diff options
author | Jon Bratseth <bratseth@verizonmedia.com> | 2019-02-18 15:19:46 +0100 |
---|---|---|
committer | Jon Bratseth <bratseth@verizonmedia.com> | 2019-02-18 15:19:46 +0100 |
commit | fbaf43ef6e0f1be149ad1a7dce11a541d485fb2c (patch) | |
tree | 49a589012bd76f6f8dd8bbbb3ddc3cd4ff6b16ea | |
parent | a2eb3041fbcdc53be8f04c190faccad62f20077b (diff) |
Add code pointers
-rw-r--r-- | TODO.md | 62 |
1 files changed, 49 insertions, 13 deletions
@@ -2,7 +2,8 @@ # List of possible future enhancements and features This lists some possible improvements to Vespa which have been considered or requested, can be developed relatively -independently of other work and are not yet under development. +independently of other work, and are not yet under development. For more information on the code structure in Vespa, see +[Code-map.md](Code-map.md). ## Query tracing including content nodes @@ -16,16 +17,9 @@ the container, not in the content nodes. This is to implement similar tracing ca integrating trace information from each content node into the container level trace. This would make it easier to understand the execution and performance consequences of various query expressions. -## Change search protocol from fnet to RPC - -**Effort:** Low<br/> -**Difficulty:** Low<br/> -**Skills:** Java, C++, networking +**Code pointers:** +- [Container-level tracing](https://github.com/vespa-engine/vespa/blob/master/processing/src/main/java/com/yahoo/processing/execution/Execution.java#L245=) -Currently, search requests happens over a very old custom protocol called "fnet". While this is efficient, it is hard to extend. -We want to replace it by RPC calls. -An RPC alternative is already implemented for summary fetch requests, but not for search requests. -The largest part of this work is to encode the Query object as a Slime structure in Java and decode that structure in C++. ## Support query profiles for document processors @@ -38,6 +32,11 @@ bundles of parameters accessible to Searchers processing queries. Writes go thro Document Processors, but have no equivalent support for parametrization. This is to allow configuration of document processor profiles by reusing the query profile support also for document processors. +**Code pointers:** +- [Query profiles](https://github.com/vespa-engine/vespa/blob/master/container-search/src/main/java/com/yahoo/search/query/profile/QueryProfile.java) +- [Document processors](https://github.com/vespa-engine/vespa/blob/master/docproc/src/main/java/com/yahoo/docproc/DocumentProcessor.java) + + ## Background reindexing **Effort:** Medium<br/> @@ -48,10 +47,32 @@ Some times there is a need to reindex existing data to refresh the set of tokens definition changes impacts the tokens produced, and changing versions of linguistics libraries also cause token changes. As content clusters store the raw data of documents it should be possible to reindex locally inside clusters in the background. However, today this is not supported and content need to be rewritten from the outside to refresh tokens, -which is inconvenient and suboptimal. This is to support (scheduled or triggered) backgroun reindexing from local data. +which is inconvenient and suboptimal. This is to support (scheduled or triggered) background reindexing from local data. This can be achieved by configuring a message bus route which feeds content from a cluster back to itself through the indexing container cluster and triggering a visiting job using this route. +**Code pointers:** +- Document API which can be used to receive a dumpt of documents: [DocumentAccess](https://github.com/vespa-engine/vespa/blob/master/documentapi/src/main/java/com/yahoo/documentapi/DocumentAccess.java) + + +## Change search protocol from fnet to RPC + +**Effort:** Medium<br/> +**Difficulty:** Low<br/> +**Skills:** Java, C++, networking + +Currently, search requests happens over a very old custom protocol called "fnet". While this is efficient, it is hard to extend. +We want to replace it by RPC calls. +An RPC alternative is already implemented for summary fetch requests, but not for search requests. +The largest part of this work is to encode the Query object as a Slime structure in Java and decode that structure in C++. + +**Code pointers:** +- FS4 protocol search invokers (to be replaced by RPC): [FS4SearchInvoker](https://github.com/vespa-engine/vespa/blob/master/container-search/src/main/java/com/yahoo/prelude/fastsearch/FS4SearchInvoker.java), [FS4FillInvoker](https://github.com/vespa-engine/vespa/blob/master/container-search/src/main/java/com/yahoo/prelude/fastsearch/FS4FillInvoker.java) +- Current Query encoding (to be replaced by Slime): [QueryPacket](https://github.com/vespa-engine/vespa/blob/master/container-search/src/main/java/com/yahoo/fs4/QueryPacket.java) +- Slime: [Java](https://github.com/vespa-engine/vespa/blob/master/vespajlib/src/main/java/com/yahoo/slime/Slime.java), [C++](https://github.com/vespa-engine/vespa/blob/master/vespalib/src/vespa/vespalib/data/slime/slime.h) +- [C++ query (to be constructed from Slime)](https://github.com/vespa-engine/vespa/blob/master/searchlib/src/vespa/searchlib/query/query.h) + + ## Java implementation of the content layer for testing **Effort:** Medium<br/> @@ -65,6 +86,10 @@ functionality would enable developers to do more testing locally within their ID performance is not a concern and some components, such as ranking expressions and features are already available as libraries (see the searchlib module). +**Code pointers:** +- Content cluster mock in Java (currently empy): [ContentCluster](https://github.com/vespa-engine/vespa/blob/master/application/src/main/java/com/yahoo/application/content/ContentCluster.java) +- The model of a search definition this must consume config from: [Search](https://github.com/vespa-engine/vespa/blob/master/config-model/src/main/java/com/yahoo/searchdefinition/Search.java) + ## Update where **Effort:** Medium<br/> @@ -74,15 +99,22 @@ libraries (see the searchlib module). Support "update where" operations which changes/removes all documents matching some document selection expression. This entails adding a new document API operation and probably supporting continuations similar to visiting. +- Document operations: [Java](https://github.com/vespa-engine/vespa/tree/master/document/src/main/java/com/yahoo/document/update), [C++](https://github.com/vespa-engine/vespa/tree/master/document/src/vespa/document/update) + + ## Indexed search in maps **Effort:** Medium<br/> **Difficulty:** Medium<br/> -**Skills:** C++, Java, multithreading, performance, indexing, data structures +**Skills:** C++, multithreading, performance, indexing, data structures Vespa supports maps and and making them searchable in memory by declaring as an attribute. However, maps cannot be indexed as text-search disk indexes. +**Code pointers:** +- [Current text indexes](https://github.com/vespa-engine/vespa/tree/master/searchlib/src/vespa/searchlib/index) + + ## Global writes **Effort:** Large<br/> @@ -104,6 +136,9 @@ rates on loss of connectivity to one (any) data center, re-establish global data recovered and support some degree of tradeoff between consistency and operation latency (although the exact modes to be supported is part of the design and analysis needed). +**Code pointers:** +- [Document API](https://github.com/vespa-engine/vespa/tree/master/documentapi/src/main/java/com/yahoo/documentapi) + ## Global dynamic tensors **Effort:** High @@ -121,4 +156,5 @@ perfect consistency. This in turn must push updates to all content nodes in a be budget, such that query execution and document write traffic is prioritized over ensuring perfect consistency of global model updates. - +**Code pointers:** +- Tensor modify operation (for document tensors): [Java](https://github.com/vespa-engine/vespa/blob/master/document/src/main/java/com/yahoo/document/update/TensorModifyUpdate.java), [C++](https://github.com/vespa-engine/vespa/blob/master/document/src/vespa/document/update/tensor_modify_update.h) |