Crash {port_exit,memory_allocation_failed} in odbc with MSSQL

[badlop]

Describe the bug

It seems the change proposed in #7000 and merged in 53ef5df introduces some problem which produces a crash:

=ERROR REPORT==== 15-Jan-2025::12:06:47.047985 ===
** Generic server <0.88.0> terminating
** Last message in was {#Port<0.7>,{exit_status,2}}
** When Server state == {state,#Port<0.7>,
                               {<0.9.0>,#Ref<0.1085833572.4179361795.157557>},
                               <0.9.0>,undefined,on,true,true,on,connected,
                               undefined,0,
                               [#Port<0.5>,#Port<0.6>],
                               #Port<0.8>,#Port<0.9>}
** Reason for termination ==
** {port_exit,memory_allocation_failed}

=CRASH REPORT==== 15-Jan-2025::12:06:47.048161 ===
  crasher:
    initial call: odbc:init/1
    pid: <0.88.0>
    registered_name: []
    exception exit: {port_exit,memory_allocation_failed}
      in function  gen_server:handle_common_reply/8 (gen_server.erl, line 1226)
    ancestors: [odbc_sup,<0.86.0>]
    message_queue_len: 1
    messages: [{'EXIT',#Port<0.7>,normal}]
    links: [<0.87.0>]
    dictionary: []
    trap_exit: true
    status: running
    heap_size: 4185
    stack_size: 28
    reductions: 12252
  neighbours:

To Reproduce

Elements required to trigger the problem:

• Erlang/OTP 26.0-rc3 or higher (affects also 27.2)
• MSSQL (both versions 2019 and 2022)
• Table created with specific contraints
• Select content from that table (table may or may not be empty)

I tried to reduce as much as possible the reproduction steps, but maybe it can be triggered with more simple table definition, or other databases.

How to reproduce the problem using my example:

• Install podman, used to setup the mssql database
• Install the Erlang/OTP version to test
• Download erlang-odbc-reproduce.zip and extract
• Run test.sh

Test success with Erlang 25.3

$ ./test.sh
...
Version: "Erlang/OTP 25 [erts-13.2] [source] [64-bit] [smp:4:4] [ds:4:4:10] [async-threads:1] [jit:ns]\n"
...
Result: {selected,["lastname2"],[{"lastname2"}]}

Test failure with Erlang 27.2

$ ./test.sh
...
Version: "Erlang/OTP 27 [erts-15.2] [source] [64-bit] [smp:4:4] [ds:4:4:10] [async-threads:1] [jit:ns]\n"
...
Result: {error,connection_closed}

=ERROR REPORT==== 15-Jan-2025::12:06:47.047985 ===
...
** Reason for termination ==
** {port_exit,memory_allocation_failed}
...

Expected behavior

The msqql server really sends the response, as seen by sniffing the traffic with sudo tcpflow -i lo -Cg port 1433. And that response is similar when using an old and a new Erlang version, so why does odbc crash?

If there's any problem in the SQL query, or in the MSSQL response, erlang should log a clear explanation about the problem.

Affected versions

Erlang/OTP 26.0-rc3 and higher, including 27.2

By reverting that commit into 27.2, the problem disappears.

Additional context

Other reports related to this problem:

• Comment added on bottom of Allow larger column sizes than 8001 in case DB supports it #7000 (comment)
• [BUG] MS SQL database backend reason: {port_exit,memory_allocation_failed} with Erlang >25.3 processone/ejabberd#4178

[IngelaAndin]

I have heard other complaints about this commit, and there has been attempts of fixing it but nothing that we have actually been able to test and commit as they changed our test case in a way that will not work out. Maybe it is a reason to revert the commit?!
This application has very low prio for us, but if anyone is interested to help we are willing to accept PRs.