Gremlin Server - Terminating the Terminator

[Stephen Mallette <sp...@gmail.com>]

It occurred to me that the Gremlin Server protocol hasn't changed in any
way in months.  A good thing, perhaps, but it also worried me a little, in
that on the other hand the reason it may not have changed is because it
hasn't been paid enough attention.  In giving it another review, I found
that I was quickly reminded of an annoying element of it - the "message
terminator".

Since Gremlin Server streams results back, the protocol need a way to
express that the stream was "done".  I'd implemented this as a terminating
message with a status code 299.  This approach had the effect of marking
the stream as "done" but came with the expense of an extra message for
every request.  If I had n requests which generated a stream of s response
messages, i'd construct (n * 2) + s response messages to get all n requests
met.  I must have grown smarter in the last few months because it only took
a few keystrokes to alter the code to get rid of the terminator leaving us
with n + s response messages which is an obvious improvement.

To accomplish this, I introduced these changes:

1. Dropped status code 299
2. Added status codes 204 (NO_CONTENT) and 206 (PARTIAL_CONTENT)
3. On a successful request with streaming results, the server will return
206 to represent that there are more results to be returned in the stream.
Those results will continue until 200 (SUCCESS) is returned. A successful
result with no streaming (e.g. 1 result in an iterator) will just return a
200.
4. All other status codes outside of 206, including the new 204 (which is
returned for things like an empty iterator) represent terminating
conditions for the stream.

I have this work in a branch at the moment, but would like to move it to
master in time for M9.  As this was a breaking change for the various
clients out there (gremlin-js, aiogremlin, etc), I thought I'd post here
first for comments before introducing it.

Thanks,

Stephen

Re: Gremlin Server - Terminating the Terminator

[Jean-Baptiste Musso <jb...@gmail.com>]

This makes a lot of sense to me. I don't think such change will bring any
issue.

Jean-Baptiste

On Monday, May 4, 2015, Stephen Mallette <sp...@gmail.com> wrote:

> It occurred to me that the Gremlin Server protocol hasn't changed in any
> way in months.  A good thing, perhaps, but it also worried me a little, in
> that on the other hand the reason it may not have changed is because it
> hasn't been paid enough attention.  In giving it another review, I found
> that I was quickly reminded of an annoying element of it - the "message
> terminator".
>
> Since Gremlin Server streams results back, the protocol need a way to
> express that the stream was "done".  I'd implemented this as a terminating
> message with a status code 299.  This approach had the effect of marking
> the stream as "done" but came with the expense of an extra message for
> every request.  If I had n requests which generated a stream of s response
> messages, i'd construct (n * 2) + s response messages to get all n requests
> met.  I must have grown smarter in the last few months because it only took
> a few keystrokes to alter the code to get rid of the terminator leaving us
> with n + s response messages which is an obvious improvement.
>
> To accomplish this, I introduced these changes:
>
> 1. Dropped status code 299
> 2. Added status codes 204 (NO_CONTENT) and 206 (PARTIAL_CONTENT)
> 3. On a successful request with streaming results, the server will return
> 206 to represent that there are more results to be returned in the stream.
> Those results will continue until 200 (SUCCESS) is returned. A successful
> result with no streaming (e.g. 1 result in an iterator) will just return a
> 200.
> 4. All other status codes outside of 206, including the new 204 (which is
> returned for things like an empty iterator) represent terminating
> conditions for the stream.
>
> I have this work in a branch at the moment, but would like to move it to
> master in time for M9.  As this was a breaking change for the various
> clients out there (gremlin-js, aiogremlin, etc), I thought I'd post here
> first for comments before introducing it.
>
> Thanks,
>
> Stephen
>


-- 
Jean-Baptiste

Re: Gremlin Server - Terminating the Terminator

[Stephen Mallette <sp...@gmail.com>]

Hearing your findings on the id issue would be great - M9 should have
sorted those out.  Confirmation on that would be appreciated.

On Tue, May 5, 2015 at 9:03 AM, Dylan Millikin <dy...@gmail.com>
wrote:

