hadoop-hdfs-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jakob Homan <jho...@yahoo-inc.com>
Subject Re: Changes to FSImage/FSEditLog
Date Thu, 18 Nov 2010 21:34:44 GMT
Glancing at 1473, it seems reasonable, but of course should get a 
thorough review.  HDFS-1465 will impact and essentially replace all the 
saving/loading code, so whether the it does so with the code's current 
state or with 1473's refactoring, it doesn't matter much.  1473 will 
probably make it easier.

It would be more useful if 1465 went in before 1070, because it would 
save Hairong work, but I believe she's already started and I just don't 
have the bandwidth to do it at the moment, but I'm hoping that it will 
be the next thing I work on.  The end result for the image, when we're 
all done, will hopefully be refactored, unified and efficient 
loading/saving code.

It should be noted that HDFS-1448, and the de-duplication JIRA that's 
been agreed to as its sequel, will also impact on all of the edits work.
-jg

Ivan Kelly wrote:
> This patch looks really good. It shouldn't affect our refactoring too much and will actually
make things a lot easier for the next part (breaking the circular dependency between namesystem
and fsimage).
> 
> -Ivan
> 
> On 18 Nov 2010, at 02:27, Todd Lipcon wrote:
> 
> I put up a sketch of a patch for 1473. It compiles and passes at least a couple of the
unit tests, but more work to be done on it. Would appreciate comments from this crowd if the
new classes seem like good divisions.
> 
> -Todd
> 
> On Wed, Nov 17, 2010 at 2:36 PM, Todd Lipcon <todd@cloudera.com<mailto:todd@cloudera.com>>
wrote:
> On Wed, Nov 17, 2010 at 2:34 PM, Hairong Kuang <kuang.hairong@fb.com<mailto:kuang.hairong@fb.com>>
wrote:
> You meant HDFS-1473 right? When do you expect that it could be done?
> 
> 
> I was working on HDFS-1473 this morning and could probably get a preliminary patch up
today.. but when I realized that we have 4 parallel refactorings going on I paused my work.
If we agree that HDFS-1473 will go in before the other ones I can finish it up, but if we're
doing the same thing we should coordinate instead.
> 
> Hairong
> 
> On 11/17/10 2:29 PM, "Eli Collins" <eli@cloudera.com<mailto:eli@cloudera.com>>
wrote:
> 
>> Agree we should do the refactoring before the features, we should also
>> add the tests with the refactoring to show they don't break things.
>>
>> On Wed, Nov 17, 2010 at 2:27 PM, Todd Lipcon <todd@cloudera.com<mailto:todd@cloudera.com>>
wrote:
>>> OK, looks like there is still quite a bit more in motion here, thanks for
>>> the list, Hairong.
>>> How would everyone feel about setting aside those improvements/fixes, doing
>>> the refactor first, and then going back to them? It should help with testing
>>> and reduce conflicts between the various ongoing projects.
>>> Maybe we can discuss quickly on this thread: what are some small refactors
>>> we can start with to help separate these things? Ivan's project looks really
>>> good but it seems like it is a pretty large change. If we can break it up
>>> into a couple smaller patches we can commit as we go.
>>> -Todd
>>> On Wed, Nov 17, 2010 at 2:08 PM, Hairong Kuang <kuang.hairong@fb.com<mailto:kuang.hairong@fb.com>>
wrote:
>>>> Hi Todd,
>>>>
>>>> Here is the list of fsimage jiras that I am working on:
>>>>
>>>> HDFS-1481 // small change
>>>> HDFS-1458 // small change
>>>> HDFS-1496  // should be checked into 022, I still have no clue how to fix
>>>> it
>>>> HDFS-1070  // this one has a simple idea, but turns out that it needs
>>>> major code re-organization. I am still struggling with how to make my code
>>>> change easier to review. But this will definitely collide with yours.
>>>>
>>>> Hairong
>>>>
>>>> On 11/17/10 11:57 AM, "Todd Lipcon" <todd@cloudera.com<mailto:todd@cloudera.com>>
wrote:
>>>>
>>>> Hey all,
>>>>
>>>> As many of you know, I've been working on HDFS-1073 for a few months, and
>>>> we're hoping to get it in for 0.22 (OOM has given the OK to merge this into
>>>> branch and Sanjay has helpfully gotten some commitment from Yahoo QA to help
>>>> test it).
>>>>
>>>> It's been difficult recently to work on the patch as a lot of changes have
>>>> gone in around FSImage and FSEditLog (eg checksumming, parallel load, etc)
>>>> so the patch falls out of date quickly and is hard to resolve the changes
>>>> since my patch moves a lot of code. Ivan Kelly over at Yahoo Barcelona has
>>>> also been working on a refactor and I think running into a similar issue.
>>>>
>>>> So, I just wanted to ping everyone who has been working on this area of
>>>> the code to find out if there are any more remaining JIRAs that you're
>>>> hoping to commit in the next several weeks that will touch FSImage and
>>>> FSEditLog. If so, we should try to agree on an order in which to commit them
>>>> so we don't end up stomping on each other, and we can help review each
>>>> other's work to move quicker.
>>>>
>>>> If you can just reply with a list of any remaining uncommitted JIRAs that
>>>> touch FSImage/FSEditLog that would be great, then we can work from there
on
>>>> how to proceed and minimize the amount of patch rebasing we all have to do.
>>>>
>>>> Thanks!
>>>> -Todd
>>>
>>>
>>> --
>>> Todd Lipcon
>>> Software Engineer, Cloudera
>>>
> 
> 
> 
> 
> --
> Todd Lipcon
> Software Engineer, Cloudera
> 
> 
> 
> --
> Todd Lipcon
> Software Engineer, Cloudera
> 


Mime
View raw message