Program Goals and Status for FY18/19

Goal Owner: Guillaume Lederrey
Program Goals for FY18/19: Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to discover and search for content.
Annual Plan: TEC8: Search Platform
- Primary Goal is Knowledge as a Service: Evolve our systems and structures
- Tech Goal: Supporting our Community of contributors

Q1 Goals

Outcome 1 / Output 1

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Incorporate Natural Language Processing (NLP) in the machine learning analysis pipeline for search

Dependency: Will need some short-term consulting help during implementation

Goal(s)

Select 1 or 2 NLP applications and prototype the features

Status

Note: July 2018

In progress

Note: August 21, 2018

In progress Contract contents written up and will start recruiting soon.

Note: September 20, 2018

Partially done this work continues but the prototype won't be completed in this quarter.

Outcome 1 / Output 2

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Evaluation of image features for search ranking

Goal(s)

Investigate and evaluate image level features for image search ranking (ie. Image quality score in ML indexing) (Stretch goal)

Status

Note: July 2018

In progress

Note: August 21, 2018

In progress We're using an older test that Miriam Redi created, and meetings / questions and answers are ongoing

Note: September 25, 2018

Done.

Outcome 1 / Output 3

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Better understanding of the effectiveness of our improvements to search and the performance of our tooling on the back end

Dependency: Analytics (Audiences)

Goal(s)

Revise search metrics and dashboard

Status

Note: July 2018

In progress

Note: August 21, 2018

In progress We'll be reaching out to the Research team for assistance and bring them together with our help from Audiences for this goal.

Note: September 20, 2018

This is currently

Stalled as the Research team is a bit busy with other priorities.

Note: September 25, 2018

In progress Analytics is working on the dashboard and documentation of search metrics

Outcome 1 / Output 4

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Improved support for multiple languages by researching and deploying new language analyzers where feasible on individual language wikis.

Goal(s)

Morphological library investigations and implementations (specific languages TBD)

Status

Note: July 2018

In progress

Note: August 21, 2018

In progress Esperanto plugin is

Partially done, Malay is

Done, both will need to be deployed into production with other small language bugs in the next couple of weeks.

Note: September 25, 2018

Done: Esperanto is

Done and has been deployed and re-indexed. Korean is still

In progress.

Outcome 1 / Output 5

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Specific media search filters for Wikidata/Wikibase and the relationships to the topics they represent will be better supported using structured data and other techniques.

Dependency: WMDE

Goal(s)

Lexeme search implementation: complete search implementation for all modes for Lemmas and Forms Done
Investigate applying machine-learning enabled ranking to Wikidata searches, start collecting click data for Wikidata completion searches and start developing machine-learning models for Wikidata search relevancy. In progress

Status

Note: July 2018

In progress

Note: August 21, 2018

In progress Lexeme search work is on-going, but running into a few small issues with the queries and how results are presented and how we want to do this in the future. Also working on extracting cirrus stuff out of Wikibase and getting the Lexemes into the Wikidata query service.

Note: September 11, 2018

In progress Lexeme search work is

Done; the models are still

In progress but most likely will be part of next quarter's work.

Outcome 2 / Output 1

Technical debt addressed and required maintenance completed for Search Platform components

Elasticsearch upgrades and server replacements

Dependency: SRE

Goal(s)

Continue to prepare for a major upgrade to Elasticsearch 6 In progress
Replace Elasticsearch servers which are at the end of their lease ((stalled))
Migrate Elasticsearch servers to RAID 0 Done

Status

Note: July 2018

In progress

Note: August 21, 2018

In progress Gehel has re-striped everything and migrated the Elasticsearch servers to Stretch. Lease expires on the servers next month; will also be working on migrating other (maps) servers to Stretch.

Note: September 20, 2018

Partially done prep work is still ongoing, but the data center switch is taking some time away from this work. The actual full upgrade will be part of Q2's work and will require a few weeks of stress testing of the upgrade; also working on the full migration sequence path/documentation/shard checks.

Note: September 25, 2018

Partially doneThere is still prep work to be done in Q2 for the ES6 upgrade, will not get to the actual upgrade until Q3, but currently working with local ES6 instances as prep work is being done. Progress on replacement Elasticsearch server procurement is stalled on the quotes and will be done in Q2.

Outcome 2 / Output 2

Technical debt addressed and required maintenance completed for Search Platform components

Higher capacity for WDQS to improve its ability to power features on-wiki for readers and the growing set of features for supporting structured data

Dependencies: SRE, WMDE

Goal(s)

Add storage to WDQS servers Done
Enable Kafka event consumption In progress
Separate the Wikidata Elasticsearch implementation into a separate extension N Stalled
Investigate Blazegraph support options and alternatives (Stretch goal) To do

Status

Note: July 2018

To do Waiting on discs to arrive.

