Percentile response time anomalies at 100%

[mblakele]

With locust 0.7.2 there seems to be a calculation problem with the 100% column. It's often reported a few ms lower than the preceding columns, which should be impossible. In the example below it's especially glaring when there was only one request for a particular endpoint: 99%: 580; 100%: 575. But I see a similar anomaly in the last line, which has 53 requests. In my data the anomalies seem to be common with less that 100 requests, but rare or at least invisible after that.

Percentage of the requests completed within given times
 Name                                                           # reqs    50%    66%    75%    80%    90%    95%    98%    99%   100%
--------------------------------------------------------------------------------------------------------------------------------------------
 GET /                                                            5346     70     81     89     96    120    130    160    170    282
 GET /rankings/networks/search/1063/                                 1    660    660    660    660    660    660    660    660    661
 GET /rankings/networks/search/1067/                                 1    630    630    630    630    630    630    630    630    625
 GET /rankings/networks/search/1068/                                 1    580    580    580    580    580    580    580    580    575
 GET /terms_of_service/?next=/                                      53    430    580    760    830    890    940    960   1000    996

[cyberw]

I think (hope) this has been solved. Let me know if this is not the case.