Options :use-connection-pool and :keep-alive make request non thread-safe about dexador HOT 7 CLOSED

With the current version (0.9.14), I believe the SBCL workaround should be:

(setf dexador.connection-cache::*threads-connection-pool*
        (make-hash-table :test 'equal :synchronized t))

not as given above as:

(setf cl-dbi::*threads-connection-pool*
        (make-hash-table :test 'equal :synchronized t))

from dexador.

(nit: your title is truncated)

from dexador.

@vindarel probably I just didn't finish it. Now it is fixed.

By the way, I've seen a similar error in Ultralisp.org's logs.

Sometimes it stop working because of that and I have to restart docker container :(
We need to invest some time and to fix it.

from dexador.

This issue is similar to fukamachi/cl-dbi#46 a lock should be added to guard hash from parallel modifications.

Or maybe some lock free structure like http://quickdocs.org/cl-skip-list/ could be used.

from dexador.

I've used cl-skip-list once, but in practice and for my use-case its performance was inferior to sbcl synchronized hash-table, and likely that will be also enough for most people. probably for sbcl only unfortunately, a very easy fix is to use (make-hashtable :synchronized t).

from dexador.

@fukamachi probably here https://github.com/fukamachi/cl-dbi/blob/7ba050dea8d137c1f85b7e704d4fc945104bf283/src/cache/thread.lisp#L19-L21 we need to use bt:with-lock-held because hash is modified and this is not a thread safe operation.

from dexador.

Thank you, @Linden6! Original post includes the code for cl-dbi, which has a similar problem.

from dexador.