aboutsummaryrefslogtreecommitdiffstats
path: root/csit.infra.dash/app/pal/news
diff options
context:
space:
mode:
authorpmikus <peter.mikus@protonmail.ch>2022-09-19 08:49:01 +0200
committerPeter Mikus <peter.mikus@protonmail.ch>2022-09-19 06:54:43 +0000
commitd6a60b5043c6f7c3dfc45853feb68d0aca5a4a5f (patch)
tree022343584d4e00a0da8ef12eb4e713d67344fc95 /csit.infra.dash/app/pal/news
parentd2ef7bc01df66f6a27f25d061db064cf4a463267 (diff)
feat(uti): Move directory
Signed-off-by: pmikus <peter.mikus@protonmail.ch> Change-Id: I7300ecfe756baaf3fbeedb020070f882cfaca445
Diffstat (limited to 'csit.infra.dash/app/pal/news')
-rw-r--r--csit.infra.dash/app/pal/news/__init__.py12
-rw-r--r--csit.infra.dash/app/pal/news/layout.py522
-rw-r--r--csit.infra.dash/app/pal/news/news.py46
-rw-r--r--csit.infra.dash/app/pal/news/tables.py176
4 files changed, 756 insertions, 0 deletions
diff --git a/csit.infra.dash/app/pal/news/__init__.py b/csit.infra.dash/app/pal/news/__init__.py
new file mode 100644
index 0000000000..5692432123
--- /dev/null
+++ b/csit.infra.dash/app/pal/news/__init__.py
@@ -0,0 +1,12 @@
+# Copyright (c) 2022 Cisco and/or its affiliates.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at:
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/csit.infra.dash/app/pal/news/layout.py b/csit.infra.dash/app/pal/news/layout.py
new file mode 100644
index 0000000000..cd1618d719
--- /dev/null
+++ b/csit.infra.dash/app/pal/news/layout.py
@@ -0,0 +1,522 @@
+# Copyright (c) 2022 Cisco and/or its affiliates.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at:
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Plotly Dash HTML layout override.
+"""
+
+import logging
+import pandas as pd
+import dash_bootstrap_components as dbc
+
+from flask import Flask
+from dash import dcc
+from dash import html
+from dash import callback_context
+from dash import Input, Output
+from yaml import load, FullLoader, YAMLError
+
+from ..data.data import Data
+from ..utils.constants import Constants as C
+from ..utils.utils import classify_anomalies, show_tooltip, gen_new_url
+from ..utils.url_processing import url_decode
+from ..data.data import Data
+from .tables import table_summary
+
+
+class Layout:
+ """The layout of the dash app and the callbacks.
+ """
+
+ def __init__(self, app: Flask, html_layout_file: str, data_spec_file: str,
+ tooltip_file: str) -> None:
+ """Initialization:
+ - save the input parameters,
+ - read and pre-process the data,
+ - prepare data for the control panel,
+ - read HTML layout file,
+ - read tooltips from the tooltip file.
+
+ :param app: Flask application running the dash application.
+ :param html_layout_file: Path and name of the file specifying the HTML
+ layout of the dash application.
+ :param data_spec_file: Path and name of the file specifying the data to
+ be read from parquets for this application.
+ :param tooltip_file: Path and name of the yaml file specifying the
+ tooltips.
