but some queries are preload queries and it is triggered in different process. (because Ecto’s optimisation mechanism)

I suppose you refer to ecto checking out a new connection for additional queries, even when they are causally related to some other queries that your app issues (preloads are dependent on the other resources that these preloads are for).

Sorry for being off-topic, but is it really an optimization? Wouldn’t it be more straightforward to run preloads on the same connection?

As for your question, maybe you can put the phoenix’s request_id into the process dictionary for the cowboy’s process handing the http request, and pass it to all other processes (like ecto) as well?

Something like

# in your controller action
request_id = conn.private.phoenix.request_id # or wherever it is stored
:erlang.put(:request_id, request_id)

# then to let ecto know about the request_id as well
Repo.transaction(fn ->
  # I suppose it's a different process, so we need to put the request id here as well
  :erlang.put(:request_id, request_id)
  # run your queries
end)

You can probably replace :erlang.put/2 with Logger.metadata, but I’m not sure.

# in your controller action
request_id = conn.private.phoenix.request_id # or wherever it is stored
Logger.metadata(request_id: request_id)

# then to let ecto know about the request_id as well
Repo.transaction(fn ->
  # I suppose it's a different process, so we need to put the request id here as well
  Logger.metadata(request_id: request_id)
  # run your queries
  # ...
  # and then 
  Logger.info("asfd") # should use request_id from before
end)