load_mon: increase cpuload update rate, get corresponding time safetly, and misc cleanup #15342
Conversation
dagar
commented
Jul 15, 2020
dagar
commented
- increase rate from 1 Hz to 3.3 Hz (every 300 ms)
- cpu load calculation grab timestamp with runtime in a critical section
- task stack only check one task per cycle (but cycle at a higher rate)
- decrease available FD threshold 3 -> 2
- minor cleanup (split out header)
dagar
force-pushed
the
pr-load_mon_updates
branch
from
1a018c9
to
5b53633
Compare
- increase rate - cpu load calculation grab timestamp atomically - only check one task per cycle (but cycle at a higher rate) - decrease available FD threshold - minor cleanup
dagar
force-pushed
the
pr-load_mon_updates
branch
from
cd2fd59
to
cb366dc
Compare
|
I increased the rate slightly because I wanted to have a bit more detail in the logs. |
dagar
changed the title
|
|
||
| // Found task low on file descriptors, report and exit. Continue here in next cycle. | ||
| if (fds_free < FDS_LOW_WARNING_THRESHOLD) { | ||
| PX4_WARN("%s low on FDs! (%i FDs left)", task_stack_info.task_name, fds_free); |
There was a problem hiding this comment.
These warnings will now appear 3 times per second and spam the log quite a bit, right?
There was a problem hiding this comment.
These warnings will now appear 3 times per second and spam the log quite a bit, right?
Only in the worst case if running the max number of tasks (32 or 64) and every one of them was either low on stack or FDs.
Overall it's kind of a lateral move. Currently in master it cycles at 1 Hz, but each iteration checks 2 tasks, and if a task entry (an array of CONFIG_MAX_TASKS) is empty it jumps forward. Now it runs slightly faster for the sake of cpuload measurement, but it's doing less work per iteration checking tasks.