+ :type app: Flask
+ :type html_layout_file: str
+ :type data_spec_file: str
+ :type tooltip_file: str
+ """
+
+ # Inputs
+ self._app = app
+ self._html_layout_file = html_layout_file
+ self._data_spec_file = data_spec_file
+ self._tooltip_file = tooltip_file
+
+ # Read the data:
+ data_stats, data_mrr, data_ndrpdr = Data(
+ data_spec_file=self._data_spec_file,
+ debug=True
+ ).read_stats(days=C.NEWS_TIME_PERIOD)
+
+ df_tst_info = pd.concat([data_mrr, data_ndrpdr], ignore_index=True)
+
+ # Prepare information for the control panel:
+ self._jobs = sorted(list(df_tst_info["job"].unique()))
+ d_job_info = {
+ "job": list(),
+ "dut": list(),
+ "ttype": list(),
+ "cadence": list(),
+ "tbed": list()
+ }
+ for job in self._jobs:
+ lst_job = job.split("-")
+ d_job_info["job"].append(job)
+ d_job_info["dut"].append(lst_job[1])
+ d_job_info["ttype"].append(lst_job[3])
+ d_job_info["cadence"].append(lst_job[4])
+ d_job_info["tbed"].append("-".join(lst_job[-2:]))
+ self.job_info = pd.DataFrame.from_dict(d_job_info)
+
+ # Pre-process the data:
+
+ def _create_test_name(test: str) -> str:
+ lst_tst = test.split(".")
+ suite = lst_tst[-2].replace("2n1l-", "").replace("1n1l-", "").\
+ replace("2n-", "")
+ return f"{suite.split('-')[0]}-{lst_tst[-1]}"
+
+ def _get_rindex(array: list, itm: any) -> int:
+ return len(array) - 1 - array[::-1].index(itm)
+
+ tst_info = {
+ "job": list(),
+ "build": list(),
+ "start": list(),
+ "dut_type": list(),
+ "dut_version": list(),
+ "hosts": list(),
+ "failed": list(),
+ "regressions": list(),
+ "progressions": list()
+ }
+ for job in self._jobs:
+ # Create lists of failed tests:
+ df_job = df_tst_info.loc[(df_tst_info["job"] == job)]
+ last_build = str(max(pd.to_numeric(df_job["build"].unique())))
+ df_build = df_job.loc[(df_job["build"] == last_build)]
+ tst_info["job"].append(job)
+ tst_info["build"].append(last_build)
+ tst_info["start"].append(data_stats.loc[
+ (data_stats["job"] == job) &
+ (data_stats["build"] == last_build)
+ ]["start_time"].iloc[-1].strftime('%Y-%m-%d %H:%M'))
+ tst_info["dut_type"].append(df_build["dut_type"].iloc[-1])
+ tst_info["dut_version"].append(df_build["dut_version"].iloc[-1])
+ tst_info["hosts"].append(df_build["hosts"].iloc[-1])
+ failed_tests = df_build.loc[(df_build["passed"] == False)]\
+ ["test_id"].to_list()
+ l_failed = list()
+ try:
+ for tst in failed_tests:
+ l_failed.append(_create_test_name(tst))
+ except KeyError:
+ l_failed = list()
+ tst_info["failed"].append(sorted(l_failed))
+
+ # Create lists of regressions and progressions:
+ l_reg = list()
+ l_prog = list()
+
+ tests = df_job["test_id"].unique()
+ for test in tests:
+ tst_data = df_job.loc[df_job["test_id"] == test].sort_values(
+ by="start_time", ignore_index=True)
+ x_axis = tst_data["start_time"].tolist()
+ if "-ndrpdr" in test:
+ tst_data = tst_data.dropna(
+ subset=["result_pdr_lower_rate_value", ]
+ )
+ if tst_data.empty:
+ continue
+ try:
+ anomalies, _, _ = classify_anomalies({
+ k: v for k, v in zip(
+ x_axis,
+ tst_data["result_ndr_lower_rate_value"].tolist()
+ )
+ })
+ except ValueError:
+ continue
+ if "progression" in anomalies:
+ l_prog.append((
+ _create_test_name(test).replace("-ndrpdr", "-ndr"),
+ x_axis[_get_rindex(anomalies, "progression")]
+ ))
+ if "regression" in anomalies:
+ l_reg.append((
