From 90ee601eb0bcda1944c2603d4a807a87842d3686 Mon Sep 17 00:00:00 2001 From: Alexander Kuzmenkov Date: Thu, 27 Feb 2020 20:57:08 +0300 Subject: [PATCH] performance comparison --- docker/test/performance-comparison/compare.sh | 4 +- .../test/performance-comparison/entrypoint.sh | 11 ++ docker/test/performance-comparison/report.py | 159 ++++++++++-------- 3 files changed, 105 insertions(+), 69 deletions(-) diff --git a/docker/test/performance-comparison/compare.sh b/docker/test/performance-comparison/compare.sh index 2680464c54a..3e5b6111288 100755 --- a/docker/test/performance-comparison/compare.sh +++ b/docker/test/performance-comparison/compare.sh @@ -369,7 +369,7 @@ create table metric_devation engine File(TSVWithNamesAndTypes, 'metric-deviation quantilesExact(0, 0.5, 1)(value) q, metric, query from (select * from unstable_run_metrics union all select * from unstable_run_traces - union all select * from unstable_run_metrics_2) + union all select * from unstable_run_metrics_2) mm join queries using query group by query, metric having d > 0.5 @@ -406,7 +406,7 @@ unset IFS # Remember that grep sets error code when nothing is found, hence the bayan # operator -grep Exception:[^:] *-err.log > run-errors.log ||: +grep -m2 Exception:[^:] *-err.log | sed 's/:/\t/' > run-errors.tsv ||: $script_dir/report.py > report.html } diff --git a/docker/test/performance-comparison/entrypoint.sh b/docker/test/performance-comparison/entrypoint.sh index 3fc79f7a110..36e6a393b28 100755 --- a/docker/test/performance-comparison/entrypoint.sh +++ b/docker/test/performance-comparison/entrypoint.sh @@ -59,6 +59,17 @@ set -m time ../compare.sh 0 $ref_sha $PR_TO_TEST $SHA_TO_TEST 2>&1 | ts "$(printf '%%Y-%%m-%%d %%H:%%M:%%S\t')" | tee compare.log set +m +# Stop the servers to free memory. Normally they are restarted before getting +# the profile info, so they shouldn't use much, but if the comparison script +# fails in the middle, this might not be the case. +for i in {1..30} +do + if ! killall clickhouse + then + break + fi +done + dmesg -T > dmesg.log 7z a /output/output.7z *.log *.tsv *.html *.txt *.rep *.svg {right,left}/db/preprocessed_configs diff --git a/docker/test/performance-comparison/report.py b/docker/test/performance-comparison/report.py index a051d175cf3..1a0de462a89 100755 --- a/docker/test/performance-comparison/report.py +++ b/docker/test/performance-comparison/report.py @@ -2,6 +2,7 @@ import collections import csv +import itertools import os import sys import traceback @@ -9,6 +10,8 @@ import traceback report_errors = [] status = 'success' message = 'See the report' +long_tests = 0 +run_errors = 0 print(""" @@ -37,6 +40,7 @@ a:hover, a:active {{ color: #F40; text-decoration: underline; }} table {{ border: 0; }} .main {{ margin-left: 10%; }} p.links a {{ padding: 5px; margin: 3px; background: #FFF; line-height: 2; white-space: nowrap; box-shadow: 0 0 0 1px rgba(0, 0, 0, 0.05), 0 8px 25px -5px rgba(0, 0, 0, 0.1); }} + .cancela,.cancela:link,.cancela:visited,.cancela:hover,.cancela:focus,.cancela:active{{ color: inherit; text-decoration: none; @@ -51,15 +55,6 @@ tr:nth-child(odd) td {{filter: brightness(95%);}}

