07 +0900 [info]: fluent/log.rb:360:info: gem 'fluentd' version '1.16.0'
2023-03-29 23:14:07 +0900 [debug]: fluent/log.rb:339:debug: No fluent logger for internal event
2023-03-29 23:14:07 +0900 [info]: fluent/log.rb:360:info: using configuration file: <ROOT>
  <source>
    @type udp
    bind "192.168.10.8"
    tag "test"
    message_length_limit 32
    <parse>
      @type "none"
    </parse>
  </source>
  <match test.**>
    @type stdout
  </match>
</ROOT>
2023-03-29 23:14:07 +0900 [info]: fluent/log.rb:360:info: starting fluentd-1.16.0 pid=4576 ruby="3.2.0"
2023-03-29 23:14:07 +0900 [info]: fluent/log.rb:360:info: spawn command to main:  cmdline=["C:/Ruby32-x64/bin/ruby.exe", "-rC:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/bundler-2.4.8/lib/bundler/setup", "-Eutf-8", "C:/Ruby32-x64/lib/ruby/gems/3.2.0/bin/fluentd", "-c", ".\\fluent\\fluent.conf", "-v", "--under-supervisor"]
2023-03-29 23:14:10 +0900 [info]: #0 fluent/log.rb:360:info: init worker0 logger path=nil rotate_age=nil rotate_size=nil
2023-03-29 23:14:10 +0900 [info]: fluent/log.rb:360:info: adding match pattern="test.**" type="stdout"
2023-03-29 23:14:10 +0900 [info]: fluent/log.rb:360:info: adding source type="udp"
2023-03-29 23:14:10 +0900 [debug]: #0 fluent/log.rb:339:debug: No fluent logger for internal event
2023-03-29 23:14:10 +0900 [info]: #0 fluent/log.rb:360:info: starting fluentd worker pid=15020 ppid=4576 worker=0
2023-03-29 23:14:10 +0900 [info]: #0 fluent/log.rb:360:info: listening udp socket bind="192.168.10.8" port=5160
2023-03-29 23:14:10 +0900 [info]: #0 fluent/log.rb:360:info: fluentd worker is now running worker=0
2023-03-29 23:15:19.367809800 +0900 test: {"message":"hoge"}
2023-03-29 23:15:39 +0900 [error]: #0 fluent/log.rb:402:error: unexpected error in processing UDP data error_class=Errno::EMSGSIZE error="A message sent on a datagram socket was larger than the internal message buffer or some other network limit, or the buffer used to receive a datagram into was smaller than the datagram itself. - recvfrom(2)"
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/server.rb:558:in `recvfrom'
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/server.rb:558:in `on_readable_with_sock'
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/cool.io-1.7.1/lib/cool.io/io.rb:186:in `on_readable'
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/cool.io-1.7.1/lib/cool.io/loop.rb:88:in `run_once'
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Ruby32-x64/lib/ruby/gems/3.2.0/gems/cool.io-1.7.1/lib/cool.io/loop.rb:88:in `run'
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/event_loop.rb:93:in `block in start'
  2023-03-29 23:15:39 +0900 [error]: #0 cool.io/loop.rb:88:run_once: C:/Users/reang/Documents/work/fluentd/fluentd/lib/fluent/plugin_helper/thread.rb:78:in `block in thread_create'

@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

On Ubuntu focal, the data over message_length_limit` are discarded without any error.

