-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/24536/#review56243
-----------------------------------------------------------
Can writing DIFFs be optional via the construcotr? I'm hoping to initially
leave this off for the Registrar for two reasons:
(1) Upgrading. It would be nice to be able to upgrade the masters seamlessly
with respect to the Registrar:
Phase 1: Masters can read DIFFs, don't write DIFFs.
Phase 2: Masters write diffs.
If we move directly to phase 2 (as per this patch), then the upgrade of the
masters needs to be done in lock-step to ensure that an old leading master
doesn't come up when there are DIFF entries written, causing unnecessary
failovers. The constructor argument would give us control over this, while
still making it immediately available to users of LogStorage.
(2) Ability to disable diffs. We need to run the Registrar benchmarks for DIFF
enabled LogStorage. However, we'd still like to be able to flip it off should
we encounter issues in a production environment. Without a flag + constructor
argument, we have to roll back to disable it.
Sound reasonable?
- Ben Mahler
On Sept. 29, 2014, 12:45 p.m., Benjamin Hindman wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/24536/
> -----------------------------------------------------------
>
> (Updated Sept. 29, 2014, 12:45 p.m.)
>
>
> Review request for mesos, Ben Mahler and Jie Yu.
>
>
> Repository: mesos-git
>
>
> Description
> -------
>
> See summary.
>
> Note that this hard codes the location of the subversion and Apache Portable
> Runtime (APR) headers.
>
>
> Diffs
> -----
>
> src/Makefile.am 27c42dfde45a449750132e416b4eaf776f8c5e3b
> src/messages/state.proto 59276e55fcbebdb754c20d39b13b402fd11c3dad
> src/state/log.cpp fd8b28a0b5d14f5ba3e6fde4695f6d09acf9c56a
>
> Diff: https://reviews.apache.org/r/24536/diff/
>
>
> Testing
> -------
>
> make check
>
>
> Thanks,
>
> Benjamin Hindman
>
>
