RFC: rename "remote bootstrap" to "tablet copy"?

[Todd Lipcon <to...@cloudera.com>]

Hi all,

Currently we use the term "remote bootstrap" to mean the process by which
one tablet server copies a tablet from another. This terminology started
long long ago, back when we thought there was some chance that it would
actually _start_ the tablet from remote storage, which isn't at all what
the current design does.

I think it would be more accurate and cause less confusion if we considered
changing our terminology here. The current terminology invites confusion
with "tablet bootstrap", the process by which a tablet opens its data,
replays its logs, and starts. A couple suggestions would be:

- Tablet copy (since we're copying a tablet from one host to another)
- Tablet snapshot transfer (a mouthful,but fairly accurate)
- Remote tablet copy (slightly more precise than "tablet copy" but also
longer)

What do people think? Changing this before we hit 1.0 would be nice just so
we have time to update docs, metric names, etc, even though it doesn't
affect any public APIs.

-Todd
-- 
Todd Lipcon
Software Engineer, Cloudera

Re: RFC: rename "remote bootstrap" to "tablet copy"?

[Todd Lipcon <to...@cloudera.com>]

I put up a preliminary patch here which appears to compile:
https://gerrit.cloudera.org/#/c/3848

I did this with fully automated find/replace substitutions. I'm now
self-reviewing the patch to spot places that some manual editing is
required to make it read fluently with the new name.

-Todd

On Mon, Aug 1, 2016 at 3:42 PM, Todd Lipcon <to...@cloudera.com> wrote:

> On Mon, Aug 1, 2016 at 3:39 PM, Mike Percy <mp...@apache.org> wrote:
>
>> +1 on the name change from remote bootstrap.
>>
>> I like tablet copy. The pro side of using the "snapshot" terminology is
>> that the Raft paper uses that terminology for the same thing. The downside
>> is that we have snapshot scans also, so we've sort of already claimed that
>> term for a totally different concept.
>>
>
> Yea, that's a good point. Also, at some point we may want to implement a
> "snapshot" feature for tables, in which case that terminology again might
> get cloudy.
>
> Leaning towards "tablet copy" given it's both short to type and doesn't
> have any overlap with existing terms for other stuff.
>
> I'll prepare a patch in the next couple days.
>
> -Todd
>
>
>>
>> On Mon, Aug 1, 2016 at 3:24 PM, Dan Burkert <da...@cloudera.com> wrote:
>>
>> > I support the rename. I like 'tablet snapshot copy/transfer' over
>> 'tablet
>> > copy', but not by a big margin.  Both are a lot clearer than 'remote
>> > bootstrap'.
>> >
>> > - Dan
>> >
>> > On Mon, Aug 1, 2016 at 3:10 PM, Adar Dembo <ad...@cloudera.com> wrote:
>> >
>> > > I don't have much of an opinion one way or the other. I'm not put off
>> > > by the current name but wouldn't object to changing it either,
>> > > provided you're signing up for the work. :)
>> > >
>> > > Tablet copy seems like a reasonable name to me.
>> > >
>> > > On Sun, Jul 31, 2016 at 5:13 PM, Alexey Serbin <as...@cloudera.com>
>> > > wrote:
>> > > > Todd,
>> > > >
>> > > > Change from 'remote bootstrap' to something like 'tablet copy'
>> sounds
>> > > > reasonable.  As I see from the description, the 'tablet copy' better
>> > > > reflects the essence of the process.  Both 'tablet copy' and 'tablet
>> > > > snapshot transfer' are better than 'remote bootstrap', IMO.  Not
>> sure
>> > > > whether the additional 'remote' brings more clarity: is it possible
>> to
>> > > have
>> > > > a 'local tablet copy' at all?
>> > > >
>> > > > BTW, if 'replica' or 'replication' terms aren't too overloaded and
>> > > > applicable in this context, consider a couple of additional options:
>> > > > - Tablet replication
>> > > > - Creating tablet replica
>> > > >
>> > > >
>> > > > Best regards,
>> > > >
>> > > > Alexey
>> > > >
>> > > > On Sun, Jul 31, 2016 at 3:45 PM, Todd Lipcon <to...@cloudera.com>
>> > wrote:
>> > > >
>> > > >> Hi all,
>> > > >>
>> > > >> Currently we use the term "remote bootstrap" to mean the process by
>> > > which
>> > > >> one tablet server copies a tablet from another. This terminology
>> > started
>> > > >> long long ago, back when we thought there was some chance that it
>> > would
>> > > >> actually _start_ the tablet from remote storage, which isn't at all
>> > what
>> > > >> the current design does.
>> > > >>
>> > > >> I think it would be more accurate and cause less confusion if we
>> > > considered
>> > > >> changing our terminology here. The current terminology invites
>> > confusion
>> > > >> with "tablet bootstrap", the process by which a tablet opens its
>> data,
>> > > >> replays its logs, and starts. A couple suggestions would be:
>> > > >>
>> > > >> - Tablet copy (since we're copying a tablet from one host to
>> another)
>> > > >> - Tablet snapshot transfer (a mouthful,but fairly accurate)
>> > > >> - Remote tablet copy (slightly more precise than "tablet copy" but
>> > also
>> > > >> longer)
>> > > >>
>> > > >> What do people think? Changing this before we hit 1.0 would be nice
>> > > just so
>> > > >> we have time to update docs, metric names, etc, even though it
>> doesn't
>> > > >> affect any public APIs.
>> > > >>
>> > > >> -Todd
>> > > >> --
>> > > >> Todd Lipcon
>> > > >> Software Engineer, Cloudera
>> > > >>
>> > >
>> >
>>
>
>
>
> --
> Todd Lipcon
> Software Engineer, Cloudera
>



