I've been evaluating You.com's code search for a few weeks as part of a workflow efficiency audit. The latency has become a significant blocker.
I'm consistently observing 15-20 second response times for simple, targeted queries. This is against a local baseline of <2s for a similar scope via grep/ripgrep on a modest dataset. Example test case:
- **Query:** `"def merge_tables" language:python`
- **Dataset context:** Public repos, under 1MB total codebase.
- **You.com result time:** 17s (observed 3 times, 2pm EST)
- **Local ripgrep:** 0.8s
The delay appears non-linear. Adding a second filter (e.g., `"dbt_utils"`) sometimes doubles the wait. This makes iterative exploration impractical.
Has anyone else conducted systematic latency tests? I'm particularly interested if the delays are query-specific or a general throughput issue. My initial hypothesis is backend queueing or overly broad index scanning.
EXPLAIN ANALYZE
Interesting test. Have you tried varying the time of day? I wonder if it's less about the specific query and more about concurrent users. I've gotten much faster results late at night, but my searches are simpler, not timed.
> The delay appears non-linear.
That's the weird part. I'd expect it to be slow but consistent. If adding a filter doubles the time, maybe it's hitting separate systems or caches weirdly. Could it be network geography? Like, are you far from their main servers?
Do you know if these delays are the same for all code types, or just Python?
That baseline comparison with ripgrep is the entire story. Why are we routing a simple static code search through a dozen cloud microservices when a regex tool on a single machine solves it in under a second?
Your hypothesis about backend queueing is probably right, but it's a symptom of overengineering. They've built a distributed search system for a problem that often doesn't need one. The non-linear delay screams "distributed join" or "fan-out request" where adding a filter spins up another subsystem.
I've seen this same pattern with other hosted code analysis tools. The latency isn't about your query complexity; it's about the overhead of their architecture. You're waiting in line for containers to spin up.
null