From: Tomek Mrugalski <tomasz@isc.org>
Date: Mon, 11 May 2015 10:48:38 +0000 (+0200)
Subject: [3793] Minor comment tweaks.
X-Git-Tag: trac3513_base~5^2~1^2~2
X-Git-Url: http://git.ipfire.org/gitweb.cgi?a=commitdiff_plain;h=cd709db1ef0dcfeead2a6d92e549eb1e345e954a;p=thirdparty%2Fkea.git

[3793] Minor comment tweaks.
---

diff --git a/src/lib/stats/Makefile.am b/src/lib/stats/Makefile.am
index 99c66cc000..f7b41cd18d 100644
--- a/src/lib/stats/Makefile.am
+++ b/src/lib/stats/Makefile.am
@@ -10,6 +10,10 @@ libkea_stats_la_SOURCES += stats_mgr.h stats_mgr.cc
 
 libkea_stats_la_LIBADD  = $(top_builddir)/src/lib/cc/libkea-cc.la
 
+# This library seem to be required when certain methods are called from
+# boost::posix_time. In particular, to_simple_string(ptime) and
+# to_simple_string(time_duration) require it on Ubuntu. It doesn't seem
+# to be required on Mac OS, though.
 libkea_stats_la_LIBADD += -lboost_date_time
 
 libkea_stats_includedir = $(includedir)/$(PACKAGE_NAME)/stats
diff --git a/src/lib/stats/stats_mgr.h b/src/lib/stats/stats_mgr.h
index 3212aba86c..f38a4c7bee 100644
--- a/src/lib/stats/stats_mgr.h
+++ b/src/lib/stats/stats_mgr.h
@@ -31,6 +31,23 @@ namespace stats {
 /// StatsMgr is a singleton class that represents a subsystem that manages
 /// collection, storage and reporting of various types of statistics.
 /// It is also the intended API for both core code and hooks.
+///
+/// As of May 2015, Tomek ran performance benchmarks (see unit-tests in
+/// stats_mgr_unittest.cc with performance in their names) and it seems
+/// the code is able to register ~2.5-3 million observations per second, even
+/// with 1000 different statistics recored. That seems sufficient for now,
+/// so there is no immediate need to develop any multi-threading solutions
+/// for now. However, should this decision be revised in the future, the
+/// best place for it would to be modify @ref addObservation method here.
+/// It's the common code point that all new observations must pass through.
+/// One possible way to enable multi-threading would be to run a separate
+/// thread handling collection. The main thread would call @ref addValue and
+/// @ref setValue methods that would end up calling @ref addObservation.
+/// That method would pass the data to separate thread to be collected and
+/// would immediately return. Further processing would be mostly as it
+/// is today, except happening in a separate thread. One unsolved issue in
+/// this approach is how to extract data, but that will remain unsolvable
+/// until we get the control socket implementation.
 class StatsMgr : public boost::noncopyable {
  public:
 
@@ -101,7 +118,7 @@ class StatsMgr : public boost::noncopyable {
 
     /// @brief determines whether a given statistic is kept as a single value
     ///        or as a number of values
-    /// 
+    ///
     /// Specifies that statistic name should be stored not as a single value,
     /// but rather as a set of values. duration determines the timespan.
     /// Samples older than duration will be discarded. This is time-constrained