Just an opinion, but I tend to think of this as distributed monolithic development... the more log correlation is needed, the more likely a requirement has been chopped up and then duct-taped back together over HTTP.
If you have a heavily used service, it's likely to have many concurrent requests logging into the same log stream and you absolutely need a request correlation identifier to be able to thread the log statements for a single request back together.
That's essential even if you don't actually call any other services as part of your implementation.