+ _create_test_name(test).replace("-ndrpdr", "-ndr"),
+ x_axis[_get_rindex(anomalies, "regression")]
+ ))
+ try:
+ anomalies, _, _ = classify_anomalies({
+ k: v for k, v in zip(
+ x_axis,
+ tst_data["result_pdr_lower_rate_value"].tolist()
+ )
+ })
+ except ValueError:
+ continue
+ if "progression" in anomalies:
+ l_prog.append((
+ _create_test_name(test).replace("-ndrpdr", "-pdr"),
+ x_axis[_get_rindex(anomalies, "progression")]
+ ))
+ if "regression" in anomalies:
+ l_reg.append((
+ _create_test_name(test).replace("-ndrpdr", "-pdr"),
+ x_axis[_get_rindex(anomalies, "regression")]
+ ))
+ else: # mrr
+ tst_data = tst_data.dropna(
+ subset=["result_receive_rate_rate_avg", ]
+ )
+ if tst_data.empty:
+ continue
+ try:
+ anomalies, _, _ = classify_anomalies({
+ k: v for k, v in zip(
+ x_axis,
+ tst_data["result_receive_rate_rate_avg"].\
+ tolist()
+ )
+ })
+ except ValueError:
+ continue
+ if "progression" in anomalies:
+ l_prog.append((
+ _create_test_name(test),
+ x_axis[_get_rindex(anomalies, "progression")]
+ ))
+ if "regression" in anomalies:
+ l_reg.append((
+ _create_test_name(test),
+ x_axis[_get_rindex(anomalies, "regression")]
+ ))
+
+ tst_info["regressions"].append(
+ sorted(l_reg, key=lambda k: k[1], reverse=True))
+ tst_info["progressions"].append(
+ sorted(l_prog, key=lambda k: k[1], reverse=True))
+
+ self._data = pd.DataFrame.from_dict(tst_info)
+
+ # Read from files:
+ self._html_layout = str()
+ self._tooltips = dict()
+
+ try:
+ with open(self._html_layout_file, "r") as file_read:
+ self._html_layout = file_read.read()
+ except IOError as err:
+ raise RuntimeError(
+ f"Not possible to open the file {self._html_layout_file}\n{err}"
+ )
+
+ try:
+ with open(self._tooltip_file, "r") as file_read:
+ self._tooltips = load(file_read, Loader=FullLoader)
+ except IOError as err:
+ logging.warning(
+ f"Not possible to open the file {self._tooltip_file}\n{err}"
+ )
+ except YAMLError as err:
+ logging.warning(
+ f"An error occurred while parsing the specification file "
+ f"{self._tooltip_file}\n{err}"
+ )
+
+ self._default_period = C.NEWS_SHORT
+ self._default_active = (False, True, False)
+ self._default_table = \
+ table_summary(self._data, self._jobs, self._default_period)
+
+ # Callbacks:
+ if self._app is not None and hasattr(self, 'callbacks'):
+ self.callbacks(self._app)
+
+ @property
+ def html_layout(self) -> dict:
+ return self._html_layout
+
+ def add_content(self):
+ """Top level method which generated the web page.
+
+ It generates:
+ - Store for user input data,
+ - Navigation bar,
+ - Main area with control panel and ploting area.
+
+ If no HTML layout is provided, an error message is displayed instead.
+
+ :returns: The HTML div with the whole page.
+ :rtype: html.Div
+ """
+
+ if self.html_layout:
+ return html.Div(
+ id="div-main",
+ className="small",
+ children=[
+ dcc.Location(id="url", refresh=False),
+ dbc.Row(
+ id="row-navbar",
+ class_name="g-0",
+ children=[
+ self._add_navbar(),
+ ]
+ ),
+ dbc.Row(
+ id="row-main",
+ class_name="g-0",
+ children=[
+ self._add_ctrl_col(),
+ self._add_plotting_col(),
+ ]
+ )
+ ]
+ )
+ else:
+ return html.Div(
+ id="div-main-error",
+ children=[
+ dbc.Alert(
+ [
+ "An Error Occured",
+ ],
+ color="danger",
+ ),
+ ]
+ )
+
+ def _add_navbar(self):
+ """Add nav element with navigation panel. It is placed on the top.
