Connection reuse is trading faster retrieval for unstable tail latency.
The service-side correction protects capacity while the rollback protects release simplicity.
Who you’d be doing this for
“My adviser asked where the sources came from, and the answer kept hanging before they loaded.”
Antti Makinen · Graduate student
Uses source-cited answers to compare current research and prepare seminar notes.
What is at stake
p99 retrieval latency has reached 910 ms and partial answers are at 2.1%. You must weigh a targeted connection-reuse change against the safer rollback path in one deployment window.
Why it isn’t already fixed
Every obvious fix costs something else. That’s the part you’d have to decide.
- tail latency vs release simplicity
- retry containment vs cluster headroom
- fast recovery vs service-level learning
Why Perplexity
Perplexity Search depends on low-latency retrieval to return source-cited answers while a query is still useful.
Written with these in mind
Not your kind of problem? 31 more at Perplexity, or browse every organization.
This is the setup. The work is inside.
Running it puts you in the room: the full situation and its constraints, stakeholders who push back in their own words, and the decisions that are yours to make. What you produce becomes a Day One Plan — work you can show someone instead of describing.