[05:56:32] <ori>	 TimStarling: is it currently possible to annotate Excimer stack traces with request latency? I _think_ the answer is no, because the flush callback receives traces from other requests (so you can't use the current request's $_SERVER['REQUEST_TIME']), and ExcimerLogEntry::getTimestamp() gives time since the excimer timer was started, which was probably prior to the start of any sampled
[05:56:34] <ori>	 request. Am I totally off?
[05:58:24] <TimStarling>	 the flush callback receives traces only from the current request
[05:59:29] <ori>	 traces, plural? probably only one, right?
[06:01:30] <ori>	 assuming site-wide profiling with a large sample period
[06:05:21] <ori>	 so you could potentially annotate the excimer samples with the request time, and use that to generate tail latency flame graphs in arc-lamp or whatever, right?
[08:49:30] <Krinkle>	 ori: one thing er do have today is flame graphs only of post send time. I'm guessing this's be similar but with the cut after a time threshold instead of parent method
[13:55:42] <ori>	 oh, right
[17:40:09] <ori>	 TimStarling: I'm confused about how Excimer works. IIRC, with HHVM's Xenon, the timer was global, so each time the timer fired, it captured stack traces from all requests that were in-flight. With Excimer, it seems that a new timer is wound by each request. Does this mean that requests with a duration of less than 60s don't get sampled at all in Wikimedia's production setup?
[17:47:09] <ori>	 if that's true it seems a lot less useful