Should retries be restricted to idempotent requests?

Detailed Description

The failsafe plugin issues retries based on the exception but completely ignores the request method, i.e. even POST requests are retried right now.

Context

The backup request already filters for safe request methods. Maybe it’s safer to do something similar for retries?

Possible Implementation

No concrete idea yet, but there is a chance of sharing some code between backup request and failsafe plugin, since all safe methods are inherently idempotent.

Your Environment

• Version used: 2.9.0-RC.3

[whiskeysierra]

@tkrop See #450 which outlines a proposal for 3.x tackling this in a more structured way.

[tkrop]

You can’t unless the server responds with 429. Still you can optimize riptide-failsafe to not fall into this trap. As I suggested:

1. Remove all read errors from the default retry behavior of transient errors.
2. Add a feature that explicitly allows clients to enable retry on read errors in addition to transient errors - only on request basis.