From 1d7035c80f791a3f70970a796ed646bfb74a9e91 Mon Sep 17 00:00:00 2001
From: Stepan Bagritsevich <stefan@dragonflydb.io>
Date: Mon, 18 Nov 2024 11:22:12 +0100
Subject: [PATCH] fix(eviction): Tune eviction threshold in cache mode

fixes dragonflydb#4011

Signed-off-by: Stepan Bagritsevich <stefan@dragonflydb.io>
---
 src/server/engine_shard.cc     |  7 +++++-
 tests/dragonfly/memory_test.py | 41 ++++++++++++++++++++++++++++++++++
 2 files changed, 47 insertions(+), 1 deletion(-)

diff --git a/src/server/engine_shard.cc b/src/server/engine_shard.cc
index 29c7957f2eb5..70a971d3b0a6 100644
--- a/src/server/engine_shard.cc
+++ b/src/server/engine_shard.cc
@@ -704,6 +704,7 @@ void EngineShard::RetireExpiredAndEvict() {
   { std::unique_lock lk(db_slice.GetSerializationMutex()); }
   constexpr double kTtlDeleteLimit = 200;
   constexpr double kRedLimitFactor = 0.1;
+  constexpr double kDenyOomThresholdMargin = 0.05;
 
   uint32_t traversed = GetMovingSum6(TTL_TRAVERSE);
   uint32_t deleted = GetMovingSum6(TTL_DELETE);
@@ -717,7 +718,11 @@ void EngineShard::RetireExpiredAndEvict() {
     ttl_delete_target = kTtlDeleteLimit * double(deleted) / (double(traversed) + 10);
   }
 
-  ssize_t eviction_redline = size_t(max_memory_limit * kRedLimitFactor) / shard_set->size();
+  const double deny_oom_threshold = 1.0 - std::max(ServerState::tlocal()->oom_deny_ratio, 0.0);
+  const double eviction_threshold =
+      std::max(deny_oom_threshold + kDenyOomThresholdMargin, kRedLimitFactor);
+
+  ssize_t eviction_redline = size_t(max_memory_limit * eviction_threshold) / shard_set->size();
 
   DbContext db_cntx;
   db_cntx.time_now_ms = GetCurrentTimeMs();
diff --git a/tests/dragonfly/memory_test.py b/tests/dragonfly/memory_test.py
index 843e7c606dc9..76889b953498 100644
--- a/tests/dragonfly/memory_test.py
+++ b/tests/dragonfly/memory_test.py
@@ -107,3 +107,44 @@ async def test_rss_oom_ratio(df_factory: DflyInstanceFactory, admin_port):
     # new client create shoud not fail after memory usage decrease
     client = df_server.client()
     await client.execute_command("set x y")
+
+
+@pytest.mark.asyncio
+@dfly_args(
+    {
+        "proactor_threads": 1,
+        "cache_mode": "true",
+        "maxmemory": "256mb",
+        "oom_deny_ratio": 0.3,
+        "enable_heartbeat_eviction": "true",
+    }
+)
+async def test_cache_eviction_with_deny_oom(
+    df_server: DflyInstance,
+):
+    """
+    Test to verify that cache eviction is triggered before reaching the deny OOM threshold.
+    """
+
+    client = df_server.client()
+
+    max_memory = 256 * 1024 * 1024  # 256 MB in bytes
+    oom_deny_ratio = 0.3
+    deny_oom_threshold = int(max_memory * oom_deny_ratio)  # 77 MB
+    key_size = 1048576  # 1 MB per key
+
+    # Populate the cache up to the deny OOM threshold (77 MB)
+    await client.execute_command("DEBUG POPULATE", deny_oom_threshold // key_size, "size", key_size)
+
+    await asyncio.sleep(1)  # Wait for RSS heartbeat update in Dragonfly
+
+    info_before_eviction = await client.info("memory")
+    assert info_before_eviction["evicted_keys"] == 0, "No eviction should have occurred yet."
+
+    # Populate the cache further to trigger eviction
+    await client.execute_command("DEBUG POPULATE 100 size", key_size)
+
+    await asyncio.sleep(1)  # Wait for RSS heartbeat update in Dragonfly
+
+    info_after_eviction = await client.info("memory")
+    assert info_after_eviction["evicted_keys"] > 0, "Eviction should have occurred."