You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@accumulo.apache.org by John Vines <vi...@apache.org> on 2014/04/07 16:54:14 UTC

Backwards compatibility of traces

Given our support of backwards compatibility, does anyone have a preference
for how traces are stored?

I ask because I have noticed some places where our internal trace hooks are
named with collisions so that you get multiple events with the same
identifier which are actually different events. I would like to change
them, ideally for 1.5.2 and 1.6.1(0?) simply because it's hampering the
ability to do performance testing and tuning.

But I'm not sure if A. we're okay making these changes from a compatibility
perspective and B. we're okay making these changes as part of a minor
release.

Re: Backwards compatibility of traces

Posted by John Vines <vi...@apache.org>.
 3457+0      shell@john-P15xEMx shell:root
    9+1603     shell@john-P15xEMx close
    9+1603
shell@john-P15xEMxorg.apache.accumulo.core.client.impl.TabletServerBatchWriter$MutationWriter
1
    8+1604
shell@john-P15xEMxorg.apache.accumulo.core.client.impl.TabletServerBatchWriter$MutationWriter
1
    8+1604         shell@john-P15xEMx sendMutations
    7+1605           tserver@localhost update
    7+1605             tserver@localhost wal
    4+1607               tserver@localhost update
    4+1607                 tserver@localhost wal
    1+1608                   tserver@localhost client:update
    1+1609                   tserver@localhost update
    1+1609                     tserver@localhost wal



On Mon, Apr 7, 2014 at 11:05 AM, Mike Drob <ma...@cloudera.com> wrote:

> Can you give an example of where this happens?
>
>
> On Mon, Apr 7, 2014 at 10:54 AM, John Vines <vi...@apache.org> wrote:
>
> > Given our support of backwards compatibility, does anyone have a
> preference
> > for how traces are stored?
> >
> > I ask because I have noticed some places where our internal trace hooks
> are
> > named with collisions so that you get multiple events with the same
> > identifier which are actually different events. I would like to change
> > them, ideally for 1.5.2 and 1.6.1(0?) simply because it's hampering the
> > ability to do performance testing and tuning.
> >
> > But I'm not sure if A. we're okay making these changes from a
> compatibility
> > perspective and B. we're okay making these changes as part of a minor
> > release.
> >
>

Re: Backwards compatibility of traces

Posted by Mike Drob <ma...@cloudera.com>.
Can you give an example of where this happens?


On Mon, Apr 7, 2014 at 10:54 AM, John Vines <vi...@apache.org> wrote:

> Given our support of backwards compatibility, does anyone have a preference
> for how traces are stored?
>
> I ask because I have noticed some places where our internal trace hooks are
> named with collisions so that you get multiple events with the same
> identifier which are actually different events. I would like to change
> them, ideally for 1.5.2 and 1.6.1(0?) simply because it's hampering the
> ability to do performance testing and tuning.
>
> But I'm not sure if A. we're okay making these changes from a compatibility
> perspective and B. we're okay making these changes as part of a minor
> release.
>

Re: Backwards compatibility of traces

Posted by Josh Elser <jo...@gmail.com>.
On 4/7/14, 10:54 AM, John Vines wrote:
> Given our support of backwards compatibility, does anyone have a preference
> for how traces are stored?
>
> I ask because I have noticed some places where our internal trace hooks are
> named with collisions so that you get multiple events with the same
> identifier which are actually different events. I would like to change
> them, ideally for 1.5.2 and 1.6.1(0?) simply because it's hampering the
> ability to do performance testing and tuning.
>
> But I'm not sure if A. we're okay making these changes from a compatibility
> perspective and B. we're okay making these changes as part of a minor
> release.

If they're ambiguous to you, then they're ambiguous to any user. If you 
don't have enough information to understand what the traces are timing, 
that is a bug. Fix it in 1.5.2.