diff options
author | Tibor Frank <tifrank@cisco.com> | 2018-04-23 09:29:29 +0200 |
---|---|---|
committer | Tibor Frank <tifrank@cisco.com> | 2018-04-23 09:52:34 +0200 |
commit | 7aff808942440ba9729bf60888738b920c024324 (patch) | |
tree | ad5f9628e3cc2e78c7733c03e8eba0ca377f7621 /resources/tools | |
parent | 7cfdac0cf07e3a7d9a8b53b7621f8b7500fd1515 (diff) |
CSIT-1041: Trending dashboard
- add long term classification
Change-Id: Iec8d2275c01da3b1e41435b9abfb70ec3fe98fd3
Signed-off-by: Tibor Frank <tifrank@cisco.com>
Diffstat (limited to 'resources/tools')
-rw-r--r-- | resources/tools/presentation/generator_tables.py | 73 | ||||
-rw-r--r-- | resources/tools/presentation/specification_CPTA.yaml | 6 |
2 files changed, 35 insertions, 44 deletions
diff --git a/resources/tools/presentation/generator_tables.py b/resources/tools/presentation/generator_tables.py index 50a6623356..59162444dc 100644 --- a/resources/tools/presentation/generator_tables.py +++ b/resources/tools/presentation/generator_tables.py @@ -677,6 +677,7 @@ def table_performance_trending_dashboard(table, input_data): # Prepare the header of the tables header = ["Test Case", "Throughput Trend [Mpps]", + "Long Trend Compliance", "Trend Compliance", "Top Anomaly [Mpps]", "Change [%]", @@ -706,12 +707,14 @@ def table_performance_trending_dashboard(table, input_data): if len(tbl_dict[tst_name]["data"]) > 2: pd_data = pd.Series(tbl_dict[tst_name]["data"]) - win_size = pd_data.size \ - if pd_data.size < table["window"] else table["window"] + win_size = min(pd_data.size, table["window"]) # Test name: name = tbl_dict[tst_name]["name"] median = pd_data.rolling(window=win_size, min_periods=2).median() + median_idx = pd_data.size - table["long-trend-window"] + median_idx = 0 if median_idx < 0 else median_idx + max_median = max(median.values[median_idx:]) trimmed_data, _ = split_outliers(pd_data, outlier_const=1.5, window=win_size) stdev_t = pd_data.rolling(window=win_size, min_periods=2).std() @@ -790,48 +793,27 @@ def table_performance_trending_dashboard(table, input_data): if rel_change_lst[idx] > rel_change_lst[index]: index = idx - # if "regression" in classification_lst[first_idx:]: - # classification = "regression" - # elif "outlier" in classification_lst[first_idx:]: - # classification = "outlier" - # elif "progression" in classification_lst[first_idx:]: - # classification = "progression" - # elif "normal" in classification_lst[first_idx:]: - # classification = "normal" - # else: - # classification = None - # - # nr_outliers = 0 - # consecutive_outliers = 0 - # failure = False - # for item in classification_lst[first_idx:]: - # if item == "outlier": - # nr_outliers += 1 - # consecutive_outliers += 1 - # if consecutive_outliers == 3: - # failure = True - # else: - # consecutive_outliers = 0 - # - # idx = len(classification_lst) - 1 - # while idx: - # if classification_lst[idx] == classification: - # break - # idx -= 1 - # - # if failure: - # classification = "failure" - # elif classification == "outlier": - # classification = "normal" - trend = round(float(median_lst[-1]) / 1000000, 2) \ - if not isnan(median_lst[-1]) else '' + if not isnan(median_lst[-1]) else '-' sample = round(float(sample_lst[index]) / 1000000, 2) \ - if not isnan(sample_lst[index]) else '' + if not isnan(sample_lst[index]) else '-' rel_change = rel_change_lst[index] \ - if rel_change_lst[index] is not None else '' + if rel_change_lst[index] is not None else '-' + if not isnan(max_median): + if not isnan(sample_lst[index]): + long_trend_threshold = max_median * \ + (table["long-trend-threshold"] / 100) + if sample_lst[index] < long_trend_threshold: + long_trend_classification = "failure" + else: + long_trend_classification = '-' + else: + long_trend_classification = "failure" + else: + long_trend_classification = '-' tbl_lst.append([name, trend, + long_trend_classification, classification, '-' if classification == "normal" else sample, '-' if classification == "normal" else rel_change, @@ -839,10 +821,13 @@ def table_performance_trending_dashboard(table, input_data): # Sort the table according to the classification tbl_sorted = list() - for classification in ("failure", "regression", "progression", "normal"): - tbl_tmp = [item for item in tbl_lst if item[2] == classification] - tbl_tmp.sort(key=lambda rel: rel[0]) - tbl_sorted.extend(tbl_tmp) + for long_trend_class in ("failure", '-'): + tbl_long = [item for item in tbl_lst if item[2] == long_trend_class] + for classification in \ + ("failure", "regression", "progression", "normal"): + tbl_tmp = [item for item in tbl_long if item[3] == classification] + tbl_tmp.sort(key=lambda rel: rel[0]) + tbl_sorted.extend(tbl_tmp) file_name = "{0}{1}".format(table["output-file"], table["output-file-ext"]) @@ -978,7 +963,7 @@ def table_performance_trending_dashboard_html(table, input_data): ref = ET.SubElement(td, "a", attrib=dict(href=url)) ref.text = item - if c_idx == 2: + if c_idx == 3: if item == "regression": td.set("bgcolor", "#eca1a6") elif item == "failure": diff --git a/resources/tools/presentation/specification_CPTA.yaml b/resources/tools/presentation/specification_CPTA.yaml index 4e2aad6d40..77fee3506a 100644 --- a/resources/tools/presentation/specification_CPTA.yaml +++ b/resources/tools/presentation/specification_CPTA.yaml @@ -228,6 +228,8 @@ outlier-const: 1.5 window: 14 evaluated-window: 14 + long-trend-window: 180 + long-trend-threshold: 80 # Percent of the highest moving median value - type: "table" @@ -247,6 +249,8 @@ outlier-const: 1.5 window: 14 evaluated-window: 14 + long-trend-window: 180 + long-trend-threshold: 80 # Percent of the highest moving median value - type: "table" @@ -266,6 +270,8 @@ outlier-const: 1.5 window: 14 evaluated-window: 14 + long-trend-window: 180 + long-trend-threshold: 80 # Percent of the highest moving median value - type: "table" |