+
+ :returns: Navigation bar.
+ :rtype: dbc.NavbarSimple
+ """
+
+ return dbc.NavbarSimple(
+ id="navbarsimple-main",
+ children=[
+ dbc.NavItem(
+ dbc.NavLink(
+ "Continuous Performance News",
+ disabled=True,
+ external_link=True,
+ href="#"
+ )
+ )
+ ],
+ brand="Dashboard",
+ brand_href="/",
+ brand_external_link=True,
+ class_name="p-2",
+ fluid=True,
+ )
+
+ def _add_ctrl_col(self) -> dbc.Col:
+ """Add column with control panel. It is placed on the left side.
+
+ :returns: Column with the control panel.
+ :rtype: dbc.Col
+ """
+
+ return dbc.Col(
+ id="col-controls",
+ children=[
+ self._add_ctrl_panel(),
+ ],
+ )
+
+ def _add_plotting_col(self) -> dbc.Col:
+ """Add column with tables. It is placed on the right side.
+
+ :returns: Column with tables.
+ :rtype: dbc.Col
+ """
+
+ return dbc.Col(
+ id="col-plotting-area",
+ children=[
+ dcc.Loading(
+ children=[
+ dbc.Row( # Failed tests
+ id="row-table",
+ class_name="g-0 p-2",
+ children=self._default_table
+ ),
+ dbc.Row(
+ class_name="g-0 p-2",
+ align="center",
+ justify="start",
+ children=[
+ dbc.InputGroup(
+ class_name="me-1",
+ children=[
+ dbc.InputGroupText(
+ style=C.URL_STYLE,
+ children=show_tooltip(
+ self._tooltips,
+ "help-url", "URL",
+ "input-url"
+ )
+ ),
+ dbc.Input(
+ id="input-url",
+ readonly=True,
+ type="url",
+ style=C.URL_STYLE,
+ value=""
+ )
+ ]
+ )
+ ]
+ )
+ ]
+ )
+ ],
+ width=9,
+ )
+
+ def _add_ctrl_panel(self) -> dbc.Row:
+ """Add control panel.
+
+ :returns: Control panel.
+ :rtype: dbc.Row
+ """
+ return dbc.Row(
+ id="row-ctrl-panel",
+ class_name="g-0",
+ children=[
+ dbc.Row(
+ class_name="g-0 p-2",
+ children=[
+ dbc.Row(
+ class_name="g-0",
+ children=[
+ dbc.Label(
+ class_name="g-0",
+ children=show_tooltip(self._tooltips,
+ "help-summary-period", "Window")
+ ),
+ dbc.Row(
+ dbc.ButtonGroup(
+ id="bg-time-period",
+ class_name="g-0",
+ children=[
+ dbc.Button(
+ id="period-last",
+ children="Last Run",
+ className="me-1",
+ outline=True,
+ color="info"
+ ),
+ dbc.Button(
+ id="period-short",
+ children=\
+ f"Last {C.NEWS_SHORT} Runs",
+ className="me-1",
+ outline=True,
+ active=True,
+ color="info"
+ ),
+ dbc.Button(
+ id="period-long",
+ children="All Runs",
+ className="me-1",
+ outline=True,
+ color="info"
+ )
+ ]
+ )
+ )
+ ]
+ )
+ ]
+ )
+ ]
+ )
+
+ def callbacks(self, app):
+ """Callbacks for the whole application.
+
+ :param app: The application.
+ :type app: Flask
+ """
+
+ @app.callback(
+ Output("row-table", "children"),
+ Output("input-url", "value"),
+ Output("period-last", "active"),
+ Output("period-short", "active"),
+ Output("period-long", "active"),
+ Input("period-last", "n_clicks"),
+ Input("period-short", "n_clicks"),
+ Input("period-long", "n_clicks"),
+ Input("url", "href")
+ )
+ def _update_application(btn_last: int, btn_short: int, btn_long: int,
+ href: str) -> tuple:
+ """Update the application when the event is detected.
