From f70dc8acb242a17a844843d022c77be99825da59 Mon Sep 17 00:00:00 2001
From: antirez <antirez@gmail.com>
Date: Wed, 26 Mar 2025 23:47:47 +0100
Subject: [PATCH] Clarify VRANDMEMBER tradeoff.

---
 vset.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/vset.c b/vset.c
index f53e20da7..33ffb8b32 100644
--- a/vset.c
+++ b/vset.c
@@ -1501,7 +1501,14 @@ int VRANDMEMBER_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv, int
             RedisModule_FreeDict(ctx, returned);
         } else {
             /* For large samples, get a random starting node and walk
-             * the list. */
+             * the list.
+             *
+             * IMPORTANT: doing so does not really generate random
+             * elements: it's just a linear scan, but we have no choices.
+             * If we generate too many random elements, more and more would
+             * fail the check of being novel (not yet collected in the set
+             * to return) if the % of elements to emit is too large, we would
+             * spend too much CPU. */
             hnswNode *start_node = hnsw_random_node(vset->hnsw, 0);
             hnswNode *current = start_node;