Windows Issue: The specified module could not be found

[cocoa-xu]

I get that same error on runtime when attempting to open a database connection

Originally posted by @DnOberon in #120 (comment)

[DnOberon]

Testing now!

[cocoa-xu]

I haven't shipped it yet but I think this should be everything (You can test it by placing the adbc_driver_manager.dll file in priv/bin directory)

[DnOberon]

Seems to work! Thank you so much for your help!

[DnOberon]

I'll keep my eye open for the release so I can switch, thanks again :)

[cocoa-xu]

Going to release v0.7.7 and retire v0.7.4-v0.7.6.

[cocoa-xu]

Okay, v0.7.7 is shipped🤞

[DnOberon]

@cocoa-xu SUCCESS

Feel free to close :)

[cocoa-xu]

That's great to hear! Closing this now

[DnOberon]

Seeing a new issue, intermittment on prod releases @cocoa-xu

What's odd is that this is on starting the application, where no code from this is actually running afaik. Sometimes it boots up my phoenix server ok, sometimes it gives me this error chain. Whats weird is that my system test script that actually uses ADBC works just fine...

{undef,[{'Elixir.Adbc.Nif.DLLLoader',add_dll_directory,[],[]},
        {'Elixir.Adbc.Nif',load_nif,0,[{file,"lib/adbc_nif.ex"},{line,13}]},
        {init,'-run_on_load_handler/2-fun-0-',1,
              [{file,"init.erl"},{line,1888}]}]}

=CRASH REPORT==== 27-Jan-2025::14:58:44.891000 ===
  crasher:
    initial call: kernel:init/1
    pid: <0.2299.0>
    registered_name: []
    exception exit: {on_load_function_failed,'Elixir.Adbc.Nif',
                        {undef,
                            [{'Elixir.Adbc.Nif.DLLLoader',add_dll_directory,
                                 [],[]},
                             {'Elixir.Adbc.Nif',load_nif,0,
                                 [{file,"lib/adbc_nif.ex"},{line,13}]},
                             {init,'-run_on_load_handler/2-fun-0-',1,
                                 [{file,"init.erl"},{line,1888}]}]}}
      in function  init:run_on_load_handlers/0 (init.erl, line 1858)
      in call from kernel:init/1 (kernel.erl, line 230)
    ancestors: [kernel_sup,<0.2292.0>]
    message_queue_len: 0
    messages: []
    links: [<0.2294.0>]
    dictionary: []
    trap_exit: false
    status: running
    heap_size: 376
    stack_size: 29
    reductions: 52
  neighbours:

=SUPERVISOR REPORT==== 27-Jan-2025::14:58:44.892000 ===
    supervisor: {local,kernel_sup}
    errorContext: start_error
    reason: {on_load_function_failed,'Elixir.Adbc.Nif',
                {undef,
                    [{'Elixir.Adbc.Nif.DLLLoader',add_dll_directory,[],[]},
                     {'Elixir.Adbc.Nif',load_nif,0,
                         [{file,"lib/adbc_nif.ex"},{line,13}]},
                     {init,'-run_on_load_handler/2-fun-0-',1,
                         [{file,"init.erl"},{line,1888}]}]}}
    offender: [{pid,undefined},
               {id,on_load},
               {mfargs,{proc_lib,start_link,[kernel,init,[on_load]]}},
               {restart_type,transient},
               {significant,false},
               {shutdown,2000},
               {child_type,worker}]

=CRASH REPORT==== 27-Jan-2025::14:58:44.892000 ===
  crasher:
    initial call: application_master:init/3
    pid: <0.2291.0>
    registered_name: []
    exception exit: {{shutdown,
                      {failed_to_start_child,on_load,
                       {on_load_function_failed,'Elixir.Adbc.Nif',
                        {undef,
                         [{'Elixir.Adbc.Nif.DLLLoader',add_dll_directory,[],
                           []},
                          {'Elixir.Adbc.Nif',load_nif,0,
                           [{file,"lib/adbc_nif.ex"},{line,13}]},
                          {init,'-run_on_load_handler/2-fun-0-',1,
                           [{file,"init.erl"},{line,1888}]}]}}}},
                     {kernel,start,[normal,[]]}}
      in function  application_master:init/3 (application_master.erl, line 143)
    ancestors: [application_controller,<0.10.0>]
    message_queue_len: 1
    messages: [{'EXIT',<0.2292.0>,normal}]
    links: [<0.2290.0>]
    dictionary: []
    trap_exit: true
    status: running
    heap_size: 376
    stack_size: 29
    reductions: 74
  neighbours:

