Per patch: multiple BMRR calls

This is to reduce sensitivity on testbed unstable performance.
Also add topo_cleanupbefore every pybot run
to avoid issues with VPP uninstall.

TRACE prints are left there to simplify investigation
of false positives if/when they happen.

Change-Id: I9b0cdcfbbe4aa0735a0596746ac32c9e93af0897
Signed-off-by: Vratko Polak <vrpolak@cisco.com>

1 files changed, 56 insertions, 25 deletions

diff --git a/resources/tools/scripts/compare_perpatch.py b/resources/tools/scripts/compare_perpatch.py
index cc9ffd8992..1f8a1cf892 100644
--- a/resources/tools/scripts/compare_perpatch.py
+++ b/resources/tools/scripts/compare_perpatch.py
@@ -14,7 +14,7 @@
 """Script for determining whether per-patch perf test votes -1.
 
 This script assumes there exist two text files with processed BMRR results,
-located at hardcoded relative paths, having several lines
+located at hardcoded relative paths (subdirs thereof), having several lines
 of json-parseable lists of float values, corresponding to testcase results.
 This script then uses jumpavg library to determine whether there was
 a regression, progression or no change for each testcase.
@@ -38,32 +38,62 @@ def hack(value_list):
     :rtype: list of float
     """
     tmp = sorted(value_list)
-    quarter = len(tmp) / 4
-    ret = tmp[quarter:-quarter]
+    eight = len(tmp) / 8
+    ret = tmp[3*eight:-eight]
     return ret
 
-parent_lines = list()
-new_lines = list()
-with open("csit_parent/results.txt") as parent_file:
-    parent_lines = parent_file.readlines()
-with open("csit_new/results.txt") as new_file:
-    new_lines = new_file.readlines()
-if len(parent_lines) != len(new_lines):
-    print "Number of passed tests does not match!"
-    sys.exit(1)
+iteration = -1
+parent_iterations = list()
+new_iterations = list()
+num_tests = None
+while 1:
+    iteration += 1
+    parent_lines = list()
+    new_lines = list()
+    filename = "csit_parent/{iter}/results.txt".format(iter=iteration)
+    try:
+        with open(filename) as parent_file:
+            parent_lines = parent_file.readlines()
+    except IOError:
+        break
+    num_lines = len(parent_lines)
+    filename = "csit_new/{iter}/results.txt".format(iter=iteration)
+    with open(filename) as new_file:
+        new_lines = new_file.readlines()
+    if num_lines != len(new_lines):
+        print "Number of tests does not match within iteration", iteration
+        sys.exit(1)
+    if num_tests is None:
+        num_tests = num_lines
+    elif num_tests != num_lines:
+        print "Number of tests does not match previous at iteration", iteration
+        sys.exit(1)
+    parent_iterations.append(parent_lines)
+    new_iterations.append(new_lines)
 classifier = BitCountingClassifier()
-num_tests = len(parent_lines)
 exit_code = 0
-for index in range(num_tests):
-    parent_values = hack(json.loads(parent_lines[index]))
-    new_values = hack(json.loads(new_lines[index]))
+for test_index in range(num_tests):
+    val_max = 1.0
+    parent_values = list()
+    new_values = list()
+    for iteration_index in range(len(parent_iterations)):
+        parent_values.extend(
+            json.loads(parent_iterations[iteration_index][test_index]))
+        new_values.extend(
+            json.loads(new_iterations[iteration_index][test_index]))
+    print "TRACE pre-hack parent: {p}".format(p=parent_values)
+    print "TRACE pre-hack new: {n}".format(n=new_values)
+    parent_values = hack(parent_values)
+    new_values = hack(new_values)
     parent_max = BitCountingMetadataFactory.find_max_value(parent_values)
     new_max = BitCountingMetadataFactory.find_max_value(new_values)
-    cmax = max(parent_max, new_max)
-    factory = BitCountingMetadataFactory(cmax)
+    val_max = max(val_max, parent_max, new_max)
+    factory = BitCountingMetadataFactory(val_max)
     parent_stats = factory.from_data(parent_values)
-    factory = BitCountingMetadataFactory(cmax, parent_stats.avg)
-    new_stats = factory.from_data(new_values)
+    new_factory = BitCountingMetadataFactory(val_max, parent_stats.avg)
+    new_stats = new_factory.from_data(new_values)
+    print "TRACE parent: {p}".format(p=parent_values)
+    print "TRACE new: {n}".format(n=new_values)
     print "DEBUG parent: {p}".format(p=parent_stats)
     print "DEBUG new: {n}".format(n=new_stats)
     common_max = max(parent_stats.avg, new_stats.avg)
@@ -71,15 +101,16 @@ for index in range(num_tests):
     print "DEBUG difference: {d}%".format(d=100 * difference)
     classified_list = classifier.classify([parent_stats, new_stats])
     if len(classified_list) < 2:
-        print "Test index {index}: normal (no anomaly)".format(
-            index=index)
+        print "Test test_index {test_index}: normal (no anomaly)".format(
+            test_index=test_index)
         continue
     anomaly = classified_list[1].metadata.classification
     if anomaly == "regression":
-        print "Test index {index}: anomaly regression".format(index=index)
+        print "Test test_index {test_index}: anomaly regression".format(
+            test_index=test_index)
         exit_code = 1
         continue
-    print "Test index {index}: anomaly {anomaly}".format(
-        index=index, anomaly=anomaly)
+    print "Test test_index {test_index}: anomaly {anomaly}".format(
+        test_index=test_index, anomaly=anomaly)
 print "DEBUG exit code {code}".format(code=exit_code)
 sys.exit(exit_code)