# Example 848 — The slo was met over a window with no requests `the_slo_was_met_over_a_window_with_no_requests.eml` - The latency SLO reads 100 per hundred met for the hour under review, and the computation is honest. How many requests that hour is computed below. ## EML ```eml # Self-authored for the EML case corpus (no external origin). The latency SLO # reads 100 per hundred met for the hour under review, and the computation is # honest. How many requests that hour is computed below. # # The SLO is measured well. It counts real requests against the threshold, not a # synthetic probe; a request over the threshold counts as a miss; the window is # the clock hour, not a rolling one chosen afterwards; and a miss rate above the # budget pages on-call. # # The hour under review had no requests, and zero misses over zero requests is # reported as fully met. 3600 => seconds_in_the_hour 0 => requests_in_the_hour 0 => requests_over_the_threshold 10000 => reported_slo_met_per_myriad 0 => alerts_that_fired 40 => minutes_the_service_was_unreachable requests_in_the_hour - requests_over_the_threshold => requests_under_the_threshold int(minutes_the_service_was_unreachable * 10000 / 60) => share_of_the_hour_unreachable_per_myriad "requests in the hour : " + str(requests_in_the_hour) ^0 " over the threshold : " + str(requests_over_the_threshold) ^0 " under the threshold : " + str(requests_under_the_threshold) ^0 "reported SLO met : " + str(reported_slo_met_per_myriad) + " per ten thousand" ^0 "alerts that fired : " + str(alerts_that_fired) ^0 "minutes the service was unreachable : " + str(minutes_the_service_was_unreachable) ^0 "share of the hour unreachable : " + str(share_of_the_hour_unreachable_per_myriad) + " per ten thousand" ^0 "" ^0 # ---- what the SLO verified ---- "the latency SLO" ^0 " counts : real requests against the threshold" ^0 " a slow request : counts as a miss" ^0 " window : the clock hour, not a rolling one" ^0 " pages when : the miss rate exceeds the budget" ^0 " requests over the threshold : 0" ^0 " verdict : 100 PER HUNDRED MET" ^0 "" ^0 " counting real requests rather than a synthetic probe is" ^0 " the part done right here, and it is why a real slow hour" ^0 " would show" ^0 "" ^0 # ---- what the ratio is over ---- "met = (requests under threshold) / (requests)" ^0 " requests this hour : " + str(requests_in_the_hour) ^0 " misses this hour : " + str(requests_over_the_threshold) ^0 " the ratio : zero over zero, reported as fully met" ^0 " what '100 per hundred met' of no requests means : every" ^0 " one of no requests was fast, trivially" ^0 " what it does not mean : that the service served anyone" ^0 "" ^0 # ---- why there were no requests ---- "the empty hour" ^0 " why no requests : the service was unreachable for " ^0 " " + str(minutes_the_service_was_unreachable) + " minutes, and the load balancer shed the rest" ^0 " what an outage looks like to a per-request SLO : an" ^0 " hour with no misses" ^0 " did availability enter this metric : no; it is over" ^0 " served requests" ^0 " is the SLO figure false : no; it is vacuously true" ^0 "" ^0 # ---- null control ---- # The same hour, with the SLO undefined when no requests are served and a # separate availability check that fires on an empty window. 10000 => nc_slo_when_empty_reads_met 0 => nc_slo_when_empty_is_undefined 1 => nc_availability_alerts_on_empty "null control - an empty window is not a met SLO" ^0 " SLO when empty reads as met : " + str(nc_slo_when_empty_reads_met) ^0 " SLO when empty is undefined : " + str(nc_slo_when_empty_is_undefined) ^0 " availability alerts on an empty window : " ^0 " " + str(nc_availability_alerts_on_empty) ^0 " no request changed; an hour with no traffic stopped" ^0 " counting as an hour that met its latency target" ^0 "" ^0 # ---- the rule ---- "what a met latency SLO guarantees" ^0 " no served request exceeded the threshold : exactly," ^0 " real requests, clock hour, budget armed" ^0 " the service met its target : not addressed; the window" ^0 " had no requests, and zero misses over zero requests is" ^0 " vacuously 100 per hundred - the service was unreachable" ^0 " " + str(minutes_the_service_was_unreachable) + " minutes and the metric fired no alert" ^0 "" ^0 "a per-request quality metric is silent when there are no requests, and silence" ^0 "reads as perfection; the hour a service serves no one is the hour its latency" ^0 "SLO looks best" ^0 "" ^0 "It counts real requests against the threshold over the clock hour and pages on" ^0 "a miss - 100 per hundred met. The hour had " + str(requests_in_the_hour) + " requests, so zero misses over" ^0 "zero is vacuously met while the service was unreachable " + str(minutes_the_service_was_unreachable) + " minutes, " ^0 "" + str(share_of_the_hour_unreachable_per_myriad) + " per ten thousand of it, under " + str(alerts_that_fired) + " alerts." ^0 ``` ## Python (deterministic transpilation) ```python seconds_in_the_hour = 3600 requests_in_the_hour = 0 requests_over_the_threshold = 0 reported_slo_met_per_myriad = 10000 alerts_that_fired = 0 minutes_the_service_was_unreachable = 40 requests_under_the_threshold = requests_in_the_hour - requests_over_the_threshold share_of_the_hour_unreachable_per_myriad = int(minutes_the_service_was_unreachable * 10000 / 60) print("requests in the hour : " + str(requests_in_the_hour)) print(" over the threshold : " + str(requests_over_the_threshold)) print(" under the threshold : " + str(requests_under_the_threshold)) print("reported SLO met : " + str(reported_slo_met_per_myriad) + " per ten thousand") print("alerts that fired : " + str(alerts_that_fired)) print("minutes the