=INFO REPORT==== 27-Jan-2025::14:58:44.892000 ===
    application: kernel
    exited: {{shutdown,
                 {failed_to_start_child,on_load,
                     {on_load_function_failed,'Elixir.Adbc.Nif',
                         {undef,
                             [{'Elixir.Adbc.Nif.DLLLoader',add_dll_directory,
                                  [],[]},
                              {'Elixir.Adbc.Nif',load_nif,0,
                                  [{file,"lib/adbc_nif.ex"},{line,13}]},
                              {init,'-run_on_load_handler/2-fun-0-',1,
                                  [{file,"init.erl"},{line,1888}]}]}}}},
             {kernel,start,[normal,[]]}}
    type: permanent

Kernel pid terminated (application_controller) ("{application_start_failure,kernel,{{shutdown,{failed_to_start_child,on_load,{on_load_function_failed,'Elixir.Adbc.Nif',{undef,[{'Elixir.Adbc.Nif.DLLLoader',add_dll_directory,[],[]},{'Elixir.Adbc.Nif',load_nif,0,[{file,\"lib/adbc_nif.ex\"},{line,13}]},{init,'-run_on_load_handler/2-fun-0-',1,[{file,\"init.erl\"},{line,1888}]}]}}}},{kernel,start,[normal,[]]}}}")

[josevalim]

It may be related to the order NIF modules are loaded. I will take a look, it may be that some release scripts are changing the order of modules.

[DnOberon]

Thanks, I'm digging into it as well. @josevalim this is on a prod build btw

[josevalim]

I can confirm that Erlang will load modules in random order in a release, so this will indeed only work from time to time. I am not sure of solutions at the moment.

[DnOberon]

I could just start an ADBC instance that I don’t ever need to interact with? Might solve it I’ll try

[josevalim]

@DnOberon I pushed a possible fix to main, untested, please give it a try.

@cocoa-xu do you think it would make sense to setup the DLL stuff within the NIF load callback and not even define a NIF function in the first place? Would that even be helpful?

[DnOberon]

Thanks, will try as soon as I’m able. Might be tomorrow EST though.

[cocoa-xu]

do you think it would make sense to setup the DLL stuff within the NIF load callback and not even define a NIF function in the first place? Would that even be helpful?

That's unfortunately impossible because adbc_nif.dll depends on adbc_driver_manager.dll so Windows will have to load adbc_driver_manager.dll first and that was the reason the dll loader helper was a standalone NIF...

Unless we switch to use function pointers and dynamically search for the adbc_driver_manager.dll file and find the corresponding symbols (but things can easily go wrong in this approach)

Another approach is just statically link adbc_driver_manager if that's possible -- theoretically this should be the best solution here but will need to test it out.

[josevalim]

We should be good on main. Please give it a try.

@cocoa-xu I meant something slightly different, here is a PR: #126. However, if we could statically load, maybe that's the best option here, as it would be by far the least amount of code?

[DnOberon]

The latest fix works perfectly @josevalim and @cocoa-xu . Thank you - you've saved a huge project demo.

[josevalim]

@cocoa-xu can you please do a new release or should I? I am pinging because both of us tend to be really fast, and therefore it is likely that we will both release at the same time. :D

[cocoa-xu]

@cocoa-xu can you please do a new release or should I? I am pinging because both of us tend to be really fast, and therefore it is likely that we will both release at the same time. :D

Haha that's true, I'll do the release in a minute!

[cocoa-xu]

Okay, tagged v0.7.8 and waiting for CI, should be ready and shipped in 5 minutes!

[cocoa-xu]

v0.7.8 is shipped, going to retire v0.7.7 as it sometimes won't work in a release, and closing this issue as it's finally resolved. :D