> Great,
>
> I'll probably have more feedback once I get a proper go at the TP3 stack.
> I'm mostly waiting for Titan to build against M9 because of the ID
> lossiness.
>
> On Mon, May 4, 2015 at 4:26 PM, Stephen Mallette <sp...@gmail.com>
> wrote:
>
> > Thanks for the fast feedback - I've merged this change back to master -
> it
> > will be available for 3.0.0.M9
> >
> > On Mon, May 4, 2015 at 2:07 PM, Dylan Millikin <dylan.millikin@gmail.com
> >
> > wrote:
> >
> > > Same, This is an easy change on the driver end and makes more sense as
> > > well.
> > >
> > > On Mon, May 4, 2015 at 7:37 AM, Stephen Mallette <spmallette@gmail.com
> >
> > > wrote:
> > >
> > > > It occurred to me that the Gremlin Server protocol hasn't changed in
> > any
> > > > way in months.  A good thing, perhaps, but it also worried me a
> little,
> > > in
> > > > that on the other hand the reason it may not have changed is because
> it
> > > > hasn't been paid enough attention.  In giving it another review, I
> > found
> > > > that I was quickly reminded of an annoying element of it - the
> "message
> > > > terminator".
> > > >
> > > > Since Gremlin Server streams results back, the protocol need a way to
> > > > express that the stream was "done".  I'd implemented this as a
> > > terminating
> > > > message with a status code 299.  This approach had the effect of
> > marking
> > > > the stream as "done" but came with the expense of an extra message
> for
> > > > every request.  If I had n requests which generated a stream of s
> > > response
> > > > messages, i'd construct (n * 2) + s response messages to get all n
> > > requests
> > > > met.  I must have grown smarter in the last few months because it
> only
> > > took
> > > > a few keystrokes to alter the code to get rid of the terminator
> leaving
> > > us
> > > > with n + s response messages which is an obvious improvement.
> > > >
> > > > To accomplish this, I introduced these changes:
> > > >
> > > > 1. Dropped status code 299
> > > > 2. Added status codes 204 (NO_CONTENT) and 206 (PARTIAL_CONTENT)
> > > > 3. On a successful request with streaming results, the server will
> > return
> > > > 206 to represent that there are more results to be returned in the
> > > stream.
> > > > Those results will continue until 200 (SUCCESS) is returned. A
> > successful
> > > > result with no streaming (e.g. 1 result in an iterator) will just
> > return
> > > a
> > > > 200.
> > > > 4. All other status codes outside of 206, including the new 204
> (which
> > is
> > > > returned for things like an empty iterator) represent terminating
> > > > conditions for the stream.
> > > >
> > > > I have this work in a branch at the moment, but would like to move it
> > to
> > > > master in time for M9.  As this was a breaking change for the various
> > > > clients out there (gremlin-js, aiogremlin, etc), I thought I'd post
> > here
> > > > first for comments before introducing it.
> > > >
> > > > Thanks,
> > > >
> > > > Stephen
> > > >
> > >
> >
>

Re: Gremlin Server - Terminating the Terminator

[Dylan Millikin <dy...@gmail.com>]

Great,

I'll probably have more feedback once I get a proper go at the TP3 stack.
I'm mostly waiting for Titan to build against M9 because of the ID
lossiness.

On Mon, May 4, 2015 at 4:26 PM, Stephen Mallette <sp...@gmail.com>
wrote:

> Thanks for the fast feedback - I've merged this change back to master - it
> will be available for 3.0.0.M9
>
> On Mon, May 4, 2015 at 2:07 PM, Dylan Millikin <dy...@gmail.com>
> wrote:
>
> > Same, This is an easy change on the driver end and makes more sense as
> > well.
> >
> > On Mon, May 4, 2015 at 7:37 AM, Stephen Mallette <sp...@gmail.com>
> > wrote:
> >
> > > It occurred to me that the Gremlin Server protocol hasn't changed in
> any
> > > way in months.  A good thing, perhaps, but it also worried me a little,
> > in
> > > that on the other hand the reason it may not have changed is because it
> > > hasn't been paid enough attention.  In giving it another review, I
> found
> > > that I was quickly reminded of an annoying element of it - the "message
> > > terminator".
> > >
> > > Since Gremlin Server streams results back, the protocol need a way to
> > > express that the stream was "done".  I'd implemented this as a
> > terminating
> > > message with a status code 299.  This approach had the effect of
> marking
> > > the stream as "done" but came with the expense of an extra message for
> > > every request.  If I had n requests which generated a stream of s
> > response
> > > messages, i'd construct (n * 2) + s response messages to get all n
> > requests
> > > met.  I must have grown smarter in the last few months because it only
> > took
> > > a few keystrokes to alter the code to get rid of the terminator leaving
> > us
> > > with n + s response messages which is an obvious improvement.
> > >
> > > To accomplish this, I introduced these changes:
> > >
> > > 1. Dropped status code 299
> > > 2. Added status codes 204 (NO_CONTENT) and 206 (PARTIAL_CONTENT)
> > > 3. On a successful request with streaming results, the server will
> return
> > > 206 to represent that there are more results to be returned in the
> > stream.
> > > Those results will continue until 200 (SUCCESS) is returned. A
> successful
> > > result with no streaming (e.g. 1 result in an iterator) will just
> return
> > a
> > > 200.
> > > 4. All other status codes outside of 206, including the new 204 (which
> is
> > > returned for things like an empty iterator) represent terminating
> > > conditions for the stream.
> > >
> > > I have this work in a branch at the moment, but would like to move it
> to
> > > master in time for M9.  As this was a breaking change for the various
> > > clients out there (gremlin-js, aiogremlin, etc), I thought I'd post
> here
> > > first for comments before introducing it.
> > >
> > > Thanks,
> > >
> > > Stephen
> > >
> >
>