$ bundle exec fluentd -c ../work/fluent.conf -v
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: init supervisor logger path=nil rotate_age=nil rotate_size=nil
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: parsing config file is succeeded path="../work/fluent.conf"
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: gem 'fluentd' version '1.16.0'
2023-03-29 23:26:20 +0900 [debug]: fluent/log.rb:339:debug: No fluent logger for internal event
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: using configuration file: <ROOT>
  <system>
  </system>
  <source>
    @type udp
    tag "test"
    bind "192.168.10.5"
    message_length_limit 32
    <parse>
      @type "none"
    </parse>
  </source>
  <match test.**>
    @type stdout
  </match>
  <source>
    @type monitor_agent
  </source>
</ROOT>
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: starting fluentd-1.16.0 pid=982411 ruby="3.2.0"
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: spawn command to main:  cmdline=["/home/daipom/.rbenv/versions/3.2.0/bin/ruby", "-r/home/daipom/.rbenv/versions/3.2.0/lib/ruby/site_ruby/3.2.0/bundler/setup", "-Eascii-8bit:ascii-8bit", "/home/daipom/.rbenv/versions/3.2.0/lib/ruby/gems/3.2.0/bin/fluentd", "-c", "../work/fluent.conf", "-v", "--under-supervisor"]
2023-03-29 23:26:20 +0900 [info]: #0 fluent/log.rb:360:info: init worker0 logger path=nil rotate_age=nil rotate_size=nil
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: adding match pattern="test.**" type="stdout"
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: adding source type="udp"
2023-03-29 23:26:20 +0900 [info]: fluent/log.rb:360:info: adding source type="monitor_agent"
2023-03-29 23:26:20 +0900 [debug]: #0 fluent/log.rb:339:debug: No fluent logger for internal event
2023-03-29 23:26:20 +0900 [info]: #0 fluent/log.rb:360:info: starting fluentd worker pid=982431 ppid=982411 worker=0
2023-03-29 23:26:20 +0900 [debug]: #0 fluent/log.rb:339:debug: listening monitoring http server on http://0.0.0.0:24220/api/plugins for worker0
2023-03-29 23:26:20 +0900 [debug]: #0 fluent/log.rb:339:debug: Start async HTTP server listening http://0.0.0.0:24220
2023-03-29 23:26:20 +0900 [debug]: #0 fluent/log.rb:339:debug:   0.0s: Async::IO::Socket
      | Binding to #<Addrinfo: 0.0.0.0:24220 TCP>
2023-03-29 23:26:20 +0900 [info]: #0 fluent/log.rb:360:info: listening udp socket bind="192.168.10.5" port=5160
2023-03-29 23:26:20 +0900 [info]: #0 fluent/log.rb:360:info: fluentd worker is now running worker=0
2023-03-29 23:26:47.859800904 +0900 test: {"message":"hoge"}
2023-03-29 23:26:52.054402922 +0900 test: {"message":"jfiaojeofpiajodkfjwoeeifojwiejok"}
$ netcat -u 192.168.10.5 5160
hoge
jfiaojeofpiajodkfjwoeeifojwiejokdjfoewioweifj

@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

If we rescue Errno::EMSGSIZE, then we can ignore incoming data over message_length_limit on Windows.
However, we can't discard only the part of data that exceeds message_length_limit in the same way as non-Windows.

def on_readable_with_sock
begin
data, addr = @sock.recvfrom(@max_bytes)
rescue Errno::EAGAIN, Errno::EWOULDBLOCK, Errno::EINTR, Errno::ECONNRESET, IOError, Errno::EBADF
return
end
@callback.call(data, UDPCallbackSocket.new(@sock, addr, close_socket: @close_socket))
rescue => e
@log.error "unexpected error in processing UDP data", error: e
@log.error_backtrace
raise if @under_plugin_development
end
end

@daipom
Copy link
Contributor Author

daipom commented Mar 29, 2023

We need to figure out what is the right specification.

@ashie
Copy link
Member

ashie commented Mar 30, 2023

Could you try with an undocumented parameter receive_buffer_size?

config_param :receive_buffer_size, :size, default: nil

https://learn.microsoft.com/en-us/windows-hardware/drivers/network/so-rcvbuf

I'm not sure whether it works as expected or not but it can tune the buffer size of a socket's receive buffer.

@daipom
Copy link
Contributor Author

daipom commented Mar 30, 2023

Could you try with an undocumented parameter receive_buffer_size?

config_param :receive_buffer_size, :size, default: nil

https://learn.microsoft.com/en-us/windows-hardware/drivers/network/so-rcvbuf

I'm not sure whether it works as expected or not but it can tune the buffer size of a socket's receive buffer.

Oh, thanks, I was not aware of that parameter.

