Slow Connections; Not valid win32 Application; OpenSSL

[metaskills]

We have had several issues with OpenSSL statically linked in the Windows gem. Issues include #250 and #290. Symptoms of using OpenSSL v1.0.x include:

• Slow initial connections.
• tiny_tds.so not a valid win32 application

Potential Workarounds

• Adding TinyTDS just before Rails starts up, for example above Bundler's setup in boot.rb.
• Setting chcp 866 for tiny_tds.so

Proposed Next Steps

One of the following two have been proposed.

• Hacks for old OpenSSL. Like using libeay32.dll.
• Use a modern OpenSSL v1.1.0

Test Appveyor Build

It would be really nice if we changed our OpenSSL version in extconsts.rb to use 1.1.0.x. I found when doing this that there are some Perl errors on Appveyor because the version is too low. We have a path config set to SET PATH=C:\MinGW\msys\1.0\bin;%PATH% and that seems to put a old version of Perl in. I did try adding SET PATH=C:\Perl\bin after the MinGW, but it does not seem to add totally fix the issue.

Update Docker Usage

I will be doing some CircleCI testing to ensure 1.1.x works for posix systems. However, there is a fair amount of work to make the native rake gem:windows tasks work for us to distribute native Windows gems.

[YvesR]

Hello @metaskills,
did your tests against Azure pass lately?
I tried 1.0.5 and AR 4.2 and still get the same problem. First connect fails, seconds works then. Also to mention that the 1.0.2g open-ssl still has this CPU bug that loading rails console takes 5 times compared to the fixed libeay32.dll.

bye Yves

[nmscholl]

@metaskills
I am currently experiencing the TinyTDS Not valid win32 Application error when executing a chef run. In some contexts, I have been able to resolve the problem by requiring TinyTDS in a particular order with the other requirements that use the conflicting components, however I have encountered a situation where this is not sufficient, I still get the error. I attempted to use chcp to resolve the issue as you suggested, however I am unsure what is meant by setting it up "for" a file. Is that different than running it in the same powershell window where tinyTDS will be used? In any case, chcp 866 was used in said powershell window, chef was executed, and the error still occurred. Any information you can provide would be appreciated.

[coderjoe]

So I've been poking at this on and off for around a day now, and I can't seem to get around the version constraints of the DevKit environment which prevent OpenSSL 1.1.x from building. My next step would be to move the responsibility of building OpenSSL outside of devkit (so not a part of the rake script), using the build environments available on appveyor, and finally consuming the built library within the tiny_tds build.

What do you guys think, is there any reason for me not to pursue this option? I'm pretty new to these concerns in windows and DevKit so I welcome any suggestions.

[coderjoe]

As an aside are there any reasons not to use one of the pre-built openssl libraries for windows mentioned on the openssl wiki? https://wiki.openssl.org/index.php/Binaries

[metaskills]

@YvesR Yes, I've run the Azure tests and they are fine. But I am mostly focusing on SQL Server on Linux and hence the latest work in master for it and CircleCI.

@nmscholl That is the symptom and reason for this issue. The fix that I am proposing is to use OpenSSL v1.1.x but that is not so easy and it requires work that I have not prioritized. @YvesR is suggesting to drop in a different DLL. Both require some work that I have not yet had time to prioritize.

@coderjoe Neat idea!!! Can we?

[coderjoe]

@metaskills honestly, not sure! But I'll keep poking at it. No promises! ;)

[YvesR]

@metaskills I have a test env where I do db access directly on start with a initialize script. This fail to start. Starting a console without first db access fails, next and following ones success.

About open SSL 1.1.x: Is it compatible? I read that the openSSL team broke up some compatibilities. So we can just use latest version what @coderjoe suggests?

[coderjoe]

@YvesR well my suggestions only really make sense if we're going to OpenSSL 1.1.x... if we keep following the LTS branch (1.0.2x) then it doesn't make sense to change the build system because it shouldn't be necessary yet.

The LTS support branch (1.0.2x) is good until December 31st 2019 so there's no immediate need to switch unless it's the only thing that fixes the binary incompatibility problem on windows. It would be much easier if switching to 1.0.2j fixed it, but I can't get the appveyor tests to pass there. (I notice you're having a similar problem @metaskills in your recent PR)

Honestly 1.1.x or 1.0.2x doesn't matter to me, I'm just desperate to solve this invalid windows binary problem since it's a critical blocker in my work.

[metaskills]

It would be much easier if switching to 1.0.2j fixed it
I notice you're having a similar problem @metaskills in your recent PR

Yup! See the issue description about C:\Perl\bin and such. I have yet to address this. Is there anything I can do to get help with that?

[coderjoe]

I meant more in PR #320 , before I started on the OpenSSL 1.1.x path I tried 1.0.2j but ran into the same build problem you are, which seems independent of the Perl version since 1.0.2x doesn't require updated Perl.

As for help, nothing I can think of now, I'm still learning the toolchain and experimenting with permutations so I don't know enough yet to help me ask the right questions. Do you know of any good guides/resources on setting up test VMs to test/verify tiny_tds on windows? Right now I'm working off of a Windows 10 Evaluation VM I set up to help me out.