ferogram-mtsender

MTProto sender pool and retry policy for ferogram.

Manages DC connections and drives the retry loop for RPC calls. ferogram sits on top of this; most users never touch it directly.

ferogram re-exports the retry types. Existing code needs no changes.

Installation

[dependencies]
ferogram-mtsender = "0.3.8"

What it does

DcPool: one DcConnection per DC, created on demand
DcConnection: owns the sender loop for a single DC
Retry policy trait with built-in AutoSleep, NoRetries, and CircuitBreaker
FLOOD_WAIT and SLOWMODE_WAIT auto-sleep with jitter
Exponential backoff for transient I/O errors

Retry policies

AutoSleep

Sleeps on FLOOD_WAIT and retries once on I/O errors. This is the default policy used by ferogram.

use ferogram_mtsender::AutoSleep;
use std::time::Duration;

let policy = AutoSleep {
    threshold: Duration::from_secs(60), // sleep through flood waits up to 60s
    io_errors_as_flood_of: Some(Duration::from_secs(1)),
};

NoRetries

Propagates every error immediately without sleeping.

use ferogram_mtsender::NoRetries;

let policy = NoRetries;

CircuitBreaker

Trips after a set number of consecutive failures and stays open for a cooldown window.

use ferogram_mtsender::CircuitBreaker;
use std::time::Duration;

// Trip after 5 consecutive errors, stay open for 30s
let policy = CircuitBreaker::new(5, Duration::from_secs(30));

Custom policy

use ferogram_mtsender::{RetryPolicy, RetryContext};
use std::ops::ControlFlow;
use std::time::Duration;

struct MyPolicy;

impl RetryPolicy for MyPolicy {
    fn should_retry(&self, ctx: &RetryContext) -> ControlFlow<(), Duration> {
        if ctx.fail_count.get() < 3 {
            ControlFlow::Continue(Duration::from_secs(1))
        } else {
            ControlFlow::Break(())
        }
    }
}

Stack position

ferogram
└ ferogram-mtsender  <-- here
  └ ferogram-connect

License

MIT or Apache-2.0, at your option. See LICENSE-MIT and LICENSE-APACHE.

Ankit Chaubey - github.com/ankit-chaubey

ferogram-mtsender 0.3.8