+
+ :returns: New values for web page elements.
+ :rtype: tuple
+ """
+
+ _, _, _ = btn_last, btn_short, btn_long
+
+ periods = {
+ "period-last": C.NEWS_LAST,
+ "period-short": C.NEWS_SHORT,
+ "period-long": C.NEWS_LONG
+ }
+ actives = {
+ "period-last": (True, False, False),
+ "period-short": (False, True, False),
+ "period-long": (False, False, True)
+ }
+
+ # Parse the url:
+ parsed_url = url_decode(href)
+ if parsed_url:
+ url_params = parsed_url["params"]
+ else:
+ url_params = None
+
+ trigger_id = callback_context.triggered[0]["prop_id"].split(".")[0]
+ if trigger_id == "url" and url_params:
+ trigger_id = url_params.get("period", list())[0]
+
+ period = periods.get(trigger_id, self._default_period)
+ active = actives.get(trigger_id, self._default_active)
+
+ ret_val = [
+ table_summary(self._data, self._jobs, period),
+ gen_new_url(parsed_url, {"period": trigger_id})
+ ]
+ ret_val.extend(active)
+ return ret_val
diff --git a/csit.infra.dash/app/pal/news/news.py b/csit.infra.dash/app/pal/news/news.py
new file mode 100644
index 0000000000..a0d05f1483
--- /dev/null
+++ b/csit.infra.dash/app/pal/news/news.py
@@ -0,0 +1,46 @@
+# Copyright (c) 2022 Cisco and/or its affiliates.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at:
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Instantiate the News Dash application.
+"""
+import dash
+
+from ..utils.constants import Constants as C
+from .layout import Layout
+
+
+def init_news(server):
+ """Create a Plotly Dash dashboard.
+
+ :param server: Flask server.
+ :type server: Flask
+ :returns: Dash app server.
+ :rtype: Dash
+ """
+
+ dash_app = dash.Dash(
+ server=server,
+ routes_pathname_prefix=C.NEWS_ROUTES_PATHNAME_PREFIX,
+ external_stylesheets=C.EXTERNAL_STYLESHEETS
+ )
+
+ layout = Layout(
+ app=dash_app,
+ html_layout_file=C.NEWS_HTML_LAYOUT_FILE,
+ data_spec_file=C.DATA_SPEC_FILE,
+ tooltip_file=C.TOOLTIP_FILE,
+ )
+ dash_app.index_string = layout.html_layout
+ dash_app.layout = layout.add_content()
+
+ return dash_app.server
diff --git a/csit.infra.dash/app/pal/news/tables.py b/csit.infra.dash/app/pal/news/tables.py
new file mode 100644
index 0000000000..7c0cc66eda
--- /dev/null
+++ b/csit.infra.dash/app/pal/news/tables.py
@@ -0,0 +1,176 @@
+# Copyright (c) 2022 Cisco and/or its affiliates.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at:
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""The tables with news.
+"""
+
+import pandas as pd
+import dash_bootstrap_components as dbc
+
+from datetime import datetime, timedelta
+
+
+def _table_info(job_data: pd.DataFrame) -> dbc.Table:
+ """Generates table with info about the job.
+
+ :param job_data: Dataframe with information about the job.
+ :type job_data: pandas.DataFrame
+ :returns: Table with job info.
+ :rtype: dbc.Table
+ """
+ return dbc.Table.from_dataframe(
+ pd.DataFrame.from_dict(
+ {
+ "Job": job_data["job"],
+ "Last Build": job_data["build"],
+ "Date": job_data["start"],
+ "DUT": job_data["dut_type"],
+ "DUT Version": job_data["dut_version"],
+ "Hosts": ", ".join(job_data["hosts"].to_list()[0])
+ }
+ ),
+ bordered=True,
+ striped=True,
+ hover=True,
+ size="sm",
+ color="info"
+ )
+
+
+def _table_failed(job_data: pd.DataFrame, failed: list) -> dbc.Table:
+ """Generates table with failed tests from the last run of the job.
