Backport #87610 to 25.9: exclude userspace cache for memory overload warning

robot-clickhouse · robot-clickhouse · commit 69bcfe7386d7 · 2025-09-25T16:12:27.000Z
diff --git a/src/Common/AsynchronousMetrics.cpp b/src/Common/AsynchronousMetrics.cpp
@@ -921,6 +921,15 @@ void AsynchronousMetrics::processWarningForMemoryOverload(const AsynchronousMetr
     double memory_resident{};
     double memory_total{};
 
+    /// Compute the user space page cache size (in bytes) and subtract from total memory used if possible
+    double user_space_cache_bytes{};
+    if (const auto * rss = getAsynchronousMetricValue(new_values, "MemoryResident"),
+        *rss_wo_pc = getAsynchronousMetricValue(new_values, "MemoryResidentWithoutPageCache");
+        rss && rss_wo_pc && rss->value >= rss_wo_pc->value)
+    {
+        user_space_cache_bytes = rss->value - rss_wo_pc->value;
+    }
+
     /// use cgroup memory metrics if available and > 0, otherwise fallback to OS memory metrics
     if (const auto *cgroup_memory_total = getAsynchronousMetricValue(new_values, "CGroupMemoryTotal"),
         *cgroup_memory_used = getAsynchronousMetricValue(new_values, "CGroupMemoryUsed");
@@ -929,7 +938,7 @@ void AsynchronousMetrics::processWarningForMemoryOverload(const AsynchronousMetr
         memory_resident = cgroup_memory_used->value;
         memory_total = cgroup_memory_total->value;
     }
-    else if (const auto * os_memory_used = getAsynchronousMetricValue(new_values, "OSMemoryResident"),
+    else if (const auto * os_memory_used = getAsynchronousMetricValue(new_values, "MemoryResident"),
              *os_memory_total = getAsynchronousMetricValue(new_values, "OSMemoryTotal");
              os_memory_used && os_memory_total && (os_memory_total->value > 0.0 && os_memory_used->value > 0.0))
     {
@@ -942,15 +951,20 @@ void AsynchronousMetrics::processWarningForMemoryOverload(const AsynchronousMetr
         return;
     }
 
+    /// Exclude user space page cache from memory resident
+    if (user_space_cache_bytes > 0.0)
+        memory_resident = std::max(0.0, memory_resident - user_space_cache_bytes);
+
     const double ratio = memory_resident / memory_total;
+    const double clamped_ratio = std::clamp(ratio, 0.0, 1.0);
 
     const auto & cfg = context->getConfigRef();
     const double mem_warn_ratio = cfg.getDouble("resource_overload_warnings.memory_overload_warn_ratio", 0.9);
     const double mem_clear_ratio = cfg.getDouble("resource_overload_warnings.memory_overload_clear_ratio", 0.85);
     const UInt64 min_duration = cfg.getUInt64("resource_overload_warnings.memory_overload_duration_seconds", 600);
 
     const auto now = Clock::now();
-    const int usage_percent = static_cast<int>(std::lround(ratio * 100.0));
+    const int usage_percent = static_cast<int>(std::lround(clamped_ratio * 100.0));
 
     const auto warning_message = PreformattedMessage::create(
         "High ClickHouse memory usage: {} of {} used ({}%) for at least {} second(s)",
