mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrei Sekretenko <asekrete...@mesosphere.io>
Subject Re: Review Request 72003: Used persisted information to decide if an executor is agent-generated.
Date Wed, 15 Jan 2020 13:55:05 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/72003/#review219270
-----------------------------------------------------------




src/slave/slave.cpp
Line 6583 (original), 6583 (patched)
<https://reviews.apache.org/r/72003/#comment307435>

    If we are not validating that `ExecutorInfo` of a task does not have `generated_by_agent`
set, maybe we can at least forcibly clear it? either here or in `Slave::runTask()`, and document
(in the previous patch) that agent **ignores** this field when received from outside?



src/slave/slave.cpp
Lines 6593 (patched)
<https://reviews.apache.org/r/72003/#comment307434>

    The logic "executor is generated by agent -> master receives no updates" is not covered
by any of the local tests; if we are not introducing the test right now, consider adding TODO
(and, probably, a separate ticket).
    
    Personally, I'm more or less OK with `upgradeExecutorInfo(...)` not covered: the necessary
setup is difficult to create, that code is hacky anyway, and we want to phase it out.



src/slave/slave.cpp
Lines 10604 (patched)
<https://reviews.apache.org/r/72003/#comment307433>

    My impression is that adding some logging here could help in case this hack misfires.
    
    Something like
    ```
    LOG(WARNING) << "Detecting whether executor " << execiutor.id() 
                 << " was generated by agent: executor command is " <<  executor.command().value()

                 << "; current path of " << MESOS_EXECUTOR << " binary is
" << executorPath.get();
    ```
    ?



src/slave/slave.cpp
Line 10601 (original), 10626 (patched)
<https://reviews.apache.org/r/72003/#comment307432>

    Maybe add/move CHECK_NOTNULL to this line, so that it does not rely on the fact that `slave`
is **declared** before `info`?


- Andrei Sekretenko


On Jan. 15, 2020, 11:25 a.m., Benjamin Bannier wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/72003/
> -----------------------------------------------------------
> 
> (Updated Jan. 15, 2020, 11:25 a.m.)
> 
> 
> Review request for mesos, Andrei Sekretenko and Benjamin Mahler.
> 
> 
> Bugs: MESOS-10084
>     https://issues.apache.org/jira/browse/MESOS-10084
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This patch modifies the agent code so it uses
> `ExecutorInfo::generated_by_agent` instead of the previous heuristic to
> detect agent-generated executors for command tasks. For that we add
> recovery logic for upgrade scenarios to inject a correct value for
> recovered executors, and set the appropriate value for newly created
> executors.
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.hpp 3d191dcbdd58e851ad5e86fbb01a890407409c64 
>   src/slave/slave.cpp 0005971717f5b90da368b45caad8e209ada95fa5 
> 
> 
> Diff: https://reviews.apache.org/r/72003/diff/1/
> 
> 
> Testing
> -------
> 
> `make check`
> 
> 
> Thanks,
> 
> Benjamin Bannier
> 
>


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