Taobench install not working as expected on Ubuntu 22.04

[KarthikL1729]

Hello!

I have tried installing the Taobench benchmark a few times on a remote server running Ubuntu 22.04. Recently however, the install has been failing. It fails with the error:

error: ‘PyArray_Descr’ {aka ‘struct _PyArray_Descr’} has no member named ‘elsize’

during the folly build, and I think that is because of a NumPy dependency issue. I was able to resolve that by force installing Numpy version 1.26.4 (since the error pops up for numpy versions >= 2.0).

Once this was fixed, there were errors with OpenSSL, as the install script has this:

if [ "$LINUX_DIST_ID" = "ubuntu" ]; then
  cp build-deps/lib/libcrypto.so.1.1 /lib/x86_64-linux-gnu/
fi

but build-deps/lib/libcrypto.so.1.1 is never generated. I commented out the OpenSSL install and let it use the default OpenSSL install from Ubuntu, and it seems to have fixed everything.

I was not sure if these were bugs or if I was doing anything wrong, so wanted to clarify. To reproduce the issue:

• Fresh install of Ubuntu 22.04
• Install dependencies:

sudo apt update
sudo apt install -y python3-pip git
sudo pip3 install click pyyaml tabulate pandas

• Taobench install fails (./benchpress_cli.py install tao_bench_autoscale)

[excelle08]

Thanks for reporting this issue! I'm looking into this and trying to come up with a more sustainable solution than copying build-deps/lib/libcrypto.so.3 to the system library path.

Removing that three lines would indeed workaround the failure, but not sure if you'll see errors like /lib/aarch64-linux-gnu/libcrypto.so.3: version 'OPENSSL_3.3.0' not found if you use ldd benchmarks/tao_bench/tao_bench_server to check the linked libraries. This is because when building tao_bench_server the linker linked the binary with the correct libssl but the wrong libcrypto, and interestingly this only happened in Ubuntu, not CentOS.

If you see the 'OPENSSL_3.3.0' not found error, you would need to export LD_PRELOAD=benchmarks/tao_bench/build-deps/lib/libcrypto.so.3 before running TaoBench.