rescue IO::TimeoutError raised by Ruby since 3.2.0 on blocking reads/writes #968
Conversation
|
See #967 |
|
Another option to fix this would be to use IOError in the rescue clauses, as IO::TimeoutError is a subclass of IOError and that class is defined for all Ruby versions. |
|
@skaes This looks ok to me, but it needs a rebase and a spec run. Can you please rebase the branch and we'll get CI running? Thanks. |
|
@petergoldstein the pull request was created from b69b24e which is the last commit on main. I don't know how to rebase it even more. But really happy this gets addressed. |
|
@skaes You were correct - there was no rebase required when I made my last comment. But I just made some commits (primarily to address test failures). Would you be able to rebase now and we can get specs running? Sorry about the confusion. |
…writes One scenario in which this happens is when a firewall silently drops connections between clients and memcached instances. With this patch the client reopens the connection and the request succeeds. Without it, IO:TimeoutError is raised for the application to handle which deviates from Dalli behavior under Ruby 3.1 and earlier.
skaes
force-pushed
the
rescue-io-timeout-error
branch
from
faec666 to
4acae51
Compare
|
@petergoldstein rebased. |
|
@skaes Thanks for rebasing. Can you please address the lints identified by Rubocop and get Rubocop passing? Then we can look at merging. Thanks. |
|
I fixed what I could. I left the task of commenting what module Protocol does to you, as I did not create that module. |
One scenario in which this happens is when a firewall silently drops connections between clients and memcached instances.
With this patch the client reopens the connection and the request succeeds. Without it, IO:TimeoutError is raised for the application to handle which deviates from Dalli behavior under Ruby 3.1 and earlier.
There re several ways to patch this up. I opted for introducing an array of timeout classes for the rescue clauses.