Merge pull request #20915 from vespa-engine/havardpe/proton-cpu-util-metrics

added cpu util metrics

4 files changed, 36 insertions, 1 deletions

diff --git a/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.cpp b/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.cpp
index 9bc20f95d13..9cf0339986c 100644
--- a/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.cpp
+++ b/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.cpp
@@ -4,6 +4,18 @@
 
 namespace proton {
 
+ResourceUsageMetrics::CpuUtilMetrics::CpuUtilMetrics(metrics::MetricSet *parent)
+  : MetricSet("cpu_util", {}, "Unnormalized cpu utilization for various categories", parent),
+    setup("setup", {}, "cpu used by system init and (re-)configuration", this),
+    read("read", {}, "cpu used by reading data from the system", this),
+    write("write", {}, "cpu used by writing data to the system", this),
+    compact("compact", {}, "cpu used by internal data re-structuring", this),
+    other("other", {}, "cpu used by work not classified as a specific category", this)
+{
+}
+
+ResourceUsageMetrics::CpuUtilMetrics::~CpuUtilMetrics() = default;
+
 ResourceUsageMetrics::ResourceUsageMetrics(metrics::MetricSet *parent)
     : MetricSet("resource_usage", {}, "Usage metrics for various resources in this search engine", parent),
       disk("disk", {}, "The relative amount of disk space used on this machine (value in the range [0, 1])", this),
@@ -15,7 +27,8 @@ ResourceUsageMetrics::ResourceUsageMetrics(metrics::MetricSet *parent)
       memoryMappings("memory_mappings", {}, "The number of mapped memory areas", this),
       openFileDescriptors("open_file_descriptors", {}, "The number of open files", this),
       feedingBlocked("feeding_blocked", {}, "Whether feeding is blocked due to resource limits being reached (value is either 0 or 1)", this),
-      mallocArena("malloc_arena", {}, "Size of malloc arena", this)
+      mallocArena("malloc_arena", {}, "Size of malloc arena", this),
+      cpu_util(this)
 {
 }
 
diff --git a/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.h b/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.h
index 774bb645c84..35100084cf7 100644
--- a/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.h
+++ b/searchcore/src/vespa/searchcore/proton/metrics/resource_usage_metrics.h
@@ -12,6 +12,17 @@ namespace proton {
  */
 struct ResourceUsageMetrics : metrics::MetricSet
 {
+    struct CpuUtilMetrics : metrics::MetricSet {
+        metrics::DoubleValueMetric setup;
+        metrics::DoubleValueMetric read;
+        metrics::DoubleValueMetric write;
+        metrics::DoubleValueMetric compact;
+        metrics::DoubleValueMetric other;
+
+        CpuUtilMetrics(metrics::MetricSet *parent);
+        ~CpuUtilMetrics();
+    };
+
     metrics::DoubleValueMetric disk;
     metrics::DoubleValueMetric diskUtilization;
     metrics::DoubleValueMetric memory;
@@ -22,6 +33,7 @@ struct ResourceUsageMetrics : metrics::MetricSet
     metrics::LongValueMetric openFileDescriptors;
     metrics::LongValueMetric feedingBlocked;
     metrics::LongValueMetric mallocArena;
+    CpuUtilMetrics           cpu_util;
 
     ResourceUsageMetrics(metrics::MetricSet *parent);
     ~ResourceUsageMetrics();
diff --git a/searchcore/src/vespa/searchcore/proton/server/proton.cpp b/searchcore/src/vespa/searchcore/proton/server/proton.cpp
index b128fe16e5e..b2a86102cd1 100644
--- a/searchcore/src/vespa/searchcore/proton/server/proton.cpp
+++ b/searchcore/src/vespa/searchcore/proton/server/proton.cpp
@@ -61,6 +61,7 @@ using vespalib::Slime;
 using vespalib::makeLambdaTask;
 using vespalib::slime::ArrayInserter;
 using vespalib::slime::Cursor;
+using CpuCategory = vespalib::CpuUsage::Category;
 
 using search::transactionlog::DomainStats;
 using vespa::config::search::core::ProtonConfig;
@@ -207,6 +208,7 @@ Proton::Proton(const config::ConfigUri & configUri,
       StatusProducer(),
       IPersistenceEngineOwner(),
       ComponentConfigProducer(),
+      _cpu_util(),
       _configUri(configUri),
       _mutex(),
       _metricsHook(std::make_unique<MetricsUpdateHook>(*this)),
@@ -775,6 +777,12 @@ Proton::updateMetrics(const metrics::MetricLockGuard &)
 #else
         metrics.resourceUsage.mallocArena.set(UINT64_C(0));
 #endif
+        auto cpu_util = _cpu_util.get_util();
+        metrics.resourceUsage.cpu_util.setup.set(cpu_util[CpuCategory::SETUP]);
+        metrics.resourceUsage.cpu_util.read.set(cpu_util[CpuCategory::READ]);
+        metrics.resourceUsage.cpu_util.write.set(cpu_util[CpuCategory::WRITE]);
+        metrics.resourceUsage.cpu_util.compact.set(cpu_util[CpuCategory::COMPACT]);
+        metrics.resourceUsage.cpu_util.other.set(cpu_util[CpuCategory::OTHER]);
     }
     {
         ContentProtonMetrics::ProtonExecutorMetrics &metrics = _metricsEngine->root().executor;
diff --git a/searchcore/src/vespa/searchcore/proton/server/proton.h b/searchcore/src/vespa/searchcore/proton/server/proton.h
index 573f215c722..90a257a0aaa 100644
--- a/searchcore/src/vespa/searchcore/proton/server/proton.h
+++ b/searchcore/src/vespa/searchcore/proton/server/proton.h
@@ -25,6 +25,7 @@
 #include <vespa/vespalib/net/json_handler_repo.h>
 #include <vespa/vespalib/net/state_explorer.h>
 #include <vespa/vespalib/util/varholder.h>
+#include <vespa/vespalib/util/cpu_usage.h>
 #include <mutex>
 #include <shared_mutex>
 
@@ -81,6 +82,7 @@ private:
         void setClusterName(const vespalib::string &clusterName, const vespalib::string &baseDir);
     };
 
+    vespalib::CpuUtil                      _cpu_util;
     const config::ConfigUri                _configUri;
     mutable std::shared_mutex              _mutex;
     std::unique_ptr<metrics::UpdateHook>   _metricsHook;