Less OOM in stress test

docker/test/stress/run.sh

@@ -53,6 +53,7 @@ function configure()
     local total_mem
     total_mem=$(awk '/MemTotal/ { print $(NF-1) }' /proc/meminfo) # KiB
     total_mem=$(( total_mem*1024 )) # bytes
+
     # Set maximum memory usage as half of total memory (less chance of OOM).
     #
     # But not via max_server_memory_usage but via max_memory_usage_for_user,
@@ -65,16 +66,17 @@ function configure()
     # max_server_memory_usage will be hard limit, and queries that should be
     # executed regardless memory limits will use max_memory_usage_for_user=0,
     # instead of relying on max_untracked_memory
-    local max_server_mem
-    max_server_mem=$((total_mem*75/100)) # 75%
-    echo "Setting max_server_memory_usage=$max_server_mem"
+
+    max_server_memory_usage_to_ram_ratio=0.5
+    echo "Setting max_server_memory_usage_to_ram_ratio to ${max_server_memory_usage_to_ram_ratio}"
     cat > /etc/clickhouse-server/config.d/max_server_memory_usage.xml <<EOL
 <clickhouse>
-    <max_server_memory_usage>${max_server_mem}</max_server_memory_usage>
+    <max_server_memory_usage_to_ram_ratio>${max_server_memory_usage_to_ram_ratio}</max_server_memory_usage_to_ram_ratio>
 </clickhouse>
 EOL
+
     local max_users_mem
-    max_users_mem=$((total_mem*50/100)) # 50%
+    max_users_mem=$((total_mem*30/100)) # 30%
     echo "Setting max_memory_usage_for_user=$max_users_mem"
     cat > /etc/clickhouse-server/users.d/max_memory_usage_for_user.xml <<EOL
 <clickhouse>
@@ -97,6 +99,13 @@ EOL
     -->
     <core_path>$PWD</core_path>
 </clickhouse>
+EOL
+
+    # Let OOM killer terminate other processes before clickhouse-server:
+    cat > /etc/clickhouse-server/config.d/oom_score.xml <<EOL
+<clickhouse>
+    <oom_score>-1000</oom_score>
+</clickhouse>
 EOL
 
     # Analyzer is not yet ready for testing

programs/server/Server.cpp

@@ -420,6 +420,31 @@ void Server::createServer(
     }
 }
 
+
+namespace
+{
+
+void setOOMScore(int value, Poco::Logger * log)
+{
+    try
+    {
+        std::string value_string = std::to_string(value);
+        DB::WriteBufferFromFile buf("/proc/self/oom_score_adj");
+        buf.write(value_string.c_str(), value_string.size());
+        buf.next();
+        buf.close();
+    }
+    catch (const Poco::Exception & e)
+    {
+        LOG_WARNING(log, "Failed to adjust OOM score: '{}'.", e.displayText());
+        return;
+    }
+    LOG_INFO(log, "Set OOM score adjustment to {}", value);
+}
+
+}
+
+
 void Server::uninitialize()
 {
     logger().information("shutting down");
@@ -881,6 +906,21 @@ try
             }
         }
     }
+
+    int default_oom_score = 0;
+
+#if !defined(NDEBUG)
+    /// In debug version on Linux, increase oom score so that clickhouse is killed
+    /// first, instead of some service. Use a carefully chosen random score of 555:
+    /// the maximum is 1000, and chromium uses 300 for its tab processes. Ignore
+    /// whatever errors that occur, because it's just a debugging aid and we don't
+    /// care if it breaks.
+    default_oom_score = 555;
+#endif
+
+    int oom_score = config().getInt("oom_score", default_oom_score);
+    if (oom_score)
+        setOOMScore(oom_score, log);
 #endif
 
     global_context->setRemoteHostFilter(config());

programs/server/config.xml

@@ -1464,4 +1464,8 @@
          I don't recommend to change this setting.
     <show_addresses_in_stack_traces>false</show_addresses_in_stack_traces>
     -->
+
+    <!-- On Linux systems this can control the behavior of OOM killer.
+    <oom_score>-1000</oom_score>
+    -->
 </clickhouse>

src/Daemon/BaseDaemon.cpp

@@ -602,34 +602,6 @@ void BaseDaemon::closeFDs()
     }
 }
 
-namespace
-{
-/// In debug version on Linux, increase oom score so that clickhouse is killed
-/// first, instead of some service. Use a carefully chosen random score of 555:
-/// the maximum is 1000, and chromium uses 300 for its tab processes. Ignore
-/// whatever errors that occur, because it's just a debugging aid and we don't
-/// care if it breaks.
-#if defined(OS_LINUX) && !defined(NDEBUG)
-void debugIncreaseOOMScore()
-{
-    const std::string new_score = "555";
-    try
-    {
-        DB::WriteBufferFromFile buf("/proc/self/oom_score_adj");
-        buf.write(new_score.c_str(), new_score.size());
-        buf.close();
-    }
-    catch (const Poco::Exception & e)
-    {
-        LOG_WARNING(&Poco::Logger::root(), "Failed to adjust OOM score: '{}'.", e.displayText());
-        return;
-    }
-    LOG_INFO(&Poco::Logger::root(), "Set OOM score adjustment to {}", new_score);
-}
-#else
-void debugIncreaseOOMScore() {}
-#endif
-}
 
 void BaseDaemon::initialize(Application & self)
 {
@@ -796,7 +768,6 @@ void BaseDaemon::initialize(Application & self)
 
     initializeTerminationAndSignalProcessing();
     logRevision();
-    debugIncreaseOOMScore();
 
     for (const auto & key : DB::getMultipleKeysFromConfig(config(), "", "graphite"))
     {