+
+ :param job_data: Dataframe with information about the job.
+ :param failed: List of failed tests.
+ :type job_data: pandas.DataFrame
+ :type failed: list
+ :returns: Table with fialed tests.
+ :rtype: dbc.Table
+ """
+ return dbc.Table.from_dataframe(
+ pd.DataFrame.from_dict(
+ {
+ (
+ f"Last Failed Tests on "
+ f"{job_data['start'].values[0]} ({len(failed)})"
+ ): failed
+ }
+ ),
+ bordered=True,
+ striped=True,
+ hover=True,
+ size="sm",
+ color="danger"
+ )
+
+
+def _table_gressions(itms: dict, color: str) -> dbc.Table:
+ """Generates table with regressions.
+
+ :param itms: Dictionary with items (regressions or progressions) and their
+ last occurence.
+ :param color: Color of the table.
+ :type regressions: dict
+ :type color: str
+ :returns: The table with regressions.
+ :rtype: dbc.Table
+ """
+ return dbc.Table.from_dataframe(
+ pd.DataFrame.from_dict(itms),
+ bordered=True,
+ striped=True,
+ hover=True,
+ size="sm",
+ color=color
+ )
+
+
+def table_news(data: pd.DataFrame, job: str, period: int) -> list:
+ """Generates the tables with news:
+ 1. Falied tests from the last run
+ 2. Regressions and progressions calculated from the last C.NEWS_TIME_PERIOD
+ days.
+
+ :param data: Trending data with calculated annomalies to be displayed in the
+ tables.
+ :param job: The job name.
+ :param period: The time period (nr of days from now) taken into account.
+ :type data: pandas.DataFrame
+ :type job: str
+ :type period: int
+ :returns: List of tables.
+ :rtype: list
+ """
+
+ last_day = datetime.utcnow() - timedelta(days=period)
+ r_list = list()
+ job_data = data.loc[(data["job"] == job)]
+ r_list.append(_table_info(job_data))
+
+ failed = job_data["failed"].to_list()[0]
+ if failed:
+ r_list.append(_table_failed(job_data, failed))
+
+ title = f"Regressions in the last {period} days"
+ regressions = {title: list(), "Last Regression": list()}
+ for itm in job_data["regressions"].to_list()[0]:
+ if itm[1] < last_day:
+ break
+ regressions[title].append(itm[0])
+ regressions["Last Regression"].append(
+ itm[1].strftime('%Y-%m-%d %H:%M'))
+ if regressions["Last Regression"]:
+ r_list.append(_table_gressions(regressions, "warning"))
+
+ title = f"Progressions in the last {period} days"
+ progressions = {title: list(), "Last Progression": list()}
+ for itm in job_data["progressions"].to_list()[0]:
+ if itm[1] < last_day:
+ break
+ progressions[title].append(itm[0])
+ progressions["Last Progression"].append(
+ itm[1].strftime('%Y-%m-%d %H:%M'))
+ if progressions["Last Progression"]:
+ r_list.append(_table_gressions(progressions, "success"))
+
+ return r_list
+
+
+def table_summary(data: pd.DataFrame, jobs: list, period: int) -> list:
+ """Generates summary (failed tests, regressions and progressions) from the
+ last week.
+
+ :param data: Trending data with calculated annomalies to be displayed in the
+ tables.
+ :param jobs: List of jobs.
+ :params period: The time period for the summary table.
+ :type data: pandas.DataFrame
+ :type job: str
+ :type period: int
+ :returns: List of tables.
+ :rtype: list
+ """
+
+ return [
+ dbc.Accordion(
+ children=[
+ dbc.AccordionItem(
+ title=job,
+ children=table_news(data, job, period)
+ ) for job in jobs
+ ],
+ class_name="gy-2 p-0",
+ start_collapsed=True,
+ always_open=True
+ )
+ ]