Now I have confirmed that it doesn't work as expected. (On Windows)

  • Only receive_buffer_size 32
    • Incoming data is emitted as is without any limitation and any error.
  • receive_buffer_size 4096 and message_length_limit 32
    • The same error occurs.

@daipom
Copy link
Contributor Author

daipom commented Mar 31, 2023

I think what we need to do is the following 2 things:

  • Fix the unexpected error and the data over message_length_limit should be simply ignored on Windows.
  • Add documentation that the specifications differ depending on whether it is Windows or not.

https://docs.fluentd.org/plugin-helper-overview/api-plugin-helper-server#server_create-title-port-proto-nil-bind-0.0.0.0-shared-true-socket-nil-backlog-nil-tls_options-nil-m

This behavior on Windows looks like not being considered so far.

This can happen in in_syslog too.
As for in_syslog, as long as we can avoid unexpected errors, it looks good.
in_syslog should cut off or ignore the data over message_length_limit.

For in_udp, we can apply the same specification.
If we work hard to cut data off instead of just ignoring it for Windows, it would not be very happy.
Cutting data off would usually only cause a failure of parsing and the data would be eventually ignored.

@ashie
Copy link
Member

ashie commented Mar 31, 2023

Refs:

https://learn.microsoft.com/en-us/windows/win32/winsock/windows-sockets-error-codes-2

Return code/value Description
WSAEMSGSIZE10040 Message too long. A message sent on a datagram socket was larger than the internal message buffer or some other network limit, or the buffer used to receive a datagram was smaller than the datagram itself.

Ruby's implementation:

WinSock might return truncated data but discarded by Ruby.

@ashie
Copy link
Member

ashie commented Mar 31, 2023

  • Fix the unexpected error and the data over message_length_limit should be simply ignored on Windows.
  • Add documentation that the specifications differ depending on whether it is Windows or not.

It looks good, please go a head 👍

@ashie
Copy link
Member

ashie commented Mar 31, 2023

If we work hard to cut data off instead of just ignoring it for Windows, it would not be very happy.
Cutting data off would usually only cause a failure of parsing and the data would be eventually ignored.

User might want to know which kind of data is lost, there is a similar feature request: #3739
You don't need to care such thing in this PR though.

@daipom
Copy link
Contributor Author

daipom commented Mar 31, 2023

If we work hard to cut data off instead of just ignoring it for Windows, it would not be very happy.
Cutting data off would usually only cause a failure of parsing and the data would be eventually ignored.

User might want to know which kind of data is lost, there is a similar feature request: #3739 You don't need to care such thing in this PR though.

Oh, I see!
I had thought that it might be better to provide an INFO log about ignored data.

I will change the target of this PR to fixing max_bytes parameter of server helper.
I'll think about logging in the case of EMSGSIZE .

@daipom
Copy link
Contributor Author

daipom commented Mar 31, 2023

I had thought that it might be better to provide an INFO log about ignored data.

Oh, now I understand. It is impossible because of

WinSock might return truncated data but discarded by Ruby.

@daipom
Copy link
Contributor Author

daipom commented Apr 7, 2023

I made another PR to fix the helper.

When that PR is merged, this PR's tests will succeed on Windows too.

@daipom daipom force-pushed the add-test-in_udp-message_length_limit branch 2 times, most recently from c874fc1 to 4aa4342 Compare April 10, 2023 13:08
@daipom daipom force-pushed the add-test-in_udp-message_length_limit branch from 4aa4342 to ea38c8f Compare April 10, 2023 15:29
@daipom daipom marked this pull request as ready for review April 11, 2023 00:43
@daipom daipom requested a review from ashie April 11, 2023 00:43
@ashie ashie merged commit 863ebca into fluent:master Apr 11, 2023
@ashie
Copy link
Member

ashie commented Apr 11, 2023

Thanks!

@daipom daipom deleted the add-test-in_udp-message_length_limit branch April 11, 2023 05:24
@daipom
Copy link
Contributor Author

daipom commented Apr 11, 2023

Thanks for your review!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants