#iterator #wrapper #timeout #peek #buffer

timeout-iterator

TimeoutIterator is a wrapper over any iterator that adds peek_timeout and next_timeout functions. The canonical use-case is parsing multi-line free-form records (such as tailing a log fime) where it is desirable to consume the very last line, and peek whether the record continues on the next time, but not block indefinitely on the peek

12 releases (7 stable)

1.1.4 Jan 3, 2021
1.1.2 Dec 31, 2020
0.5.0 Apr 29, 2020
0.4.0 Apr 29, 2020
0.1.0 Sep 23, 2019

#57 in Asynchronous

Download history 4/week @ 2020-10-28 10/week @ 2020-11-04 17/week @ 2020-11-11 3/week @ 2020-11-18 7/week @ 2020-11-25 16/week @ 2020-12-02 14/week @ 2020-12-09 24/week @ 2020-12-16 55/week @ 2020-12-23 56/week @ 2020-12-30 15/week @ 2021-01-06 27/week @ 2021-01-13 25/week @ 2021-01-20 1/week @ 2021-01-27 55/week @ 2021-02-03 29/week @ 2021-02-10

68 downloads per month

Custom license

26KB
552 lines

Build Status

timeout-iterator

timeout_iterator::TimeoutIterator is a wrapper over any iterator that adds two additional functions:

  • peek_timeout()
  • next_timeout()

The canonical use-case is parsing multi-line free-form records (such as tailing a log fime) where it is desirable to consume the very last line, and peek whether the record continues on the next time, without blocking indefinitely on the peek().

This was built for parsing Kernel logs from /dev/kmsg for instance. A kernel log record may look like this:

6,361,518496,-;ahci 0000:00:05.0: AHCI 0001.0300 32 slots 6 ports 6 Gbps 0x1 impl SATA mode
 SUBSYSTEM=pci
 DEVICE=+pci:0000:00:05.0

The end of such a record is only known when the next line begins a new record. However, if this were the last record emitted, then it is possible that it never gets flushed/parsed because there is no next record to trigger it.

This is where an iterator with a timeout capability helps to break the deadlock.

Synchronous Iteration

When feature sync is used.

The iterator is obvious and intuitive:

use timeout_iterator::synchronous::TimeoutIterator;

let numbers: Vec<u32> = vec![1, 2, 3, 4, 5];
let mut ti = TimeoutIterator::with_iter(numbers.into_iter()).unwrap();

This will iterate over all the integers in the Vector.

However, if the underlying iterator is bursty, you can break out of iteration after a certain timeout duration. This has practical applications when tailing log files - you can wait for a timeout to publish or process groups of lines of log.

use timeout_iterator::synchronous::TimeoutIterator;
use timeout_iterator::error::TimeoutIteratorError;
use std::io::{BufReader, BufRead};
use std::fs::File;
use std::time::Duration;

let file = File::open("log.txt").unwrap();
let lines = BufReader::new(file).lines();
let mut ti = TimeoutIterator::with_iter(lines).unwrap();

loop {
   match ti.next_timeout(Duration::from_secs(1)) {
       Ok(Ok(next_line)) => {
           println!("{}", next_line);
       },
       Ok(Err(_)) => {
           // TimeoutIterator succeeded, underlying iterator provided an error
       },
       Err(TimeoutIteratorError::TimedOut) => {
           // timed out waiting for underlying iterator to provide something
       },
       Err(_) => {
           // other TimeoutIterator error
       }
   }
}

There is a similar peek_timeout function to peek the next entry (or timeout doing so), so you can see if anything is coming down the line without consuming it.

Asynchronous Iteration

When feature async is used.

The stream wrapper is intuitive:

    let numbers: Vec<u32> = vec![1, 2, 3, 4, 5];
    let numbers_stream = iter(numbers.into_iter());

    let mut ti = TimeoutStream::with_stream(numbers_stream).await?;

You can use it like any other Stream:

            assert_eq!(ti.next().await?, 1);
            assert_eq!(ti.next().await?, 2);
            assert_eq!(ti.next().await?, 3);
            assert_eq!(ti.next().await?, 4);
            assert_eq!(ti.next().await?, 5);

You can peek the next value without consuming it:

    // Can peek many times
    assert_eq!(ti.peek().await.unwrap(), 1);
    assert_eq!(ti.peek().await.unwrap(), 1);
    assert_eq!(ti.peek().await.unwrap(), 1);
    assert_eq!(ti.peek().await.unwrap(), 1);

    // And then consume with 'next`
    assert_eq!(ti.next().await.unwrap(), 3);

You can peek or consume with a timeout and catch the Error::TimedOut error:


    let numbers: Vec<u32> = vec![1, 2, 3, 4, 5];

    // Slow down the numbers to stream
    let throttled_numbers = iter(numbers.into_iter())
        .throttle(Duration::from_secs(1));

    let mut ti = TimeoutStream::with_stream(throttled_numbers).await.unwrap();

    // First number is always available
    assert_eq!(ti.next().await.unwrap(), 1);

    // 2nd number will timeout at half a second in
    assert_matches!(ti.next_timeout(Duration::from_millis(500)).await.unwrap_err(), Error::TimedOut);

    // Will consume it if called blocking
    assert_eq!(ti.next().await.unwrap(), 2);

    // Peek with timeout will... timeout
    assert_matches!(ti.peek_timeout(Duration::from_millis(500)).await.unwrap_err(), Error::TimedOut);

    // a blocking peek will eventually succeed
    // we dereference the peek because it's a reference (not move)
    assert_eq!(*ti.peek().await.unwrap(), 3);

    // The number will be consumed
    assert_eq!(ti.next().await.unwrap(), 3);

    // As proven by the next number
    assert_eq!(ti.next().await.unwrap(), 4);

Dependencies

~0–750KB
~13K SLoC