Increase start_server thread join timeout

[k0kubun]

The 30-second base timeout can be insufficient on slow CI runners, causing spurious failures in tests like test_server_session_cache. Double the base timeout to 60 seconds to reduce flakiness.

Example failure: https://github.com/ruby/ruby/actions/runs/23631288760/job/68830977060?pr=16576

[rhenium]

Do you see this happening often? This particular test takes 0.04 seconds on the GitHub-provided runner for ruby/openssl, and I wouldn't expect it to become 700x slower even with high parallelism normally.

https://github.com/ruby/openssl/actions/runs/23633478701/job/68837745603?pr=1020

 test_server_session_cache:			.: (0.039300)

[rhenium]

    1) Failure:
  OpenSSL::TestSSLSession#test_server_session_cache [/Users/runner/work/ruby/ruby/src/test/openssl/utils.rb:277]:
  exceptions on 1 threads:
  #<Thread:0x0000000128403480 /Users/runner/work/ruby/ruby/src/test/openssl/utils.rb:251 dead>:
  /Users/runner/work/ruby/ruby/build/.ext/common/openssl/buffering.rb:76:in 'OpenSSL::SSL::SSLSocket#sysread': [start_server] thread did not exit in 30 secs (RuntimeError)
  	from /Users/runner/work/ruby/ruby/build/.ext/common/openssl/buffering.rb:76:in 'OpenSSL::Buffering#fill_rbuff'
  	from /Users/runner/work/ruby/ruby/build/.ext/common/openssl/buffering.rb:238:in 'OpenSSL::Buffering#gets'
  	from /Users/runner/work/ruby/ruby/src/test/openssl/test_ssl_session.rb:207:in 'block (3 levels) in OpenSSL::TestSSLSession#test_server_session_cache'
  	from /Users/runner/work/ruby/ruby/src/test/openssl/test_ssl_session.rb:436:in 'OpenSSL::TestSSLSession#server_connect_with_session'
  	from /Users/runner/work/ruby/ruby/src/test/openssl/test_ssl_session.rb:206:in 'block (2 levels) in OpenSSL::TestSSLSession#test_server_session_cache'
  	from /Users/runner/work/ruby/ruby/src/test/openssl/test_ssl_session.rb:202:in 'Integer#times'
  	from /Users/runner/work/ruby/ruby/src/test/openssl/test_ssl_session.rb:202:in 'block in OpenSSL::TestSSLSession#test_server_session_cache'
  	from /Users/runner/work/ruby/ruby/src/test/openssl/utils.rb:255:in 'block (2 levels) in OpenSSL::SSLTestCase#start_server'

It seems either #sysread decided to wait for the underlying socket when it shouldn't have, or rb_io_wait() in #sysread didn't wake up, but I have no idea how it happened. I doubt doubling timeout would help in either case.

[k0kubun]

It's a RUBY_DEBUG=1 build (slower interpreter) + --zjit-call-threshold=1 (compile everything with ZJIT) build, so it's not just about parallelism.

However, we've just fixed ZJIT-specific flakiness in an openssl test ruby/ruby#16576 ruby/ruby#16585, and I'm wondering if this might have been just a case that's caused by the same problem but with a different symptom. So, I actually want to let it sit without this to see if it's true. Let me close this until I see this again.

Thank you for your comment.