-- 
Todd Lipcon
Software Engineer, Cloudera

Re: RFC: rename "remote bootstrap" to "tablet copy"?

[Todd Lipcon <to...@cloudera.com>]

On Mon, Aug 1, 2016 at 3:39 PM, Mike Percy <mp...@apache.org> wrote:

> +1 on the name change from remote bootstrap.
>
> I like tablet copy. The pro side of using the "snapshot" terminology is
> that the Raft paper uses that terminology for the same thing. The downside
> is that we have snapshot scans also, so we've sort of already claimed that
> term for a totally different concept.
>

Yea, that's a good point. Also, at some point we may want to implement a
"snapshot" feature for tables, in which case that terminology again might
get cloudy.

Leaning towards "tablet copy" given it's both short to type and doesn't
have any overlap with existing terms for other stuff.

I'll prepare a patch in the next couple days.

-Todd


>
> On Mon, Aug 1, 2016 at 3:24 PM, Dan Burkert <da...@cloudera.com> wrote:
>
> > I support the rename. I like 'tablet snapshot copy/transfer' over 'tablet
> > copy', but not by a big margin.  Both are a lot clearer than 'remote
> > bootstrap'.
> >
> > - Dan
> >
> > On Mon, Aug 1, 2016 at 3:10 PM, Adar Dembo <ad...@cloudera.com> wrote:
> >
> > > I don't have much of an opinion one way or the other. I'm not put off
> > > by the current name but wouldn't object to changing it either,
> > > provided you're signing up for the work. :)
> > >
> > > Tablet copy seems like a reasonable name to me.
> > >
> > > On Sun, Jul 31, 2016 at 5:13 PM, Alexey Serbin <as...@cloudera.com>
> > > wrote:
> > > > Todd,
> > > >
> > > > Change from 'remote bootstrap' to something like 'tablet copy' sounds
> > > > reasonable.  As I see from the description, the 'tablet copy' better
> > > > reflects the essence of the process.  Both 'tablet copy' and 'tablet
> > > > snapshot transfer' are better than 'remote bootstrap', IMO.  Not sure
> > > > whether the additional 'remote' brings more clarity: is it possible
> to
> > > have
> > > > a 'local tablet copy' at all?
> > > >
> > > > BTW, if 'replica' or 'replication' terms aren't too overloaded and
> > > > applicable in this context, consider a couple of additional options:
> > > > - Tablet replication
> > > > - Creating tablet replica
> > > >
> > > >
> > > > Best regards,
> > > >
> > > > Alexey
> > > >
> > > > On Sun, Jul 31, 2016 at 3:45 PM, Todd Lipcon <to...@cloudera.com>
> > wrote:
> > > >
> > > >> Hi all,
> > > >>
> > > >> Currently we use the term "remote bootstrap" to mean the process by
> > > which
> > > >> one tablet server copies a tablet from another. This terminology
> > started
> > > >> long long ago, back when we thought there was some chance that it
> > would
> > > >> actually _start_ the tablet from remote storage, which isn't at all
> > what
> > > >> the current design does.
> > > >>
> > > >> I think it would be more accurate and cause less confusion if we
> > > considered
> > > >> changing our terminology here. The current terminology invites
> > confusion
> > > >> with "tablet bootstrap", the process by which a tablet opens its
> data,
> > > >> replays its logs, and starts. A couple suggestions would be:
> > > >>
> > > >> - Tablet copy (since we're copying a tablet from one host to
> another)
> > > >> - Tablet snapshot transfer (a mouthful,but fairly accurate)
> > > >> - Remote tablet copy (slightly more precise than "tablet copy" but
> > also
> > > >> longer)
> > > >>
