mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benjamin Mahler <bmah...@apache.org>
Subject Re: Review Request 68304: Fixed a backoff overflow bug in agent authentication retry logic.
Date Wed, 15 Aug 2018 21:29:36 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68304/#review207356
-----------------------------------------------------------


Fix it, then Ship it!





src/slave/constants.hpp
Lines 68-69 (original), 68-74 (patched)
<https://reviews.apache.org/r/68304/#comment290723>

    How about:
    
    ```
    // The minimum timeout used when authenticating against the master.
    //
    // TODO(mzhu): Make this configurable.
    constexpr Duration AUTHENTICATION_TIMEOUT_MIN = Seconds(5);
    
    // The maximum timeout used when authenticating against the master.
    //
    // TODO(mzhu): Make this configurable.
    constexpr Duration AUTHENTICATION_TIMEOUT_MAX = Minutes(1);
    ```



src/slave/slave.cpp
Lines 1288 (patched)
<https://reviews.apache.org/r/68304/#comment290722>

    "authentication" typo and newline:
    
    ```
        // Wait for a random amount of time before authentication or
        // registration.
        //
        // TODO(mzhu): Specialize this for authetication.
    ```



src/slave/slave.cpp
Lines 1401-1408 (original), 1421-1428 (patched)
<https://reviews.apache.org/r/68304/#comment290724>

    It seems a little easier to understand the overall approach if you show the math / instead
of directly using variable names?
    
    ```
    // Grow the timeout range using exponential backoff:
    //
    //   [min, min + factor * 2^0]
    //   [min, min + factor * 2^1]
    //   ...
    //   [min, min + factor * 2^N]
    //   ...
    //   [min, max] // Stop at max.   
    ```
    
    Once I understand this, I can easily map it to the variables


- Benjamin Mahler


On Aug. 15, 2018, 6:47 p.m., Meng Zhu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68304/
> -----------------------------------------------------------
> 
> (Updated Aug. 15, 2018, 6:47 p.m.)
> 
> 
> Review request for mesos, Benjamin Mahler and Gastón Kleiman.
> 
> 
> Bugs: MESOS-9147
>     https://issues.apache.org/jira/browse/MESOS-9147
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This patch fixed the backoff time calculation
> overflow bug described in MESOS-9147.
> 
> The old approach times out an authentication request after
> 5s and then retries after some backoff time. This is not
> optimal because, if the agent is going to backoff some time
> before retry, we might as well wait that long for the
> previous authentication request (instead of timeout early).
> 
> This patch combines the authentication timeout and
> authentication retry backoff interval into a single
> wait time interval. Now agent will timeout the previous
> authentication request after the wait time interval and
> then immediately retry.
> 
> 
> Diffs
> -----
> 
>   src/slave/constants.hpp 0bd9f371ca24df66055f31bca0d57625dc7642d9 
>   src/slave/slave.hpp 802d4eb9e9eba2f1175dd85e56dcc80a61e32f74 
>   src/slave/slave.cpp 78e8666f402be58af5b6e20a715da4998af2615c 
>   src/tests/authentication_tests.cpp c9a8f85951a50e278ae509f4efa7105755015ce9 
> 
> 
> Diff: https://reviews.apache.org/r/68304/diff/5/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Meng Zhu
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message