From 7a9382b115374ee8aa6e5d07a6b160eff4ebf4a2 Mon Sep 17 00:00:00 2001
From: Kerollmops <clement@meilisearch.com>
Date: Thu, 30 Jan 2025 11:29:41 +0100
Subject: [PATCH] Better document the rayon limitation condition

---
 crates/milli/src/vector/ollama.rs | 4 ++++
 crates/milli/src/vector/openai.rs | 4 ++++
 crates/milli/src/vector/rest.rs   | 4 ++++
 3 files changed, 12 insertions(+)

diff --git a/crates/milli/src/vector/ollama.rs b/crates/milli/src/vector/ollama.rs
index ef5cfd937..d2a80d6b5 100644
--- a/crates/milli/src/vector/ollama.rs
+++ b/crates/milli/src/vector/ollama.rs
@@ -118,6 +118,8 @@ impl Embedder {
         text_chunks: Vec<Vec<String>>,
         threads: &ThreadPoolNoAbort,
     ) -> Result<Vec<Vec<Embedding>>, EmbedError> {
+        // This condition helps reduce the number of active rayon jobs
+        // so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
         if threads.active_operations() >= REQUEST_PARALLELISM {
             text_chunks.into_iter().map(move |chunk| self.embed(&chunk, None)).collect()
         } else {
@@ -137,6 +139,8 @@ impl Embedder {
         texts: &[&str],
         threads: &ThreadPoolNoAbort,
     ) -> Result<Vec<Vec<f32>>, EmbedError> {
+        // This condition helps reduce the number of active rayon jobs
+        // so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
         if threads.active_operations() >= REQUEST_PARALLELISM {
             let embeddings: Result<Vec<Vec<Embedding>>, _> = texts
                 .chunks(self.prompt_count_in_chunk_hint())
diff --git a/crates/milli/src/vector/openai.rs b/crates/milli/src/vector/openai.rs
index afb48bdcd..c7aec5d93 100644
--- a/crates/milli/src/vector/openai.rs
+++ b/crates/milli/src/vector/openai.rs
@@ -255,6 +255,8 @@ impl Embedder {
         text_chunks: Vec<Vec<String>>,
         threads: &ThreadPoolNoAbort,
     ) -> Result<Vec<Vec<Embedding>>, EmbedError> {
+        // This condition helps reduce the number of active rayon jobs
+        // so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
         if threads.active_operations() >= REQUEST_PARALLELISM {
             text_chunks.into_iter().map(move |chunk| self.embed(&chunk, None)).collect()
         } else {
@@ -274,6 +276,8 @@ impl Embedder {
         texts: &[&str],
         threads: &ThreadPoolNoAbort,
     ) -> Result<Vec<Vec<f32>>, EmbedError> {
+        // This condition helps reduce the number of active rayon jobs
+        // so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
         if threads.active_operations() >= REQUEST_PARALLELISM {
             let embeddings: Result<Vec<Vec<Embedding>>, _> = texts
                 .chunks(self.prompt_count_in_chunk_hint())
diff --git a/crates/milli/src/vector/rest.rs b/crates/milli/src/vector/rest.rs
index 49be155c1..58d805aaf 100644
--- a/crates/milli/src/vector/rest.rs
+++ b/crates/milli/src/vector/rest.rs
@@ -188,6 +188,8 @@ impl Embedder {
         text_chunks: Vec<Vec<String>>,
         threads: &ThreadPoolNoAbort,
     ) -> Result<Vec<Vec<Embedding>>, EmbedError> {
+        // This condition helps reduce the number of active rayon jobs
+        // so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
         if threads.active_operations() >= REQUEST_PARALLELISM {
             text_chunks.into_iter().map(move |chunk| self.embed(chunk, None)).collect()
         } else {
@@ -207,6 +209,8 @@ impl Embedder {
         texts: &[&str],
         threads: &ThreadPoolNoAbort,
     ) -> Result<Vec<Embedding>, EmbedError> {
+        // This condition helps reduce the number of active rayon jobs
+        // so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
         if threads.active_operations() >= REQUEST_PARALLELISM {
             let embeddings: Result<Vec<Vec<Embedding>>, _> = texts
                 .chunks(self.prompt_count_in_chunk_hint())