FAQ
The new git repositories for HDFS and Map/Reduce,
git://git.apache.org/hadoop-{mapreduce,hdfs}.git are up.

There are two open issues:
(1) What to do with the defunct git mirror that points to core
(2) What to call the common mirror

For (1), I'd advocate simply deleting it. There doesn't seem to be a
compelling case for maintaining an archive of a mirror next to its
active replacements.

For (2), the git mirror is currently being built as hadoop-common.git.
Since it will contain the pre-0.21 tagged releases/branches and
developers pulling from the old mirror will error out either way, I'd
lean toward calling it hadoop.git, but am mostly ambivalent on this.

The ticket for this is INFRA-2108. -C

Search Discussions

  • Steve Loughran at Jun 29, 2009 at 11:15 am

    Chris Douglas wrote:
    The new git repositories for HDFS and Map/Reduce,
    git://git.apache.org/hadoop-{mapreduce,hdfs}.git are up.

    There are two open issues:
    (1) What to do with the defunct git mirror that points to core
    (2) What to call the common mirror

    For (1), I'd advocate simply deleting it. There doesn't seem to be a
    compelling case for maintaining an archive of a mirror next to its
    active replacements.
    How about pre-announce a planned deletion some months ahead, give people
    time to migrate off

    For (2), the git mirror is currently being built as hadoop-common.git.
    Since it will contain the pre-0.21 tagged releases/branches and
    developers pulling from the old mirror will error out either way, I'd
    lean toward calling it hadoop.git, but am mostly ambivalent on this.

    The ticket for this is INFRA-2108. -C
  • Chris Douglas at Jun 29, 2009 at 7:02 pm

    How about pre-announce a planned deletion some months ahead, give people
    time to migrate off
    If the repository has been cloned, are there any operations that
    require the remote? We could leave the old mirror there to be polled
    for updates- which it will never have- but I was under the impression
    that a cloned git repo was independent. Nobody should clone the old
    mirror if they're starting anew, so I'd rather just delete it than
    confuse people who reasonably expect "hadoop.git" to contain a current
    version, pull down something that appears to contain core, mapred, and
    hdfs, and then need to migrate their work to the active repositories.
    -C
  • Aaron Kimball at Jun 30, 2009 at 2:18 am
    Chris,

    No operations in git ever require connectivity to an upstream remote, except
    for the obvious ones of "pull more down from remote" and "push local refs up
    to remote." All history and associated metadata is fully replicated to each
    clone.

    - Aaron
    On Mon, Jun 29, 2009 at 12:01 PM, Chris Douglas wrote:

    How about pre-announce a planned deletion some months ahead, give people
    time to migrate off
    If the repository has been cloned, are there any operations that
    require the remote? We could leave the old mirror there to be polled
    for updates- which it will never have- but I was under the impression
    that a cloned git repo was independent. Nobody should clone the old
    mirror if they're starting anew, so I'd rather just delete it than
    confuse people who reasonably expect "hadoop.git" to contain a current
    version, pull down something that appears to contain core, mapred, and
    hdfs, and then need to migrate their work to the active repositories.
    -C
  • Vinod KV at Jun 29, 2009 at 1:30 pm

    Steve Loughran wrote:
    Chris Douglas wrote:
    The new git repositories for HDFS and Map/Reduce,
    git://git.apache.org/hadoop-{mapreduce,hdfs}.git are up.

    There are two open issues:
    (1) What to do with the defunct git mirror that points to core
    (2) What to call the common mirror

    For (1), I'd advocate simply deleting it. There doesn't seem to be a
    compelling case for maintaining an archive of a mirror next to its
    active replacements.
    How about pre-announce a planned deletion some months ahead, give people
    time to migrate off
    +1, for leaving it alone for some time as Steve says.
    For (2), the git mirror is currently being built as hadoop-common.git.
    Since it will contain the pre-0.21 tagged releases/branches and
    developers pulling from the old mirror will error out either way, I'd
    lean toward calling it hadoop.git, but am mostly ambivalent on this.

    The ticket for this is INFRA-2108. -C
    I think hadoop-common.git is better - pre-21 hadoop-common contains
    everything, post that, it only has common stuff. But like you, have no
    strong opinion on this.

    +Vinod
  • Todd Lipcon at Jun 30, 2009 at 12:19 am
    Has anyone started using the new split repositories yet? I have a major
    concern, which is that there is no history for the -mapreduce and -hdfs
    repositories. The initial commit is:

    commit 546d96754ffee3142bcbbf4563c624c053d0ed0d
    Author: Owen O'Malley <omalley@apache.org>
    Date: Tue May 19 04:20:40 2009 +0000

    HADOOP-4687 Moving src directories on branch

    Unfortunately this means that a "git log" or "git blame" on any file in
    MapReduce or HDFS is unable to give any useful information. I do these
    operations several times every day, and it's going to be a big pain to have
    to go back and forth to the "last pre-split 0.21-dev" in hadoop-common to
    track changes.

    -Todd
    On Mon, Jun 29, 2009 at 6:25 AM, Vinod KV wrote:

    Steve Loughran wrote:
    Chris Douglas wrote:
    The new git repositories for HDFS and Map/Reduce,
    git://git.apache.org/hadoop-{mapreduce,hdfs}.git<http://git.apache.org/hadoop-%7Bmapreduce,hdfs%7D.git>are up.
    There are two open issues:
    (1) What to do with the defunct git mirror that points to core
    (2) What to call the common mirror
    For (1), I'd advocate simply deleting it. There doesn't seem to be a
    compelling case for maintaining an archive of a mirror next to its
    active replacements.
    How about pre-announce a planned deletion some months ahead, give people
    time to migrate off

    +1, for leaving it alone for some time as Steve says.
    For (2), the git mirror is currently being built as hadoop-common.git.
    Since it will contain the pre-0.21 tagged releases/branches and
    developers pulling from the old mirror will error out either way, I'd
    lean toward calling it hadoop.git, but am mostly ambivalent on this.
    The ticket for this is INFRA-2108. -C
    I think hadoop-common.git is better - pre-21 hadoop-common contains
    everything, post that, it only has common stuff. But like you, have no
    strong opinion on this.

    +Vinod

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
groupcommon-dev @
categorieshadoop
postedJun 29, '09 at 1:29a
activeJun 30, '09 at 2:18a
posts6
users6
websitehadoop.apache.org...
irc#hadoop

People

Translate

site design / logo © 2022 Grokbase