Note: August 21, 2018

In progress Most of the discs have now arrived, need to take the servers offline and re-image them next as growing the cluster is difficult.

Note: September 20, 2018

Partially done Kafka event consumption is continuing, but separating the Wikidata implementation into a different extension work will be moved to Q2 for completion.

Q2 Goals

Outcome 1 / Output 1

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Incorporate Natural Language Processing (NLP) in the machine learning analysis pipeline for search

Dependency: Will need some short-term consulting help during implementation

Goal(s)

Find and hire a contractor to help with NLP work
Begin working on one internal NLP project (wrong keyboard detection, starting with Russian/English) (T138958)

Status

Note: October 29, 2018

In progress as interviews are on-going, we'll probably start the work in early Q3. Work is now ongoing with wrong keyboard detection.

Note: November 29, 2018

Wrong keyboard work is

In progress and NLP contractor work will still happen in Q3.

Note: December 13, 2018

Contractor has been identified and the contract is starting to be worked on, considering this to be

Done.

Outcome 1 / Output 2

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Better understanding of the effectiveness of our improvements to search and the performance of our tooling on the backend

Goal(s)

Improve autocomplete of Wikidata items
- Expand our machine learning to Wikidata and Commons (autocompletes for relevance, considering multilingual)
Prototype a feature that is based on collected data
- Continued work from Q1 with the collection of click logs for the autocomplete feature

Status

Note: October 29, 2018

This is actively being worked on and is fully

In progress.

Note: November 29, 2018

This is actively being worked on and is fully

In progress.

Note: December 13, 2018

Trey has this

In progress but will probably flow into early Q3 with the NLP contractor.

Outcome 1 / Output 3

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Improved support for multiple languages by researching and deploying new language analyzers where feasible on individual language wikis.

Goal(s)

Finish up the Korean morphological library analysis and deploy into production (carry-over work from Q1)
- This work is dependent on the upgrade to ElasticSearch 6 finishing
General language support (i.e, misc language-specific bugs)

Status

Note: October 29, 2018

The Korean goal is now

Done but won't be updated until the ElasticSearch 6.

Outcome 1 / Output 4

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Specific media search filters for Wikidata/Wikibase and the relationships to the topics they represent will be better supported using structured data and other techniques.

Dependency: WMDE

Goal(s)

[figuring out what SDC wants]
Search for licenses in Commons

Status

Note: October 29, 2018

Stalled We are waiting on a few things yet from the SDoC program - hard specs on what is expected from the Search Platform to develop features based off of presentations.

Note: November 29, 2018

Stalled We are waiting on a few things yet from the SDoC program - hard specs on what is expected from the Search Platform to develop features based off of presentations.

Note: December 13, 2018

This goal is still

Stalled as we await further instructions from SDoC program.

Outcome 2 / Output 1

Technical debt addressed and required maintenance completed for Search Platform components

Elasticsearch upgrades and server replacements

Dependency: SRE, WMDE

Goal(s)

Split the search clusters to increase stability Partially done
Continue replacing ElasticSearch servers (end of life maintenance) Done
Separate the Wikidata ElasticSearch implementation into a separate extension Partially done
Migrate ElasticSearch cluster restart scripts as cookbooks using Spicerack In progress
Stretch goal: Start working on the CloudElastic replicas (and perform a proof of concept with a few select wikis) To do

Status

Note: October 29, 2018

In progress Quotes from vendors were much higher than expected, Gehel, Erik, and SRE are going over requests and refining configurations to get more within our budget.

Note: November 29, 2018

This is still

In progress — purchases for the new servers had been made and servers are getting delivered / installed.

Note: December 13, 2018

Splitting the search clusters has been

Done and testing is still

In progress (should be done in the next week); replacing the older servers can be considered

Done; separating the Wikidata ES implementation is

Done but the new extension is still

In progress and will be completed in Q3. Migrating the ES cluster to Spicerack is still

In progress and will need testing.

The stretch goal has been pushed to Q3.

Outcome 2 / Output 2

Technical debt addressed and required maintenance completed for Search Platform components

Higher capacity for WDQS to improve its ability to power features on-wiki for readers and the growing set of features for supporting structured data

Dependencies: SRE, WMDE

Goal(s)

Performance and bug fixes for WDQS In progress
Service Level Objective (SLO) work for WDQS (T199228) In progress
Carryover from Q1: continue to investigate Blazegraph support options and alternatives Done

Status

Note: October 29, 2018

The SLO work is a bit

Stalled, but still somewhat

In progress at this point due to varying conversations, but lots of work has been ongoing with the fixes on WDQS. The team is still investigating long term usage of Blazegraph and what that means to the Foundation.