> > > >> What do people think? Changing this before we hit 1.0 would be nice
> > > just so
> > > >> we have time to update docs, metric names, etc, even though it
> doesn't
> > > >> affect any public APIs.
> > > >>
> > > >> -Todd
> > > >> --
> > > >> Todd Lipcon
> > > >> Software Engineer, Cloudera
> > > >>
> > >
> >
>



-- 
Todd Lipcon
Software Engineer, Cloudera

Re: RFC: rename "remote bootstrap" to "tablet copy"?

[Mike Percy <mp...@apache.org>]

+1 on the name change from remote bootstrap.

I like tablet copy. The pro side of using the "snapshot" terminology is
that the Raft paper uses that terminology for the same thing. The downside
is that we have snapshot scans also, so we've sort of already claimed that
term for a totally different concept.

Mike


On Mon, Aug 1, 2016 at 3:24 PM, Dan Burkert <da...@cloudera.com> wrote:

> I support the rename. I like 'tablet snapshot copy/transfer' over 'tablet
> copy', but not by a big margin.  Both are a lot clearer than 'remote
> bootstrap'.
>
> - Dan
>
> On Mon, Aug 1, 2016 at 3:10 PM, Adar Dembo <ad...@cloudera.com> wrote:
>
> > I don't have much of an opinion one way or the other. I'm not put off
> > by the current name but wouldn't object to changing it either,
> > provided you're signing up for the work. :)
> >
> > Tablet copy seems like a reasonable name to me.
> >
> > On Sun, Jul 31, 2016 at 5:13 PM, Alexey Serbin <as...@cloudera.com>
> > wrote:
> > > Todd,
> > >
> > > Change from 'remote bootstrap' to something like 'tablet copy' sounds
> > > reasonable.  As I see from the description, the 'tablet copy' better
> > > reflects the essence of the process.  Both 'tablet copy' and 'tablet
> > > snapshot transfer' are better than 'remote bootstrap', IMO.  Not sure
> > > whether the additional 'remote' brings more clarity: is it possible to
> > have
> > > a 'local tablet copy' at all?
> > >
> > > BTW, if 'replica' or 'replication' terms aren't too overloaded and
> > > applicable in this context, consider a couple of additional options:
> > > - Tablet replication
> > > - Creating tablet replica
> > >
> > >
> > > Best regards,
> > >
> > > Alexey
> > >
> > > On Sun, Jul 31, 2016 at 3:45 PM, Todd Lipcon <to...@cloudera.com>
> wrote:
> > >
> > >> Hi all,
> > >>
> > >> Currently we use the term "remote bootstrap" to mean the process by
> > which
> > >> one tablet server copies a tablet from another. This terminology
> started
> > >> long long ago, back when we thought there was some chance that it
> would
> > >> actually _start_ the tablet from remote storage, which isn't at all
> what
> > >> the current design does.
> > >>
> > >> I think it would be more accurate and cause less confusion if we
> > considered
> > >> changing our terminology here. The current terminology invites
> confusion
> > >> with "tablet bootstrap", the process by which a tablet opens its data,
> > >> replays its logs, and starts. A couple suggestions would be:
> > >>
> > >> - Tablet copy (since we're copying a tablet from one host to another)
> > >> - Tablet snapshot transfer (a mouthful,but fairly accurate)
> > >> - Remote tablet copy (slightly more precise than "tablet copy" but
> also
> > >> longer)
> > >>
> > >> What do people think? Changing this before we hit 1.0 would be nice
> > just so
> > >> we have time to update docs, metric names, etc, even though it doesn't
> > >> affect any public APIs.
> > >>
> > >> -Todd
> > >> --
> > >> Todd Lipcon
> > >> Software Engineer, Cloudera
> > >>
> >
>

