Class: Ask::Agent::Middleware::RetryOnFailure

Inherits:
Base
  • Object
show all
Defined in:
lib/ask/agent/middleware/retry_on_failure.rb

Overview

Retries LLM provider calls on transient failures.

Retries on:

  • RateLimitError with exponential backoff + jitter
  • ServerError (5xx) with exponential backoff

Does NOT retry on:

  • Unauthorized — credentials are wrong, retrying won't help
  • InvalidCredential — same
  • ModelNotFound — model doesn't exist
  • ConfigurationError — user config issue

Examples:

pipeline.use :retry_on_failure, max_retries: 5

Constant Summary collapse

DEFAULT_MAX_RETRIES = 
3

Instance Method Summary collapse

Constructor Details

#initialize(max_retries: DEFAULT_MAX_RETRIES) ⇒ RetryOnFailure

Returns a new instance of RetryOnFailure.



23
24
25
 
# File 'lib/ask/agent/middleware/retry_on_failure.rb', line 23

def initialize(max_retries: DEFAULT_MAX_RETRIES)
  @max_retries = max_retries
end

Instance Method Details

#around_request(provider, request) ⇒ Object 



27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
 
# File 'lib/ask/agent/middleware/retry_on_failure.rb', line 27

def around_request(provider, request)
  last_error = nil

  @max_retries.times do |attempt|
    begin
      return yield
    rescue Ask::RateLimitError => e
      raise if attempt >= @max_retries - 1

      delay = e.retry_after || compute_backoff(attempt)
      sleep(delay)
      last_error = e
    rescue Ask::ServerError, Ask::ServiceUnavailable => e
      raise if attempt >= @max_retries - 1

      sleep(compute_backoff(attempt))
      last_error = e
    rescue Ask::Unauthorized, Ask::InvalidCredential,
           Ask::ModelNotFound, Ask::ConfigurationError
      raise
    end
  end

  raise last_error if last_error
end