Note: November 29, 2018

This is now

In progress

Note: December 13, 2018

These goals are still

In progress; but the SLO work will flow into Q3. Blazegraph support options investigation is now

Done and we are determining next steps.

Q3 Goals

Outcome 1 / Output 1

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Incorporate Natural Language Processing (NLP) in the machine learning analysis pipeline for search

Dependency: Will need some short-term consulting help during implementation

Tech Goal(s)

C: Improve our own feature set

Goal(s)

Work with NLP contractor on spelling and did-you-mean result tuning (T212884) and (T212888)
Complete Russian/English wrong keyboard detection (T138958)

Status

Note: January 22, 2019

* The wrong keyboard detection work is

In progress; the NLP contract work is

Stalled for right now due to certain legal issues.

Note: February 25, 2019

* The wrong keyboard work has been

Postponed for now, as the NLP work has kicked off and is fully

In progress

Note: March 19, 2019

Work with Julia (contractor) is In progress but the work getting into production will most likely go into Q4.
Deployed wrong-keyboard language-ID models needed for future use. (T213931 / T216083)

Outcome 1 / Output 2

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Better understanding of the effectiveness of our improvements to search and the performance of our tooling on the backend

Tech Goal(s)

C: Improve our own feature set

Goal(s)

Deploy improved autocomplete for English Entities
Use machine learning to improve autocomplete of Wikidata items for three more languages

Status

Note: January 22, 2019

* The patch for this is

Done and going out with this week's train and the A/B tests are going to be starting soon.

Note: February 25, 2019

* Tested Spanish, English, and French - all good with the testing and is now deployed in production

Done

Outcome 1 / Output 3

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Improved support for multiple languages by researching and deploying new language analyzers where feasible on individual language wikis.

Tech Goal(s)

A: Foundation Goal: Reach

B: Serving our audiences

C: Improve our own feature set

D: Technical Debt

Goal(s)

Deploy Korean morphological library into production (after Elasticsearch is upgraded)
General language support (i.e, misc language-specific bugs)

Status

Note: January 22, 2019

* We're still working on upgrading ElasticSearch, so the deployment of Korean library is still

Stalled at this point. General lang support is

In progress and ongoing.

Note: February 25, 2019

* The Korean update is

Partially done and planned for next week to be rolled out as part of the ES6 upgrade. General lang support is

In progress and ongoing.

Note: March 19, 2019

* The Korean update work is

Partially done and will be completed when ES6 is

Done

* Improved analysis for Greek, Turkish, and Irish with language-specific lowercasing (T203117/T217602)

Outcome 2 / Output 1

Technical debt addressed and required maintenance completed for Search Platform components

Elasticsearch 6 upgrades and server replacements

Dependency: SRE, WMDE

Tech Goal(s)

C: Improve our own feature set

D: Technical Debt

Goal(s)

Upgrade ElasticSearch to v6
Complete separation of Wikidata ElasticSearch implementation into a separate extension
JVM and ElasticSearch upgrade Spicerack cookbooks
Stretch goal: Start working on the CloudElastic replicas (and perform a proof of concept with a few select wikis)

Status

Note: January 22, 2019

* The upgrade is still

In progress and troubleshooting issues with cluster states and building a whole new big bunch of shards to handle the increased load.

* The separation of Wikidata ElasticSearch implementation is still

In progress, but other work is also on-going right at the moment.

* JVM and ES upgrade is

In progress; we are doing first passes at the integration testing, and will have a first set of cookbooks soon.

* Stretch goal might be a bit closer than we initially thought, but once ES6 is in production, we can finish this up, maybe in mid February.

Note: February 25, 2019

* ES6 is

In progress planned to be deployed the week of March 4th

* Wikidata ElasticSearch implementation is still

In progress and we hope to get it mostly done by end of this quarter, but might go over into next quarter for the deployment.

* JVM and ElasticSearch upgrade is still

In progress - ES6 cookbooks are

Done and we only have a little bit to do with the rest of the WS6 upgrade. (T202885)

* Stretch goal is

In progress (T214921)

Note: March 19, 2019

Relforge update will be Done soon (this week).
Stretch goal (CloudElastica replica) is still In progress but will extend into Q4's work

Outcome 2 / Output 2

Technical debt addressed and required maintenance completed for Search Platform components

Higher capacity for WDQS to improve its ability to power features on-wiki for readers and the growing set of features for supporting structured data

Dependencies: SRE, WMDE

Tech Goal(s)

B: Serving our audiences

C: Improve our own feature set

D: Technical Debt

Goal(s)

Performance and bug fixes for WDQS
Service Level Objective (SLO) work for WDQS (T199228) In progress
Release official Blazegraph code package and start using Blazegraph CI
Roadmap planning for WDQS

