misc fixes #7633
misc fixes #7633
Conversation
github-actions
bot
added
devops
| @@ -41,6 +41,8 @@ services: | |||
| - METADATA_SERVICE_AUTH_ENABLED=false | |||
| - JAVA_TOOL_OPTIONS=-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=*:5001 | |||
| - BOOTSTRAP_SYSTEM_UPDATE_WAIT_FOR_SYSTEM_UPDATE=false | |||
| - SEARCH_SERVICE_ENABLE_CACHE=false | |||
There was a problem hiding this comment.
was search caching causing the failure in the test?
There was a problem hiding this comment.
This is not used when running smoke tests. This is run when developing locally (enables debugger for example 2 lines above), not used when running quickstart by default, nor in smoke-tests. I think that cache should be disabled for local development by default because one of the first things I did was to search, then realized I was missing data, loaded data, and the results were cached. This is a quality of life for local development.
...c/main/java/com/linkedin/metadata/search/elasticsearch/query/request/SearchQueryBuilder.java
Show resolved
Hide resolved
| @@ -39,7 +40,12 @@ public class SearchQueryBuilderTest { | |||
| exactMatchConfiguration.setCaseSensitivityFactor(0.7f); | |||
| exactMatchConfiguration.setEnableStructured(true); | |||
|
|
|||
| PartialConfiguration partialConfiguration = new PartialConfiguration(); | |||
| partialConfiguration.setFactor(0.4f); | |||
| # Field weight annotations are typically calibrated for exact match, if partial match is possible on the field use these adjustments | ||
| partial: | ||
| urnFactor: ${ELASTICSEARCH_QUERY_PARTIAL_URN_FACTOR:0.7} # multiplier on Urn token match, a partial match on Urn > non-Urn is assumed | ||
| factor: ${ELASTICSEARCH_QUERY_PARTIAL_FACTOR:0.4} # multiplier on possible non-Urn token match |
There was a problem hiding this comment.
Wonder if we could set this on a per-field basis inside the Searchable annotation...
There was a problem hiding this comment.
Probably useful, but ultimately likely to be replaced by a proper search runtime configuration yaml. The model annotations are fine for index time decisions, we can detect and reindex as needed. However query-time/runtime configuration as model annotations is difficult to override at runtime unless you also are introducing custom models replacing the existing ones ....not even sure that would work. The query-time search configuration via annotations in my opinion are likely to be removed for a dedicated configuration file that is used at runtime. cc: @RyanHolstien @iprentic
There was a problem hiding this comment.
Yeah this makes sense to me. The model should host static (non-changing) configurations only.
david-leifker
requested review from
leifker,
RyanHolstien and
iprentic
and removed request for
leifker
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
fix slim trivy scan adjust urn partial further
Included: #7628
Checklist