From: bert hubert <bert.hubert@netherlabs.nl>
Date: Mon, 1 Apr 2019 12:47:59 +0000 (+0200)
Subject: This provides CPU usage statistics per thread (worker & distributor).
X-Git-Tag: dnsdist-1.4.0-alpha1~18^2
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=refs%2Fpull%2F7649%2Fhead;p=thirdparty%2Fpdns.git

This provides CPU usage statistics per thread (worker & distributor).

(cherry picked from commit 4ae78615a5a0fb0721ebe1a58a1235e75c47d400)
---

diff --git a/pdns/pdns_recursor.cc b/pdns/pdns_recursor.cc
index 231232248c..1c9f5746fc 100644
--- a/pdns/pdns_recursor.cc
+++ b/pdns/pdns_recursor.cc
@@ -3031,6 +3031,7 @@ template string broadcastAccFunction(const boost::function<string*()>& fun); //
 template uint64_t broadcastAccFunction(const boost::function<uint64_t*()>& fun); // explicit instantiation
 template vector<ComboAddress> broadcastAccFunction(const boost::function<vector<ComboAddress> *()>& fun); // explicit instantiation
 template vector<pair<DNSName,uint16_t> > broadcastAccFunction(const boost::function<vector<pair<DNSName, uint16_t> > *()>& fun); // explicit instantiation
+template ThreadTimes broadcastAccFunction(const boost::function<ThreadTimes*()>& fun);
 
 static void handleRCC(int fd, FDMultiplexer::funcparam_t& var)
 {
diff --git a/pdns/rec_channel_rec.cc b/pdns/rec_channel_rec.cc
index 453885bad4..77bbd872e8 100644
--- a/pdns/rec_channel_rec.cc
+++ b/pdns/rec_channel_rec.cc
@@ -762,6 +762,48 @@ static uint64_t getUserTimeMsec()
   return (ru.ru_utime.tv_sec*1000ULL + ru.ru_utime.tv_usec/1000);
 }
 
+/* This is a pretty weird set of functions. To get per-thread cpu usage numbers,
+   we have to ask a thread over a pipe. We could do so surgically, so if you want to know about
+   thread 3, we pick pipe 3, but we lack that infrastructure.
+
+   We can however ask "execute this function on all threads and add up the results".
+   This is what the first function does using a custom object ThreadTimes, which if you add
+   to each other keeps filling the first one with CPU usage numbers
+*/
+
+static ThreadTimes* pleaseGetThreadCPUMsec()
+{
+  uint64_t ret=0;
+#ifdef RUSAGE_THREAD
+  struct rusage ru;
+  getrusage(RUSAGE_THREAD, &ru);
+  ret = (ru.ru_utime.tv_sec*1000ULL + ru.ru_utime.tv_usec/1000);
+  ret += (ru.ru_stime.tv_sec*1000ULL + ru.ru_stime.tv_usec/1000);
+#endif
+  return new ThreadTimes{ret};
+}
+
+/* Next up, when you want msec data for a specific thread, we check
+   if we recently executed pleaseGetThreadCPUMsec. If we didn't we do so
+   now and consult all threads.
+
+   We then answer you from the (re)fresh(ed) ThreadTimes.
+*/
+static uint64_t doGetThreadCPUMsec(int n)
+{
+  static std::mutex s_mut;
+  static time_t last = 0;
+  static ThreadTimes tt;
+
+  std::lock_guard<std::mutex> l(s_mut);
+  if(last != time(nullptr)) {
+   tt = broadcastAccFunction<ThreadTimes>(pleaseGetThreadCPUMsec);
+   last = time(nullptr);
+  }
+
+  return tt.times.at(n);
+}
+
 static uint64_t calculateUptime()
 {
   return time(nullptr) - g_stats.startupTime;
@@ -1082,6 +1124,9 @@ void registerAllStats()
   addGetStat("user-msec", getUserTimeMsec);
   addGetStat("sys-msec", getSysTimeMsec);
 
+  for(unsigned int n=0; n < g_numThreads; ++n)
+    addGetStat("cpu-msec-thread-"+std::to_string(n), boost::bind(&doGetThreadCPUMsec, n));
+
 #ifdef MALLOC_TRACE
   addGetStat("memory-allocs", boost::bind(&MallocTracer::getAllocs, g_mtracer, string()));
   addGetStat("memory-alloc-flux", boost::bind(&MallocTracer::getAllocFlux, g_mtracer, string()));
diff --git a/pdns/recursordist/docs/metrics.rst b/pdns/recursordist/docs/metrics.rst
index 1fbf805763..c04ebbc01f 100644
--- a/pdns/recursordist/docs/metrics.rst
+++ b/pdns/recursordist/docs/metrics.rst
@@ -185,6 +185,10 @@ concurrent-queries
 ^^^^^^^^^^^^^^^^^^
 shows the number of MThreads currently   running
 
+cpu-msec-thread-n
+^^^^^^^^^^^^^^^^^
+shows the number of milliseconds spent in thread n. Available since 4.1.12.
+
 dlg-only-drops
 ^^^^^^^^^^^^^^
 number of records dropped because of :ref:`setting-delegation-only` setting
diff --git a/pdns/syncres.hh b/pdns/syncres.hh
index bba453e2cf..e2562a1c2f 100644
--- a/pdns/syncres.hh
+++ b/pdns/syncres.hh
@@ -1037,3 +1037,14 @@ void doCarbonDump(void*);
 void primeHints(void);
 
 extern __thread struct timeval g_now;
+
+struct ThreadTimes
+{
+  uint64_t msec;
+  vector<uint64_t> times;
+  ThreadTimes& operator+=(const ThreadTimes& rhs)
+  {
+    times.push_back(rhs.msec);
+    return *this;
+  }
+};