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 70390: Added a test to ensure correct headroom accounting.
Date Thu, 04 Apr 2019 22:02:26 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70390/#review214385
-----------------------------------------------------------


Fix it, then Ship it!




Does this test fail on master prior to the fixes?


src/tests/hierarchical_allocator_tests.cpp
Lines 6176-6178 (patched)
<https://reviews.apache.org/r/70390/#comment300576>

    just as an FYI, when I read this I was expecting it would test:
    
    "a" <-- has quota
    "a/b"
    
    (where it ensures that "a/b"'s allocation counts towards "a"'s quota)
    
    I guess we don't have a test for this, but it looks like a variant of 'QuotaWithNestedRoleReservation'
will accomplish it.
    
    Maybe call this one QuotaHeadroomWithNested...



src/tests/hierarchical_allocator_tests.cpp
Lines 6180-6186 (patched)
<https://reviews.apache.org/r/70390/#comment300579>

    Simple numbers like "cpus:1;mem:100" seem to make the tests easier to read, any reason
not to do that?



src/tests/hierarchical_allocator_tests.cpp
Lines 6180-6193 (patched)
<https://reviews.apache.org/r/70390/#comment300585>

    I found that while debugging some of these tests, the "imperative" style of explaining
them (e.g. do x then y then x) makes the tests a bit tougher to understand, and that "declarative"
explanations might be more readable? E.g.
    
    ```
    // Setup:
    //   agents: 2 * R
    //    roles: "a"   --> allocated R
    //           "a/b" --> allocated R
    //           "quota-role" --> guarantee R w/ no framework
    //
    // Test: Add 1 more agent with R.
    //       Ensure agent is held back for "quota-role".
    //       Add 1 more agent with R.
    //       Ensure only 1 of the two extra agents goes to "a" or "a/b"
    //         (since there is enough headroom for "quota-role")
    ```
    
    Here I can more easily see what setup state looks like and what is then tested off of
that setup.



src/tests/hierarchical_allocator_tests.cpp
Lines 6199-6200 (patched)
<https://reviews.apache.org/r/70390/#comment300590>

    I realize it's not just this test, but this brace syntax but it seems less readable than:
    
    const string CHILD_ROLE = "a/b";
    
    Does that not compile?



src/tests/hierarchical_allocator_tests.cpp
Lines 6221 (patched)
<https://reviews.apache.org/r/70390/#comment300588>

    parent role?



src/tests/hierarchical_allocator_tests.cpp
Lines 6241 (patched)
<https://reviews.apache.org/r/70390/#comment300586>

    quota-role?


- Benjamin Mahler


On April 4, 2019, 8:31 p.m., Meng Zhu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70390/
> -----------------------------------------------------------
> 
> (Updated April 4, 2019, 8:31 p.m.)
> 
> 
> Review request for mesos and Benjamin Mahler.
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This test verifies that the headroom accounting is correct in the
> presence of subrole allocation. In particular, we need to ensure
> that a subrole's allocation is accounted only once (not multiple
> times in itself as well as all of its ancestors) when calculating
> available the quota headroom.
> 
> 
> Diffs
> -----
> 
>   src/tests/hierarchical_allocator_tests.cpp c7edee2f49bfac0796c9506265d3bb766416ee63

> 
> 
> Diff: https://reviews.apache.org/r/70390/diff/1/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Meng Zhu
> 
>


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