Feature #10674: Net::HTTP retries idempotent requests once after a timeout, but its not configurable

• Status: Closed
• Assignee: naruse (Yui NARUSE)
• Target version: -

[jlecour (Jérémy Lecour)]

Hi,

I've noticed that Net::HTTP retries idempotent requests once after a time-out. It seems to adhere to the RFC 2616 (http://tools.ietf.org/html/rfc2616), but it is not always the best thing to do.

When you don't know exactly how a remote web service works, it may be wise to retry after a time-out.
But when you know that a retry won't change anything, it should be possible to disable this automatic retry. Otherwise, the server will have to deal with 2 requests (instead of 1) and the client will effectively time-out twice later than expected (time for the first and second time-out).

Having a basic "#retry=" (like we have "#read_timeout=" or "#open_timeout" would be good first step.
The value could be a simple boolean, or even an integer for the maximum number of retries.

The "Retry" middleware in Faraday has a more comprehensive feature set with max retries, exponential back-off, … : https://github.com/lostisland/faraday/blob/master/lib/faraday/request/retry.rb

[shulmang (Garett Shulman)]

I can confirm this exact behavior on 2.1.7

[cben (Beni Cherniavsky-Paskin)]

Can confirm on 2.4.1.
The responsible code is here: https://github.com/ruby/ruby/blob/v2_4_1/lib/net/http.rb#L1468
I'm thinking of making a simple patch adding a boolean "#retry=". Application or higher-level gems can then implement any retry logic...

[naruse (Yui NARUSE)]

• Status changed from Open to Assigned
• Assignee changed from drbrain (Eric Hodel) to naruse (Yui NARUSE)

[naruse (Yui NARUSE)]

• Status changed from Assigned to Closed

Applied in changeset trunk|r60035.

---

Make retries for Net::HTTP configurable [Feature #10674]

by stereobooster
fix https://github.com/ruby/ruby/pull/1654