5 Commits

Author	SHA1	Message	Date
Janmejay Singh	da8c0b9e92	Double impstats flush-interval because 1-flush-per-second does not give the test enough tolerance on slow machines.	2017-04-20 12:26:55 +05:30
Janmejay Singh	5f071af97f	Fixed dyn-stats unused-metric-ttl bug which was was expected to garbage-collect accumulators that were unused for time-period >= ttl, but used to wipe all accumulators. This worked from memory-footprint and metric-life point-of-view, but used to reset accumulators even when they were configured to not be reset. Now it maintains survivor(shadow) table which keeps the accumulator value around from ttl up-to (2 * ttl) and reclaims it only if no dyn_inc call resurrects it by the end of this period. This allows us to resurrect surviving counters if they are used within ttl duration. This keeps a counter alive up-to (2 * ttl) in the worst case(late eviction) and ttl(timely eviction) in the best case.	2016-04-14 21:03:35 +05:30
Janmejay Singh	7a96794b12	made some changes to dyn-stats tests to make it more resillient to timing-problems (prevents bad-timing introduced occasional failures)	2015-11-17 17:48:59 +05:30
Janmejay Singh	33cb11db57	dynstats reset-of-accumulated-dyn-metric-names and more-metrics-than-expected flows fixed and tested	2015-11-16 16:43:51 +05:30
Janmejay Singh	e24c175cae	- fixed vg-tests to actually check exit code ... - fixed and excercised reset dyn-metrics-table flow (where dynamic-metric names are discarded. this excercises unusedMetricLife timer expiry.)	2015-11-13 20:59:07 +05:30