abort on backend failures at startup and retry while running
LGTM, but what happens when backends get instantiated on other threads/for other causes?
From the top of my head API should be fine (will just cause internal server error to be returned), but no idea about other callers.
this indeed works on several scenarios for the pgsql backend
add missing throw in pipe backend
retry once after a backend failure
mysql autoreconnect is a really bad idea
prepared statementes are released at reconnect
increase the default value for the maximum number of TCP connections …
Is merged onto master, but GH fails for some reason