thread 'tokio-runtime-worker' panicked at src/results.rs:252:31: index out of bounds: the len is 1 but the index is 1

[paguilomanas]

I am trying to run the benchmark in kind throughput setting a number of max vus and many of the times I try to do so the results are gathered but then they are not written in the results json because I get this out of bounds error.

[markcholland]

I get the same panic. New to rust, but doesn't fn:quantile_duration() in results.rs compare i to the length of the array but then access the array using i+1?

thread 'tokio-runtime-worker' (128041) panicked at src/results.rs:252:31:
index out of bounds: the len is 1 but the index is 1
stack backtrace:
0: __rustc::rust_begin_unwind
1: core::panicking::panic_fmt
2: core::panicking::panic_bounds_check
3: inference_benchmarker::results::BenchmarkResults::quantile_duration
4: inference_benchmarker::results::BenchmarkResults::time_to_first_token_percentile
5: inference_benchmarker::writers::BenchmarkReportWriter::try_new
6: inference_benchmarker::run::{{closure}}
7: inference_benchmarker::main::{{closure}}::{{closure}}
8: tokio::runtime::task::core::Core<T,S>::poll
9: tokio::runtime::task::harness::Harness<T,S>::poll
10: tokio::runtime::scheduler::multi_thread::worker::Context::run_task
11: tokio::runtime::context::scoped::Scoped::set
12: tokio::runtime::context::runtime::enter_runtime
2025-11-05T12:50:25.476588177+00:00tINFOrStarting benchmark
2025-11-05T12:50:22.087519671+00:00rINFOSDownloading dataset
15: tokio::run1i09:9ta932400rness::HarneBenchmark warmup ended
16: tokio::run5i0e:872483897::pool::InneBenchm rk throughput ended

[markcholland]

I was able to get past the crash with this diff. I'm not 100% sure I was true to the intention of the quantile here. I believe this corner case only occurs when the length of the vector is 1. In that case, what did the designer of the code intend for, say, the 10th percentile? I've chosen to interpret it as "if the length of the vector is 1, every percentile is equal to the sole value in the vector".

diff --git a/src/results.rs b/src/results.rs
index da4f475..06c69b2 100644
--- a/src/results.rs
+++ b/src/results.rs
@@ -247,10 +247,14 @@ impl BenchmarkResults {
             let delta = (data.len() - 1) as f64 * quantile - i;
             if i as usize >= data.len() {
                 return Err(anyhow::anyhow!(NoResponses));
-            }
-            let quantile = (1. - delta) * data[i as usize].as_secs_f64()
-                + delta * data[i as usize + 1].as_secs_f64();
-            Ok(quantile)
+            } else if i as usize + 1 == data.len() {
+               let quantile = data[ i as usize ].as_secs_f64();
+               Ok(quantile)
+            } else {
+               let quantile = (1. - delta) * data[i as usize].as_secs_f64()
+                    + delta * data[i as usize + 1].as_secs_f64();
+               Ok(quantile)
+           }
         } else {
             Err(anyhow::anyhow!(NoResponses))
         }

[Hugoch]

Same panic as #36 — fixed in #45. The guard in quantile_duration didn't cover the data[i+1] read when a percentile bucket ends up with a single sample.