Phoenix Channel sometimes stop working because :handle_info function is no longer exported

Hi there

I’ve run into a really strange problem in one of my apps recently, which I sadly cannot reproduce reliably but do see from time to time in :dev and :prod environments.

So I’ve got an app where users can create “cards” with some content on them, and we’ve got a list of all cards the user owns where he can search for his cards. The searches can be quite complicated, so what we’ve implemented is that the search request is send to the server which performs the search and then sends back the result in form of rendered html as a string. The result itself is sent using a channel called “search:unique_identifier_for_each_search_instance”, to which the client subscribes on loading of the app. The html is then received on the client and the current table with the results is replaced with the received new result.

This all is a bit more complicated then necessary, but a lot of this functionality has grown over time, so for now it is what it is :slight_smile:

Now the socket that the channel belongs to hosts multiple channels, e.g. another one is “alert:user_id” in order to send the user information if something goes wrong, of ir new cards are received from other users, just in case this might be important.

So finally the problem: Usually all this works fine, but from time to time what can happen is, that if the user navigates back in the browser (the frontend is an SPA, so the socket is not remounted), suddenly the search results are no longer sent to the client. Instead I get an error like this:

[warning] Phoenix.Channel.Server #PID<0.1296.0> received unexpected message in handle_info/2: 
{:new_result,   lots_of_html_here_that_is_truncated ......

So looking into the code of Phoenix.Channel.Server I found the place that triggers the message, and this only happens if function_exported?(MyApp.Channel.Search, :handle_info, 2) returns false. However the MyApp.Channel.Search module does actually export that function.

Next luck would have it that I triggered this error shortly afterwards while running the server using iex -S phx.server, and quickly checking the output of function_exported?(MyApp.Channel.Search, :handle_info, 2) in the console manually, the call does indeed return false. However after some digging around and trying it again, it now returns true, and the search also starts working as expected again. Also if I try function_exported?(MyApp.Channel.Search, :handle_info, 2) directly after starting the server, it also always returns true.

Does anyone have any idea how this can happen?

Thanks for your time already :smile:

Are you running a release? With a release, the BEAM is in the embedded mode and barring from hot code reloading, the function_exported? shall always return the same value.

Iex is interactive mode, so the code may not be loaded, if no one has called any function in that module.

1 Like

Hmm, I had the same thought yesterday. After checking the log files on the server I could actually not find the same error message, but I’m not sure if some sort of similiar problem still exists because in Phoenix.Channel.Server the logs are written using the Erlang :error_logger instead of the Elixir Logger. I’m just not sure if those logs are captured as well, but I’d guess so.

I’ll continue investigating on :prod, though I still don’t understand why in the :dev environment sometimes the function ceases to be exported…

Ok, I did some more investigating and it seems that @derek-zhou is correct in his answer above, thanks again. In :prod the error is not being logged, so I guess on :prod the issue is triggered by something else…