Use coarsetime #92
Use coarsetime #92
Conversation
| event->start_time = ctx->PHP_FUNCTION_ARG_START_TIME; | ||
| event->end_time = ctx->PHP_FUNCTION_ARG_END_TIME; | ||
| event->timestamp = bpf_ktime_get_ns(); | ||
| event->elapsed = ctx->PHP_FUNCTION_ARG_ELAPSED; |
There was a problem hiding this comment.
That is actually clever.
Because you are now no longer affecting performance you can get the ts in the event instead of from the actual function in PHP. Event ordering is still giving the parent hierarchy, right?
Another way - though slightly more effort is to create a real realtime clock TS at start of request.
And then pass this realtime start, monotonic start and monotonic end always to the probe! - that way it’s both realtime and not really.
There was a problem hiding this comment.
It has been working well. I really like decoupling the timestamps and letting the probe me a metric.
| @@ -30,16 +30,16 @@ unsafe extern "C" fn execute_ex(execute_data: *mut sys::zend_execute_data) { | |||
| }; | |||
|
|
|||
| // Run the upstream function and record the duration. | |||
| let start = get_unix_timestamp_micros(); | |||
| let start = Instant::recent(); | |||
There was a problem hiding this comment.
One problem is that you only update on function end. I don’t know how realistic it is to never call a small function that returns, but maybe a good way to solve this is to have a call stack depth % 20 or so and call update() every once in a while.
That can be a stack local variable - right now.
There was a problem hiding this comment.
I'll take a look at options. Thanks for the review @LionsAd !
What does this change