Andrzej Wisłowski

Paul Anderson photo

Software Engineer
Allegro.pl

An experienced software engineer in the development and design of search engines and NoSql architectures. Last 3 years spent developing a search engine based on solr for allegro.pl site. Before he worked as a developer of elasticsearch solutions. Worked as a jvm developer for over 20 years. An author of company blog posts: https://allegro.tech/authors/andrzej.wislowski/

Andrzej Wisłowski is speaking at the following session/s

Battle for Fast Real-time Document Grouping using Solr Search Engine

Thursday | 3:20PM - 4:00PM |

At Allegro we use Solr as our main search engine. Due to our search engine traffic and index size, we need to develop custom optimizations. In offer listings, we were supposed to group together offers representing different variants of the same product so that they would be displayed as a single item. We started with configuring built in solr collapsing query parser, but it was not competitive enough. So we developed a custom filter and compared its implementations with sorted indexes and priority queue. After a few months we needed to choose the best offer in a group by one sort expression, then finally re-sort the results by the second sort expression. We tried a few approaches, starting with our custom collapser with a defined size priority queue, then grouping a long type field. Finally, we changed built-in Solr collapser plugin, which in our case allowed to decrease query time by 40%. Our improvement was merged into Solr 8.1.

Attendee Takeaway
Learn how to boost Solr performance and customize query handling to better fit our use case. I will show an example of how maintaining custom changes in the search code is possible to be handled at the plugins level.

Intended Audience
Learn how to boost Solr performance and customise query handling to better fit our use case. I will show an example how maintaining custom changes in the search code is possible to be handled at the plugins level.

Level:
All Levels