Status

Note: January 22, 2019

Perf / bug fixing is on-going and In progress along with a new roadmap on how to figure out which ones come first.
SLO is In progress, we'll talk about it during All Hands week
We are having issues with Blazegraph right now, but we're working them, so In progress for the issues, but the official code package is To do

Note: February 25, 2019

Perf / bug fixing for WDQS is on-going and In progress
SLO work is in a waiting status for us, but we're keeping it In progress
Blazegraph's CI is still getting attention and is In progress (T216855)

Note: March 19, 2019

Perf / bug fixing for WDQS is on-going and In progress
SLO work for Blazegraph is a bit N Stalled right now as we're continuing a few conversations about public and internal facing WDQS up time and expectations of the service.
The release official Blazegraph code package is Done and we're working on a few config changes for using Blazegraph in CI, but can be considered Done at this point.
Road map planning for Blazegraph is Done as well

Q4 Goals

Outcome 1 / Output 1

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Incorporate Natural Language Processing (NLP) in the machine learning analysis pipeline for search

Dependency: Using short-term consulting help

Tech Goal(s)

C: Improve our own feature set

Goal(s)

Complete 'did you mean' (DYM) project with NLP contractor (T212884)
- Assess performance and investigate beginning integration work for deployment
Enable new Korean analysis, reindex clusters, and retrain the 'learning to rank' (LTR) model (T216738)

Status

Note: May 28, 2019

The 'did you mean' work has been in progress and the team is In progress reviewing the code patches (method 1 is Done, method 2 is In progress).
The work on the Korean analysis is Done and we are currently gathering data after the reindex. The retraining is still To do.

Note: June 25, 2019

DYM is pretty much Done at this point, as far as the work with our contractor. Method 2 works but we need to find another way to evaluate this method - still evaluating In progress and this will continue into next FY/quarter.
Assessing perf and integration work is In progress
Korean retraining is still To do and will move into next FY/quarter. We are wanting to setup an automated pipeline.

Outcome 1 / Output 3

Through incremental Search Platform component improvements, teams and developers can deliver more and better ways for readers and editors to search for content across languages.

Improved support for multiple languages by researching and deploying new language analyzers where feasible on individual language wikis.

Tech Goal(s)

A: Foundation Goal: Reach

B: Serving our audiences

C: Improve our own feature set

D: Technical Debt

Goal(s)

Search "Local Impact": making making bigger improvements for smaller (or underrepresented) communities T218613
Investigate and continue to revise search metrics and dashboards (T216055)

Status

Note: May 28, 2019

Completion Suggester improvements under T117217 are Done
Other improvements under epic T218613 are still In progress

Note: June 25, 2019

This work will continue into next FY/quarter.

Outcome 2 / Output 1

Technical debt addressed and required maintenance completed for Search Platform components

Elasticsearch 6 upgrades and server replacements

Dependency: SRE

Tech Goal(s)

C: Improve our own feature set

D: Technical Debt

Goal(s)

Adding ability for cloud environment to use ElasticSearch and perform a proof of concept with a few select wikis (T214921)

This work is often referred to as "Cloud Elastic Replicas".

Status

To do April 2019

There might be some work needed to do this month after the ES6 upgrade that should be finished in q3

Note: May 28, 2019

Upgrade to ES6 is Done and the CloudElastic work is In progress. Might need to do some more work with the deprecation loggers (minor cleanup still In progress on https://phabricator.wikimedia.org/T218994

Note: June 25, 2019

CloudElastic work is still In progress - we're waiting on the load balancer to be activated and then we can automate the updates.
The deprecation loggers work is Partially done and will wrap up early next quarter and continue until the next elastic upgrade.

Outcome 2 / Output 2

Technical debt addressed and required maintenance completed for Search Platform components

Higher capacity for WDQS to improve its ability to power features on-wiki for readers and the growing set of features for supporting structured data

Dependencies: SRE, WMDE

Tech Goal(s)

B: Serving our audiences

C: Improve our own feature set

D: Technical Debt

Goal(s)

Improve the performance and functionality of Blazegraph (the graph storage engine behind Wikidata Query Service).
- Increase Blazegraph support for SPARQL operations. Example: T200612
- Improve query performance of Blazegraph in areas of particular interest for WMF. Example: T152773

Status

Note: May 28, 2019

Improving Blazegraph is still In progress and we are working on the existing bugs with SPARQL operations. The query performance work is going well and ready for deployment with a full db reload Partially done, should be completely done in June.

Note: June 25, 2019

Blazegraph improvements for performance and functionality is ongoing and continuing ...lots of bug fixes are In progress, and we need to investigate how to do rebuilds while having different configurations. This work will continue into next FY.