Bug #1012
no stats provided by smuxi-server
| Status: | New | Start: | 01/05/2015 | |
| Priority: | Normal | Due date: | ||
| Assigned to: | - | % Done: | 0% |
|
| Category: | Engine | |||
| Target version: | - | |||
| Complexity: | Found in Version: | |||
| Votes: | 1 (View) |
Description
something like that but also with network usage (nethogs provide this but not for user and not a single process as i could see)
pidstat -uIlhd -p `pgrep -fl mono | grep server | awk '{print $1}'` 2 5
A possible reason why such functionality could be wanted is, before doing the maintance work it would be usefull to see how heavy it is used by the end users.Another usecase might be to collect perfomance and usage data of the server, so the schedule of the maintance gets easier.
Usefull information i would be interested in
- connected user/frontends
- uptime
- time since last connects (could be merged in --list-users)
- io or netio
- (size of buffers when they are limited)
History
Updated by Mirco Bauer 3642 days ago
Not sure how this works. Can you provide a working/assumed example? It this using signals and prints to stdout or so?
Updated by Andre n 3642 days ago
Mirco Bauer wrote:
Not sure how this works. Can you provide a working/assumed example? It this using signals and prints to stdout or so?
me neither. (to be sure)
But i would split the functionality between an internal "API" to query and a dedicated binary for that to not affect the engine, when statistics are hanging or producing too much load.
For connected user/frontends i have to less insights how smuxi works, it might be good to react on a SIGnal or some method to query the state.
you find network statistics for a process in:
cat /proc/`pgrep -fl mono | grep server | awk '{print $1}'`/net/dev
a tool to observe this for all processes might be nethogs, but it is written in c++ :/ So i'm not very motiviated to analyze what they do
since i cannot say from monitoring the application from the outside if there is any user connected, i cannot
i didn't find any way to discover if users are connected to it.
If there is interest in i could to wrap the the provided oneliner up in a script.
If the performance impact is too high someone can still try to migrate from accessing proc or sysfs to using some lower API.