hcq/MaxScale
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MXS-1777 ResponseStat improvements

Browse Source 
Fix comments.
Fix a bug in make_valid().
Change sync time (when the average should be pushed to the server EMA)
to only depend on time, not use sample_max. This decreases the amount of
sync calls, and allows for a much shorter sync time. Testing shows this to be
more stabel and allow better control of the sample_max.
This commit is contained in:
Niclas Antti
2018-10-03 16:35:06 +03:00
parent 268e689dc5
commit 5892ef18af
2 changed files with 14 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
include/maxscale/response_stat.hh
View File

@ -16,42 +16,33 @@
#include <maxbase/stopwatch.hh> #include <maxbase/stopwatch.hh>
#include <maxbase/average.hh> #include <maxbase/average.hh>
/** This could arguably be a utility, but is written specifically for rwsplit
* so it stays here at least for now.
*/
namespace maxscale namespace maxscale
{ {
/** /**
* Query response statistics. Uses median of N samples to filter noise, then * Query response time average for a backend. Uses median of N samples to filter noise,
* uses those medians to calculate the average response time. * then uses those medians to calculate the average response time.
* The class makes an average of the durations between calls to query_started()
* and query_ended(). Once the stats are good, sync_time_reached(int max) returns true,
* based on the average containing at least max samples (or medians), or the time
* sync_duration (constructor arg) has passed since the last reset().
*/ */
class ResponseStat class ResponseStat
{ {
public: public:
/* /*
* @param num_filter_samples - collect num samples, use median * @param num_filter_samples - collect num samples, use median (digital filter)
* @param num_synch_medians - this many medians before the average should be synced, or * @param sync_duration - this much time between synchronize to server stats.
* @param sync_duration - this much time between syncs.
*/ */
ResponseStat(int num_filter_samples = 5, ResponseStat(int num_filter_samples = 9,
int num_synch_medians = 500, maxbase::Duration sync_duration = std::chrono::milliseconds(250));
maxbase::Duration sync_duration = std::chrono::seconds(5));
void query_started(); void query_started();
void query_ended();// ok to call without a query_started void query_ended(); // ok to call without a query_started
bool make_valid(); // make valid even if there are too few filter_samples bool make_valid(); // make valid even if there are only filter_samples
bool is_valid() const; bool is_valid() const;
int num_samples() const; int num_samples() const;
maxbase::Duration average() const; maxbase::Duration average() const;
bool sync_time_reached(); // is it time to apply the average? bool sync_time_reached(); // is it time to apply the average to the server?
void reset(); void reset();
private: private:
const int m_num_filter_samples; const int m_num_filter_samples;
const int m_num_synch_medians;
const maxbase::Duration m_sync_duration; const maxbase::Duration m_sync_duration;
int m_sample_count; int m_sample_count;
std::vector<maxbase::Duration> m_samples; // N sampels from which median is used std::vector<maxbase::Duration> m_samples; // N sampels from which median is used

9
server/core/response_stat.cc
View File

@ -17,10 +17,8 @@
namespace maxscale namespace maxscale
{ {
ResponseStat::ResponseStat(int num_filter_samples, ResponseStat::ResponseStat(int num_filter_samples,
int num_synch_medians,
maxbase::Duration sync_duration) maxbase::Duration sync_duration)
: m_num_filter_samples {num_filter_samples} : m_num_filter_samples {num_filter_samples}
, m_num_synch_medians{num_synch_medians}
, m_sync_duration{sync_duration} , m_sync_duration{sync_duration}
, m_sample_count{0} , m_sample_count{0}
, m_samples(num_filter_samples) , m_samples(num_filter_samples)
@ -39,7 +37,7 @@ void ResponseStat::query_ended()
{ {
if (m_last_start == maxbase::TimePoint()) if (m_last_start == maxbase::TimePoint())
{ {
// m_last_start is defaulted. Ignore, avoids extra logic in call sites. // m_last_start is defaulted. Ignore, avoids extra logic at call sites.
return; return;
} }
m_samples[m_sample_count] = maxbase::Clock::now() - m_last_start; m_samples[m_sample_count] = maxbase::Clock::now() - m_last_start;
@ -58,9 +56,11 @@ bool ResponseStat::make_valid()
{ {
if (!m_average.num_samples() && m_sample_count) if (!m_average.num_samples() && m_sample_count)
{ {
std::sort(m_samples.begin(), m_samples.begin() + m_sample_count);
maxbase::Duration new_sample = m_samples[m_sample_count / 2]; maxbase::Duration new_sample = m_samples[m_sample_count / 2];
m_average.add(std::chrono::duration<double>(new_sample).count()); m_average.add(std::chrono::duration<double>(new_sample).count());
m_sample_count = 0; m_sample_count = 0;
m_last_start = maxbase::TimePoint();
} }
return is_valid(); return is_valid();
@ -84,8 +84,7 @@ maxbase::Duration ResponseStat::average() const
bool ResponseStat::sync_time_reached() bool ResponseStat::sync_time_reached()
{ {
auto now = maxbase::Clock::now(); auto now = maxbase::Clock::now();
bool reached = m_next_sync < now bool reached = m_next_sync < now;
|| m_average.num_samples() >= m_num_synch_medians;
if (reached) if (reached)
{ {