Re: RFC: rename "remote bootstrap" to "tablet copy"?

[Dan Burkert <da...@cloudera.com>]

I support the rename. I like 'tablet snapshot copy/transfer' over 'tablet
copy', but not by a big margin.  Both are a lot clearer than 'remote
bootstrap'.

- Dan

On Mon, Aug 1, 2016 at 3:10 PM, Adar Dembo <ad...@cloudera.com> wrote:

> I don't have much of an opinion one way or the other. I'm not put off
> by the current name but wouldn't object to changing it either,
> provided you're signing up for the work. :)
>
> Tablet copy seems like a reasonable name to me.
>
> On Sun, Jul 31, 2016 at 5:13 PM, Alexey Serbin <as...@cloudera.com>
> wrote:
> > Todd,
> >
> > Change from 'remote bootstrap' to something like 'tablet copy' sounds
> > reasonable.  As I see from the description, the 'tablet copy' better
> > reflects the essence of the process.  Both 'tablet copy' and 'tablet
> > snapshot transfer' are better than 'remote bootstrap', IMO.  Not sure
> > whether the additional 'remote' brings more clarity: is it possible to
> have
> > a 'local tablet copy' at all?
> >
> > BTW, if 'replica' or 'replication' terms aren't too overloaded and
> > applicable in this context, consider a couple of additional options:
> > - Tablet replication
> > - Creating tablet replica
> >
> >
> > Best regards,
> >
> > Alexey
> >
> > On Sun, Jul 31, 2016 at 3:45 PM, Todd Lipcon <to...@cloudera.com> wrote:
> >
> >> Hi all,
> >>
> >> Currently we use the term "remote bootstrap" to mean the process by
> which
> >> one tablet server copies a tablet from another. This terminology started
> >> long long ago, back when we thought there was some chance that it would
> >> actually _start_ the tablet from remote storage, which isn't at all what
> >> the current design does.
> >>
> >> I think it would be more accurate and cause less confusion if we
> considered
> >> changing our terminology here. The current terminology invites confusion
> >> with "tablet bootstrap", the process by which a tablet opens its data,
> >> replays its logs, and starts. A couple suggestions would be:
> >>
> >> - Tablet copy (since we're copying a tablet from one host to another)
> >> - Tablet snapshot transfer (a mouthful,but fairly accurate)
> >> - Remote tablet copy (slightly more precise than "tablet copy" but also
> >> longer)
> >>
> >> What do people think? Changing this before we hit 1.0 would be nice
> just so
> >> we have time to update docs, metric names, etc, even though it doesn't
> >> affect any public APIs.
> >>
> >> -Todd
> >> --
> >> Todd Lipcon
> >> Software Engineer, Cloudera
> >>
>

Re: RFC: rename "remote bootstrap" to "tablet copy"?

