[Gluster-devel] DHT2 for Gl4.0: Project update

[Shyam]

Hi,

Here is an update on DHT2 progress as we exit our self imposed first 3 
week iteration on the same.

The soft update:
----------------
(if you are looking at technical details skip to the next section :) )

Target: The target of the iteration was to exit with some FOPs 
implemented both in the new POSIX xlator and in DHT2. Also, to firm up 
the design to a large extent.

Reality: POSIX xlator changes are WIP, none of the FOPs implemented here 
are tested. DHT2 empty xlator including its required configuration 
options is created. The design itself has come a long way, but is about 
50-60% done.

Next iteration plan:
- Implement basic FOPs for a proof of concept DH2 implementation
- Mature the design further to decide on which approach to take (see 
P1/2 below
- Identify and understand how to resolve dependencies across other xlators

The technical notes:
--------------------
DHT2 has 2 possible implementation options, one where the file name and 
its inode is separated to an MDS and a DS. The other where, in addition 
to the file name and inode, its data extents are also separated, into 
the MDS and DS. Each has its own advantages and possible shortcomings.

These 2 approaches are defined as Phase1 (P1) and Phase2 (P2) of the 
DHT2 prototyping effort. With P1 being the former as stated above.

Here are more details on the design,
P1:
- Design of DHT2: 
https://raw.githubusercontent.com/ShyamsundarR/glusterfs/gl40_dht_playground/xlators/cluster/dht2/docs/DHT2_FirstPrototypeDesign.md
- Functional spec of POSIX: https://review.gerrithub.io/#/c/243885/

P2:
- Initial thoughts on DHT2-P2: https://review.gerrithub.io/#/c/244752/

Here are some interesting problems that arise from DHT2:
1. Quota xlator as it exists today, cannot path walk and update 
accounting information in real time. As it may need to traverse across 
MDS nodes to do the same.
2. Small file creation scale considerations

See the WSR for details on these and some other issues here, [1] [2]

How to track/follow/contribute to the project:
----------------------------------------------
We are hosting the design, code and other artifacts for DHT2 in the 
following gerrithub project,
https://review.gerrithub.io/#/admin/projects/ShyamsundarR/glusterfs

This tracks the github project, 
https://github.com/ShyamsundarR/glusterfs/tree/gl40_dht_playground

Most of the initial commits are separated as design, WSR (weekly status 
reports) and RFC (code patches), in the gerrithub project above.

Those interested can start above, and ping us on gluster-devel in case 
information is lacking on any front.

Thanks,
KP, Venky, Shyam

[1] 
https://raw.githubusercontent.com/ShyamsundarR/glusterfs/gl40_dht_playground/xlators/cluster/dht2/docs/DHT2_WSR_Week002_Aug_20_2015.md
[2] 
https://raw.githubusercontent.com/ShyamsundarR/glusterfs/gl40_dht_playground/xlators/cluster/dht2/docs/DHT2_WSR_Week003_Aug_28_2015.md

On 08/10/2015 01:29 PM, Shyam wrote:
> Hi,
>
> DHT2 initial prototype development is targeted in gerrithub [1] with its
> backed github repos from [2].
>
> The git repos at [1/2] should serve as the single point for all DHT2
> related information going forward.
>
> Here are the important directories to monitor for documentation and code
> for DHT2,
> - xlator/cluster/dht2/docs
> - xlator/cluster/dht2/src
>
> The initial prototype design is derived from the design documents
> previously posted at [3, 4] and, can be found here [5].
>
> The milestones and plans would also be put into the same directory as
> [5] and we intend to maintain weekly update reports on progress et. al.
> at the same place (i.e we would open a review, update it and commit it
> into the repo.)
>
> The current work is to write POSIX layer to suit the on disk layout as
> needed by DHT2, so that we can then start writing DHT2 code.
>
> Work till now can be found in the gerrithub project.
>
> Shyam
>
> [1] https://review.gerrithub.io/#/admin/projects/ShyamsundarR/glusterfs
> [2] https://github.com/ShyamsundarR/glusterfs/tree/gl40_dht_playground
> [3] http://www.gluster.org/pipermail/gluster-devel/2015-June/045449.html
> [4] http://www.gluster.org/pipermail/gluster-devel/2015-June/045591.html
> [5]
> https://github.com/ShyamsundarR/glusterfs/blob/gl40_dht_playground/xlators/cluster/dht2/docs/DHT2_FirstPrototypeDesign.md
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-devel