Dalli requires support for the connect_timeout option to TCPSocket#initialize

[nirvdrum]

We currently do not support the connect_timeout option to TCPSocket#initialize. This has recently come up as a problem with Dalli 3.2.7, which started using connect_timeout. This will impact those running Rails with memcache as its cache store.

TypeError: no implicit conversion of {:connect_timeout=>1} into Integer
--
  | /usr/local/ruby/lib/truffle/socket/truffle.rb:168:in `coerce_to_string'
  | /usr/local/ruby/lib/truffle/socket/tcp_socket.rb:66:in `initialize'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/socket.rb:93:in `open'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/protocol/connection_manager.rb:210:in `memcached_socket'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/protocol/connection_manager.rb:55:in `establish_connection'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/protocol/base.rb:207:in `connect'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/protocol/base.rb:196:in `ensure_connected!'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/protocol/base.rb:58:in `alive?'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/ring.rb:46:in `server_for_key'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/client.rb:425:in `perform'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/client.rb:208:in `set_cas'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/dalli-3.2.7/lib/dalli/client.rb:201:in `set'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache/mem_cache_store.rb:265:in `block (2 levels) in write_serialized_entry'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/connection_pool-2.4.1/lib/connection_pool.rb:110:in `block (2 levels) in with'
  | <internal:core> core/thread.rb:94:in `handle_interrupt'
  | <internal:core> core/thread.rb:94:in `handle_interrupt'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/connection_pool-2.4.1/lib/connection_pool.rb:109:in `block in with'
  | <internal:core> core/thread.rb:94:in `handle_interrupt'
  | <internal:core> core/thread.rb:94:in `handle_interrupt'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/connection_pool-2.4.1/lib/connection_pool.rb:106:in `with'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache/mem_cache_store.rb:265:in `block in write_serialized_entry'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache/mem_cache_store.rb:336:in `rescue_error_with'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache/mem_cache_store.rb:262:in `write_serialized_entry'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache/strategy/local_cache.rb:154:in `write_serialized_entry'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache/mem_cache_store.rb:252:in `write_entry'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache.rb:668:in `block in write'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache.rb:1030:in `block in _instrument'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/notifications.rb:208:in `instrument'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache.rb:1029:in `_instrument'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache.rb:1006:in `instrument'
  | /tmp/bundle/truffleruby/3.2.2.23.1.0/gems/activesupport-7.1.3/lib/active_support/cache.rb:666:in `write'

[eregon]

Thanks for the report.
The corresponding CRuby ticket is https://bugs.ruby-lang.org/issues/17187.
From a quick look at the PR they seem to use the fact that socket file descriptors are immediately marked as non-blocking in CRuby and then try one connect() and if not immediately successful wait using select/poll with the timeout.

In lib/truffle/socket/tcp_socket.rb the logic seems to currently use a blocking socket and blocking connect() call.
So we can either do the same as CRuby but that might involve changing quite a bit of Socket code, or use Timeout.timeout instead, but that occurs the cost of an extra thread the first time it is used and likely has more overhead.

Likely relevant snippet from man 2 connect:

       EINPROGRESS
              The socket is nonblocking and the connection cannot be completed immediately.  (UNIX domain  sockets  failed
              with  EAGAIN  instead.)   It  is possible to select(2) or poll(2) for completion by selecting the socket for
              writing.  After select(2) indicates writability, use getsockopt(2) to read  the  SO_ERROR  option  at  level
              SOL_SOCKET  to  determine  whether  connect()  completed  successfully  (SO_ERROR is zero) or unsuccessfully
              (SO_ERROR is one of the usual error codes listed here, explaining the reason for the failure).

[eregon]

If this is aimed for 24.0 I think best to use the Timeout.timeout approach as it seems less risky (changing all socket fds to be non-blocking is likely a big change with many side effects).