Re: Gremlin Server - Terminating the Terminator

[Stephen Mallette <sp...@gmail.com>]

Thanks for the fast feedback - I've merged this change back to master - it
will be available for 3.0.0.M9

On Mon, May 4, 2015 at 2:07 PM, Dylan Millikin <dy...@gmail.com>
wrote:

> Same, This is an easy change on the driver end and makes more sense as
> well.
>
> On Mon, May 4, 2015 at 7:37 AM, Stephen Mallette <sp...@gmail.com>
> wrote:
>
> > It occurred to me that the Gremlin Server protocol hasn't changed in any
> > way in months.  A good thing, perhaps, but it also worried me a little,
> in
> > that on the other hand the reason it may not have changed is because it
> > hasn't been paid enough attention.  In giving it another review, I found
> > that I was quickly reminded of an annoying element of it - the "message
> > terminator".
> >
> > Since Gremlin Server streams results back, the protocol need a way to
> > express that the stream was "done".  I'd implemented this as a
> terminating
> > message with a status code 299.  This approach had the effect of marking
> > the stream as "done" but came with the expense of an extra message for
> > every request.  If I had n requests which generated a stream of s
> response
> > messages, i'd construct (n * 2) + s response messages to get all n
> requests
> > met.  I must have grown smarter in the last few months because it only
> took
> > a few keystrokes to alter the code to get rid of the terminator leaving
> us
> > with n + s response messages which is an obvious improvement.
> >
> > To accomplish this, I introduced these changes:
> >
> > 1. Dropped status code 299
> > 2. Added status codes 204 (NO_CONTENT) and 206 (PARTIAL_CONTENT)
> > 3. On a successful request with streaming results, the server will return
> > 206 to represent that there are more results to be returned in the
> stream.
> > Those results will continue until 200 (SUCCESS) is returned. A successful
> > result with no streaming (e.g. 1 result in an iterator) will just return
> a
> > 200.
> > 4. All other status codes outside of 206, including the new 204 (which is
> > returned for things like an empty iterator) represent terminating
> > conditions for the stream.
> >
> > I have this work in a branch at the moment, but would like to move it to
> > master in time for M9.  As this was a breaking change for the various
> > clients out there (gremlin-js, aiogremlin, etc), I thought I'd post here
> > first for comments before introducing it.
> >
> > Thanks,
> >
> > Stephen
> >
>

Re: Gremlin Server - Terminating the Terminator

[Dylan Millikin <dy...@gmail.com>]

Same, This is an easy change on the driver end and makes more sense as well.

On Mon, May 4, 2015 at 7:37 AM, Stephen Mallette <sp...@gmail.com>
wrote:

> It occurred to me that the Gremlin Server protocol hasn't changed in any
> way in months.  A good thing, perhaps, but it also worried me a little, in
> that on the other hand the reason it may not have changed is because it
> hasn't been paid enough attention.  In giving it another review, I found
> that I was quickly reminded of an annoying element of it - the "message
> terminator".
>
> Since Gremlin Server streams results back, the protocol need a way to
> express that the stream was "done".  I'd implemented this as a terminating
> message with a status code 299.  This approach had the effect of marking
> the stream as "done" but came with the expense of an extra message for
> every request.  If I had n requests which generated a stream of s response
> messages, i'd construct (n * 2) + s response messages to get all n requests
> met.  I must have grown smarter in the last few months because it only took
> a few keystrokes to alter the code to get rid of the terminator leaving us
> with n + s response messages which is an obvious improvement.
>
> To accomplish this, I introduced these changes:
>
> 1. Dropped status code 299
> 2. Added status codes 204 (NO_CONTENT) and 206 (PARTIAL_CONTENT)
> 3. On a successful request with streaming results, the server will return
> 206 to represent that there are more results to be returned in the stream.
> Those results will continue until 200 (SUCCESS) is returned. A successful
> result with no streaming (e.g. 1 result in an iterator) will just return a
> 200.
> 4. All other status codes outside of 206, including the new 204 (which is
> returned for things like an empty iterator) represent terminating
> conditions for the stream.
>
> I have this work in a branch at the moment, but would like to move it to
> master in time for M9.  As this was a breaking change for the various
> clients out there (gremlin-js, aiogremlin, etc), I thought I'd post here
> first for comments before introducing it.
>
> Thanks,
>
> Stephen
>