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

Make response time statistics worker-local

Browse Source 
Given that the response times across threads are extremely likely to be
nearly identical, the data can be partitioned by worker thread.
This commit is contained in:
Markus Mäkelä
2018-09-11 00:01:02 +03:00
parent 675182e9eb
commit 62788f39e1
2 changed files with 10 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
server/core/internal/server.hh
View File

@ -19,6 +19,7 @@
#include <maxbase/average.hh>
#include <maxscale/server.h>
#include <maxscale/resultset.hh>
#include <maxscale/routingworker.hh>
std::unique_ptr<ResultSet> serverGetList();
@ -27,19 +28,24 @@ class Server : public SERVER
{
public:
Server()
: response_time(maxbase::EMAverage {0.04, 0.35, 500})
{
}
int response_time_num_samples() const
{
return response_time.num_samples();
return response_time->num_samples();
}
double response_time_average() const
{
return response_time.average();
return response_time->average();
}
void response_time_add(double ave, int num_samples)
{
response_time.add(ave, num_samples);
response_time->add(ave, num_samples);
}
private:
@ -47,7 +53,7 @@ private:
// can be calculated at runtime. The "500" or sample_max affects how often a
// session should updates this stat. sample_max should be slightly lower than max sample
// rate (which is less than qps due to the noise filter).
maxbase::EMAverage response_time = {0.04, 0.35, 500};
mxs::rworker_local<maxbase::EMAverage> response_time;
};
void server_free(Server* server);