Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(197)

Unified Diff: tools/telemetry/telemetry/results/buildbot_output_formatter.py

Issue 1157793003: [Telemetry] Remove BuildbotOutputFormatter (Closed) Base URL: https://chromium.googlesource.com/chromium/src.git@master
Patch Set: fix test :\ Created 5 years, 7 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
Index: tools/telemetry/telemetry/results/buildbot_output_formatter.py
diff --git a/tools/telemetry/telemetry/results/buildbot_output_formatter.py b/tools/telemetry/telemetry/results/buildbot_output_formatter.py
deleted file mode 100644
index 2e39ef9f6fcc0fc47a817850d92b104ef1842b99..0000000000000000000000000000000000000000
--- a/tools/telemetry/telemetry/results/buildbot_output_formatter.py
+++ /dev/null
@@ -1,109 +0,0 @@
-# Copyright 2014 The Chromium Authors. All rights reserved.
-# Use of this source code is governed by a BSD-style license that can be
-# found in the LICENSE file.
-
-from telemetry import perf_tests_helper
-from telemetry.results import output_formatter
-from telemetry import value as value_module
-from telemetry.value import summary as summary_module
-
-
-class BuildbotOutputFormatter(output_formatter.OutputFormatter):
- def __init__(self, output_stream, trace_tag=''):
- super(BuildbotOutputFormatter, self).__init__(output_stream)
- self._trace_tag = trace_tag
-
- def _PrintPerfResult(self, measurement, trace, v, units,
- result_type='default'):
- output = perf_tests_helper.PrintPerfResult(
- measurement, trace, v, units, result_type, print_to_stdout=False)
- self.output_stream.write(output + '\n')
- self.output_stream.flush()
-
- def Format(self, page_test_results):
- """Print summary data in a format expected by buildbot for perf dashboards.
-
- If any failed pages exist, only output individual page results, and do
- not output any average data.
- """
- had_failures = len(page_test_results.failures) > 0
-
- # Print out the list of unique pages.
- perf_tests_helper.PrintPages(
- [page.display_name for page in page_test_results.pages_that_succeeded])
- summary = summary_module.Summary(page_test_results.all_page_specific_values)
- for value in summary.interleaved_computed_per_page_values_and_summaries:
- if value.page:
- self._PrintComputedPerPageValue(value)
- else:
- self._PrintComputedSummaryValue(value, had_failures)
- self._PrintOverallResults(page_test_results)
-
- def _PrintComputedPerPageValue(self, value):
- # We dont print per-page-values when there is a trace tag.
- if self._trace_tag:
- return
-
- # Actually print the result.
- buildbot_value = value.GetBuildbotValue()
- buildbot_data_type = value.GetBuildbotDataType(
- output_context=value_module.PER_PAGE_RESULT_OUTPUT_CONTEXT)
- if buildbot_value is None or buildbot_data_type is None:
- return
-
- buildbot_measurement_name, buildbot_trace_name = (
- value.GetChartAndTraceNameForPerPageResult())
- self._PrintPerfResult(buildbot_measurement_name,
- buildbot_trace_name,
- buildbot_value, value.units, buildbot_data_type)
-
- def _PrintComputedSummaryValue(self, value, had_failures):
- # If there were any page errors, we typically will print nothing.
- #
- # Note: this branch is structured less-densely to improve legibility.
- if had_failures:
- return
-
- buildbot_value = value.GetBuildbotValue()
- buildbot_data_type = value.GetBuildbotDataType(
- output_context=value_module.COMPUTED_PER_PAGE_SUMMARY_OUTPUT_CONTEXT)
- if buildbot_value is None or buildbot_data_type is None:
- return
-
- buildbot_measurement_name, buildbot_trace_name = (
- value.GetChartAndTraceNameForComputedSummaryResult(
- self._trace_tag))
- self._PrintPerfResult(buildbot_measurement_name,
- buildbot_trace_name,
- buildbot_value, value.units, buildbot_data_type)
-
- def _PrintOverallResults(self, page_test_results):
- # If there were no failed pages, output the overall results (results not
- # associated with a page).
- had_failures = len(page_test_results.failures) > 0
- if not had_failures:
- for value in page_test_results.all_summary_values:
- buildbot_value = value.GetBuildbotValue()
- buildbot_data_type = value.GetBuildbotDataType(
- output_context=value_module.SUMMARY_RESULT_OUTPUT_CONTEXT)
- buildbot_measurement_name, buildbot_trace_name = (
- value.GetChartAndTraceNameForComputedSummaryResult(
- self._trace_tag))
- self._PrintPerfResult(
- buildbot_measurement_name,
- buildbot_trace_name,
- buildbot_value,
- value.units,
- buildbot_data_type)
-
- # Print the number of failed and errored pages.
- self._PrintPerfResult('telemetry_page_measurement_results', 'num_failed',
- [len(page_test_results.failures)], 'count',
- 'unimportant')
-
- # TODO(chrishenry): Remove this in a separate patch to reduce the risk
- # of rolling back due to buildbot breakage.
- # Also fix src/tools/bisect-perf-regression_test.py when this is
- # removed.
- self._PrintPerfResult('telemetry_page_measurement_results', 'num_errored',
- [0], 'count', 'unimportant')
« no previous file with comments | « tools/telemetry/telemetry/internal/story_runner_unittest.py ('k') | tools/telemetry/telemetry/results/results_options.py » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698