| 1 |
|
defmodule WalletIntegrations.RetryPolicy do |
| 2 |
|
@moduledoc """ |
| 3 |
|
Timeout, retry, and backoff configuration for outbound provider operations (ADR 0008). |
| 4 |
|
|
| 5 |
|
Each operation class has distinct timeout bounds appropriate to its latency profile. |
| 6 |
|
Retry is permitted only for transient/retryable failures. |
| 7 |
|
""" |
| 8 |
|
|
| 9 |
|
# Operation classes and their timeout policy (milliseconds) |
| 10 |
|
@policies %{ |
| 11 |
|
initiate_payment: %{ |
| 12 |
|
connect_timeout: 5_000, |
| 13 |
|
recv_timeout: 15_000, |
| 14 |
|
deadline: 20_000, |
| 15 |
|
max_attempts: 3, |
| 16 |
|
base_delay_ms: 1_000, |
| 17 |
|
max_delay_ms: 30_000 |
| 18 |
|
}, |
| 19 |
|
get_payment_status: %{ |
| 20 |
|
connect_timeout: 3_000, |
| 21 |
|
recv_timeout: 8_000, |
| 22 |
|
deadline: 12_000, |
| 23 |
|
max_attempts: 5, |
| 24 |
|
base_delay_ms: 500, |
| 25 |
|
max_delay_ms: 15_000 |
| 26 |
|
}, |
| 27 |
|
cancel_payment: %{ |
| 28 |
|
connect_timeout: 5_000, |
| 29 |
|
recv_timeout: 10_000, |
| 30 |
|
deadline: 15_000, |
| 31 |
|
max_attempts: 2, |
| 32 |
|
base_delay_ms: 2_000, |
| 33 |
|
max_delay_ms: 10_000 |
| 34 |
|
}, |
| 35 |
|
refund_payment: %{ |
| 36 |
|
connect_timeout: 5_000, |
| 37 |
|
recv_timeout: 15_000, |
| 38 |
|
deadline: 20_000, |
| 39 |
|
max_attempts: 2, |
| 40 |
|
base_delay_ms: 2_000, |
| 41 |
|
max_delay_ms: 20_000 |
| 42 |
|
} |
| 43 |
|
} |
| 44 |
|
|
| 45 |
|
# HTTP status codes that indicate a transient failure safe to retry |
| 46 |
|
@retryable_status_codes [429, 500, 502, 503, 504] |
| 47 |
|
|
| 48 |
|
@doc "Return the timeout/retry policy map for a given operation atom." |
| 49 |
|
@spec for_operation(operation :: atom()) :: map() |
| 50 |
|
def for_operation(operation) do |
| 51 |
5 |
Map.get(@policies, operation, %{ |
| 52 |
|
connect_timeout: 5_000, |
| 53 |
|
recv_timeout: 15_000, |
| 54 |
|
deadline: 20_000, |
| 55 |
|
max_attempts: 3, |
| 56 |
|
base_delay_ms: 1_000, |
| 57 |
|
max_delay_ms: 30_000 |
| 58 |
|
}) |
| 59 |
|
end |
| 60 |
|
|
| 61 |
|
@doc """ |
| 62 |
|
Calculate exponential backoff with full jitter for a given attempt number. |
| 63 |
|
Returns delay in milliseconds. |
| 64 |
|
Formula: jitter(0, min(cap, base * 2^attempt)) |
| 65 |
|
""" |
| 66 |
|
@spec calculate_backoff(attempt :: non_neg_integer(), base_ms :: pos_integer(), cap_ms :: pos_integer()) :: |
| 67 |
|
pos_integer() |
| 68 |
:-( |
def calculate_backoff(attempt, base_ms \\ 1_000, cap_ms \\ 30_000) do |
| 69 |
35 |
ceiling = min(cap_ms, trunc(base_ms * :math.pow(2, attempt))) |
| 70 |
|
# Full jitter: random in [0, ceiling] + 1 to ensure positive |
| 71 |
35 |
:rand.uniform(max(ceiling, 1)) |
| 72 |
|
end |
| 73 |
|
|
| 74 |
|
@doc "True when the HTTP status code indicates a transient, retryable failure." |
| 75 |
|
@spec retryable_status_code?(code :: non_neg_integer()) :: boolean() |
| 76 |
10 |
def retryable_status_code?(code), do: code in @retryable_status_codes |
| 77 |
|
|
| 78 |
|
@doc "True when the error reason indicates a transient, retryable condition." |
| 79 |
|
@spec retryable_error?(reason :: term()) :: boolean() |
| 80 |
1 |
def retryable_error?(:timeout), do: true |
| 81 |
1 |
def retryable_error?(:connect_timeout), do: true |
| 82 |
1 |
def retryable_error?(:econnrefused), do: true |
| 83 |
1 |
def retryable_error?(:closed), do: true |
| 84 |
1 |
def retryable_error?({:exit, _}), do: true |
| 85 |
3 |
def retryable_error?(_), do: false |
| 86 |
|
end |