service was unreachable : " + str(minutes_the_service_was_unreachable)) print("share of the hour unreachable : " + str(share_of_the_hour_unreachable_per_myriad) + " per ten thousand") print("") print("the latency SLO") print(" counts : real requests against the threshold") print(" a slow request : counts as a miss") print(" window : the clock hour, not a rolling one") print(" pages when : the miss rate exceeds the budget") print(" requests over the threshold : 0") print(" verdict : 100 PER HUNDRED MET") print("") print(" counting real requests rather than a synthetic probe is") print(" the part done right here, and it is why a real slow hour") print(" would show") print("") print("met = (requests under threshold) / (requests)") print(" requests this hour : " + str(requests_in_the_hour)) print(" misses this hour : " + str(requests_over_the_threshold)) print(" the ratio : zero over zero, reported as fully met") print(" what '100 per hundred met' of no requests means : every") print(" one of no requests was fast, trivially") print(" what it does not mean : that the service served anyone") print("") print("the empty hour") print(" why no requests : the service was unreachable for ") print(" " + str(minutes_the_service_was_unreachable) + " minutes, and the load balancer shed the rest") print(" what an outage looks like to a per-request SLO : an") print(" hour with no misses") print(" did availability enter this metric : no; it is over") print(" served requests") print(" is the SLO figure false : no; it is vacuously true") print("") nc_slo_when_empty_reads_met = 10000 nc_slo_when_empty_is_undefined = 0 nc_availability_alerts_on_empty = 1 print("null control - an empty window is not a met SLO") print(" SLO when empty reads as met : " + str(nc_slo_when_empty_reads_met)) print(" SLO when empty is undefined : " + str(nc_slo_when_empty_is_undefined)) print(" availability alerts on an empty window : ") print(" " + str(nc_availability_alerts_on_empty)) print(" no request changed; an hour with no traffic stopped") print(" counting as an hour that met its latency target") print("") print("what a met latency SLO guarantees") print(" no served request exceeded the threshold : exactly,") print(" real requests, clock hour, budget armed") print(" the service met its target : not addressed; the window") print(" had no requests, and zero misses over zero requests is") print(" vacuously 100 per hundred - the service was unreachable") print(" " + str(minutes_the_service_was_unreachable) + " minutes and the metric fired no alert") print("") print("a per-request quality metric is silent when there are no requests, and silence") print("reads as perfection; the hour a service serves no one is the hour its latency") print("SLO looks best") print("") print("It counts real requests against the threshold over the clock hour and pages on") print("a miss - 100 per hundred met. The hour had " + str(requests_in_the_hour) + " requests, so zero misses over") print("zero is vacuously met while the service was unreachable " + str(minutes_the_service_was_unreachable) + " minutes, ") print("" + str(share_of_the_hour_unreachable_per_myriad) + " per ten thousand of it, under " + str(alerts_that_fired) + " alerts.") ``` ## stdout (executed) ```text requests in the hour : 0 over the threshold : 0 under the threshold : 0 reported SLO met : 10000 per ten thousand alerts that fired : 0 minutes the service was unreachable : 40 share of the hour unreachable : 6666 per ten thousand the latency SLO counts : real requests against the threshold a slow request : counts as a miss window : the clock hour, not a rolling one pages when : the miss rate exceeds the budget requests over the threshold : 0 verdict : 100 PER HUNDRED MET counting real requests rather than a synthetic probe is the part done right here, and it is why a real slow hour would show met = (requests under threshold) / (requests) requests this hour : 0 misses this hour : 0 the ratio : zero over zero, reported as fully met what '100 per hundred met' of no requests means : every one of no requests was fast, trivially what it does not mean : that the service served anyone the empty hour why no requests : the service was unreachable for 40 minutes, and the load balancer shed the rest what an outage looks like to a per-request SLO : an hour with no misses did availability enter this metric : no; it is over served requests is the SLO figure false : no; it is vacuously true null control - an empty window is not a met SLO SLO when empty reads as met : 10000 SLO when empty is undefined : 0 availability alerts on an empty window : 1 no request changed; an hour with no traffic stopped counting as an hour that met its latency target what a met latency SLO guarantees no served request exceeded the threshold : exactly, real requests, clock hour, budget armed the service met its target : not addressed; the window had no requests, and zero misses over zero requests is vacuously 100 per hundred - the service was unreachable 40 minutes and the metric fired no alert a per-request quality metric is silent when there are no requests, and silence reads as perfection; the hour a service serves no one is the hour its latency SLO looks best It counts real requests against the threshold over the clock hour and pages on a miss - 100 per hundred met. The hour had 0 requests, so zero misses over zero is vacuously met while the service was unreachable 40 minutes, 6666 per ten thousand of it, under 0 alerts. ``` ## Round-trip `ok: true` — round-trip fixpoint reached (python1 == python2) ## Trace event types eml:run:start · eml:assign · eml:output · eml:run:done