Re:Show HN: Full text search on 630M US court cases
by tmikaeld on Saturday February 19, 2022 @04:45PM
(#30400825)
How much ram does that use up? What’s the latency? Is it sharded? Is it a cluster? So many questions
[ Parent
| on HN ]
Re:Show HN: Full text search on 630M US court cases
by richardbarosky on Saturday February 19, 2022 @04:57PM
(#30400912)
There are 2 search boxes going. One for storing the search index without source and another which stores the source, which is only used for highlighting. Searches usually take under 200ms and SRP and individual pages usually take less than 20ms. The 2 ES nodes are not formally part of a single cluster due to the index storage difference. Another box uses a traditional LAMP setup. Feel free to send a message on reddit if interested in more detail.
Re:Show HN: Full text search on 630M US court cases
by ramoz on Saturday February 19, 2022 @08:11PM
(#30402228)
How large is the index?
How do you manage that between RAM or SSD?
Re:Show HN: Full text search on 630M US court cases
by richardbarosky on Saturday February 19, 2022 @08:27PM
(#30402325)
Search - Index is ~373GB. AMD Epyc 7371 - 16c/32t - 3.1 GHz/3.8 GHz. 512 GB ECC 2400 MHz. 2×1.92 TB SSD NVMe
Highlight - Index is ~620GB. Xeon-D 2141I - 8c/16t - 2.2 GHz/3 GHz. 64 GB ECC 2133 MHz. 2×1.92 TB SSD NVMe
Search and highlighting handled async from queue.