[Adar Dembo <ad...@cloudera.com>]

I don't have much of an opinion one way or the other. I'm not put off
by the current name but wouldn't object to changing it either,
provided you're signing up for the work. :)

Tablet copy seems like a reasonable name to me.

On Sun, Jul 31, 2016 at 5:13 PM, Alexey Serbin <as...@cloudera.com> wrote:
> Todd,
>
> Change from 'remote bootstrap' to something like 'tablet copy' sounds
> reasonable.  As I see from the description, the 'tablet copy' better
> reflects the essence of the process.  Both 'tablet copy' and 'tablet
> snapshot transfer' are better than 'remote bootstrap', IMO.  Not sure
> whether the additional 'remote' brings more clarity: is it possible to have
> a 'local tablet copy' at all?
>
> BTW, if 'replica' or 'replication' terms aren't too overloaded and
> applicable in this context, consider a couple of additional options:
> - Tablet replication
> - Creating tablet replica
>
>
> Best regards,
>
> Alexey
>
> On Sun, Jul 31, 2016 at 3:45 PM, Todd Lipcon <to...@cloudera.com> wrote:
>
>> Hi all,
>>
>> Currently we use the term "remote bootstrap" to mean the process by which
>> one tablet server copies a tablet from another. This terminology started
>> long long ago, back when we thought there was some chance that it would
>> actually _start_ the tablet from remote storage, which isn't at all what
>> the current design does.
>>
>> I think it would be more accurate and cause less confusion if we considered
>> changing our terminology here. The current terminology invites confusion
>> with "tablet bootstrap", the process by which a tablet opens its data,
>> replays its logs, and starts. A couple suggestions would be:
>>
>> - Tablet copy (since we're copying a tablet from one host to another)
>> - Tablet snapshot transfer (a mouthful,but fairly accurate)
>> - Remote tablet copy (slightly more precise than "tablet copy" but also
>> longer)
>>
>> What do people think? Changing this before we hit 1.0 would be nice just so
>> we have time to update docs, metric names, etc, even though it doesn't
>> affect any public APIs.
>>
>> -Todd
>> --
>> Todd Lipcon
>> Software Engineer, Cloudera
>>

Re: RFC: rename "remote bootstrap" to "tablet copy"?

[Alexey Serbin <as...@cloudera.com>]

Todd,

Change from 'remote bootstrap' to something like 'tablet copy' sounds
reasonable.  As I see from the description, the 'tablet copy' better
reflects the essence of the process.  Both 'tablet copy' and 'tablet
snapshot transfer' are better than 'remote bootstrap', IMO.  Not sure
whether the additional 'remote' brings more clarity: is it possible to have
a 'local tablet copy' at all?

BTW, if 'replica' or 'replication' terms aren't too overloaded and
applicable in this context, consider a couple of additional options:
- Tablet replication
- Creating tablet replica


Best regards,

Alexey

On Sun, Jul 31, 2016 at 3:45 PM, Todd Lipcon <to...@cloudera.com> wrote:

> Hi all,
>
> Currently we use the term "remote bootstrap" to mean the process by which
> one tablet server copies a tablet from another. This terminology started
> long long ago, back when we thought there was some chance that it would
> actually _start_ the tablet from remote storage, which isn't at all what
> the current design does.
>
> I think it would be more accurate and cause less confusion if we considered
> changing our terminology here. The current terminology invites confusion
> with "tablet bootstrap", the process by which a tablet opens its data,
> replays its logs, and starts. A couple suggestions would be:
>
> - Tablet copy (since we're copying a tablet from one host to another)
> - Tablet snapshot transfer (a mouthful,but fairly accurate)
> - Remote tablet copy (slightly more precise than "tablet copy" but also
> longer)
>
> What do people think? Changing this before we hit 1.0 would be nice just so
> we have time to update docs, metric names, etc, even though it doesn't
> affect any public APIs.
>
> -Todd
> --
> Todd Lipcon
> Software Engineer, Cloudera
>