GitBubble
commented
6 years ago DFX REPORT (PART 2)
Netdata investigation report-> to learn how netdata re-invent the wheels
After read the code for one hour or so , I realized netdata is a truely carefully written project.
It construct its sources from scratch with small utility functions where as normal open source project will using of the existed one. You may ask WHY?
netdata is aimed to run on every piece of cheap hardware and platform . that's a common goals with hi-benchmarks. Different system and architecture may have different effect when using the linux headers. some even not POSIX. In order to adapt netdata on these platforms, netdata re-written some system level functions .
the re-invent wheel does helpful in the project. to improve portability and performance. i think it's a one of the most important reasons why netdata is a popular project in github.
for example, in src/inlined.h
static inline char *strncpyz(char *dst, const char *src, size_t n) {
char * = dst;
while (*src && n --)
*dst++ = *src++;
*dst = '\0';
return p;it's a nearly standard implementation, the project consider the underlying environment may even without glibc header files.
compared with glibc implementations : https://github.com/lattera/glibc/blob/master/string/strncpy.c then, we will get a clue that these little tricks in all will make netdata a better performance monitor tool run every scenario.
#ifndef STRNCPY
#define STRNCPY strncpy
#endif
char *
STRNCPY (char *s1, const char *s2, size_t n)
{
char c;
char *s = s1;
--s1;
if (n >= 4)
{
size_t n4 = n >> 2;
for (;;)
{
c = *s2++;
*++s1 = c;
if (c == '\0')
break;
c = *s2++;
*++s1 = c;
if (c == '\0')
break;
c = *s2++;
*++s1 = c;
if (c == '\0')
break;
c = *s2++;
*++s1 = c;
if (c == '\0')
break;
if (--n4 == 0)
goto last_chars;
}
n = n - (s1 - s) - 1;
if (n == 0)
return s;
goto zero_fill;
}
last_chars:
n &= 3;
if (n == 0)
return s;
do
{
c = *s2++;
*++s1 = c;
if (--n == 0)
return s;
}
while (c != '\0');
zero_fill:
do
*++s1 = '\0';
while (--n > 0);
return s;
}but one thing confused me is that it declare every function with extern , extern is a default prefix to let compiler know there is a definition somewhere accordingly to expose the function to global wide even you don't add it .
foodtooth
DFX REPORT (PART 1)
Netdata investigation report-> how to monitor our ROS system
netdata is great tool for server monitoring , we try to learn some architecture from it. so here is a brief report to guide our plan in the near future.
1, something interested
netdata collect data on each $host$ just one second behind real status. they mean real-time.
netdata can mirror collected data to designated server. they call real time data replication and mirroring
netdata use a plugin language to add plugin to collect more metrics and can visualize the data as charts if only the item have some countable data.
netdata collect all its metrics locally.
netdata has alarms with pre-configured events and threshhold.
netdata can pipe messages to syslog and various of other tools. such as
(email addresses slack channels discord channels IRC channels pushover pushbullet telegram.org pagerduty twilio messagebird alerta flock twillo kavenegar)
2，the monitor and health alert is great
Monitor is interesting feature of netdata, specially interested in the following 2 area -monitor systemd service is useful -monitor IPMI tools
https://github.com/firehol/netdata/wiki/health-monitoring
3，Run time cost in different architecture
seems netdata is a not light weight monitor now. A liar !
50 threads, 14 shared libraries...
18 threads, 10 shared libraries:
17 threads, 11 shared libraries
4, Summary:
Sources written in a very clear style, but the build system is sophisticated. including autoconf files and CMake with well debugged script. The flexible architecture behind its design philosophy is a plugin system . it create some static thread to run plugin which take advantage of bash/python/systemd/etc utilities to collect statistic locally.it is fast but also linked too many libraries we will not need in ROS system.
5, conclusion:
we can adopt the plugin system ,and some parts of its http-server. but it linked to much burden we don't need. we need to optimize the source to customize the ros scenario. especially, the monitor and alert system is really a good wheel which can be used by us to create hi-benchmarks.