ClickHouse performance comparison

""".format()) - -table_template = """ -

{caption}

- -{header} -{rows} -
-""" - table_anchor = 0 row_anchor = 0 @@ -79,16 +74,19 @@ def tr(x): #return '{x}'.format(a=a, x=str(x)) return '{x}'.format(a=a, x=str(x)) -def td(value, cell_class = ''): - return '{value}'.format( - cell_class = cell_class, +def td(value, cell_attributes = ''): + return '{value}'.format( + cell_attributes = cell_attributes, value = value) def th(x): return '' + str(x) + '' -def tableRow(r): - return tr(''.join([td(f) for f in r])) +def tableRow(cell_values, cell_attributes = []): + return tr(''.join([td(v, a) + for v, a in itertools.zip_longest( + cell_values, cell_attributes, + fillvalue = '')])) def tableHeader(r): return tr(''.join([th(f) for f in r])) @@ -103,7 +101,7 @@ def tableStart(title): def tableEnd(): return '' -def tsvRowsRaw(n): +def tsvRows(n): result = [] try: with open(n, encoding='utf-8') as fd: @@ -115,64 +113,81 @@ def tsvRowsRaw(n): pass return [] -def tsvTableRows(n): - rawRows = tsvRowsRaw(n) +def htmlRows(n): + rawRows = tsvRows(n) result = '' for row in rawRows: result += tableRow(row) return result -print(table_template.format( - anchor = nextTableAnchor(), - caption = 'Tested commits', - header = tableHeader(['Old', 'New']), - rows = tableRow([open('left-commit.txt').read(), open('right-commit.txt').read()]))) - -print(table_template.format( - anchor = nextTableAnchor(), - caption = 'Changes in performance', - header = tableHeader(['Old, s', 'New, s', - 'Relative difference (new - old)/old', - 'Randomization distribution quantiles [5%, 50%, 95%]', - 'Test', 'Query']), - rows = tsvTableRows('changed-perf.tsv'))) - -print(table_template.format( - anchor = nextTableAnchor(), - caption = 'Slow on client', - header = tableHeader(['Client time, s', 'Server time, s', 'Ratio', 'Query']), - rows = tsvTableRows('slow-on-client.tsv'))) - -print(table_template.format( - anchor = nextTableAnchor(), - caption = 'Unstable queries', - header = tableHeader(['Old, s', 'New, s', - 'Relative difference (new - old)/old', - 'Randomization distribution quantiles [5%, 50%, 95%]', - 'Test', 'Query']), - rows = tsvTableRows('unstable-queries.tsv'))) - -print(table_template.format( - anchor = nextTableAnchor(), - caption = 'Run errors', - header = tableHeader(['A', 'B']), - rows = tsvTableRows('run-errors.log'))) - -print(table_template.format( - anchor = nextTableAnchor(), - caption = 'Tests with most unstable queries', - header = tableHeader(['Test', 'Unstable', 'Changed perf', 'Total not OK']), - rows = tsvTableRows('bad-tests.tsv'))) - -def print_test_times(): - rows = tsvRowsRaw('test-times.tsv') +def printSimpleTable(caption, columns, rows): if not rows: return - print(rows, file=sys.stderr) + print(tableStart(caption)) + print(tableHeader(columns)) + for row in rows: + print(tableRow(row)) + print(tableEnd()) - print(tableStart('Test times')) - print(tableHeader([ +printSimpleTable('Tested commits', ['Old', 'New'], + [[open('left-commit.txt').read(), open('right-commit.txt').read()]]) + +def print_changes(): + rows = tsvRows('changed-perf.tsv') + if not rows: + return + + print(tableStart('Changes in performance')) + columns = [ + 'Old, s', # 0 + 'New, s', # 1 + 'Relative difference (new - old)/old', # 2 + 'Randomization distribution quantiles [5%, 50%, 95%]', # 3 + 'Test', # 4 + 'Query', # 5 + ] + + print(tableHeader(columns)) + + attrs = ['' for c in columns] + for row in rows: + if float(row[2]) > 0.: + attrs[2] = 'style="background: #adbdff"' + else: + attrs[2] = 'style="background: #ffb0a0"' + + print(tableRow(row, attrs)) + + print(tableEnd()) + +print_changes() + +printSimpleTable('Slow on client', + ['Client time, s', 'Server time, s', 'Ratio', 'Query'], + tsvRows('slow-on-client.tsv')) + +printSimpleTable('Unstable queries', + [ + 'Old, s', 'New, s', 'Relative difference (new - old)/old', + 'Randomization distribution quantiles [5%, 50%, 95%]', + 'Test', 'Query' + ], + tsvRows('unstable-queries.tsv')) + +printSimpleTable('Run errors', ['Test', 'Error'], tsvRows('run-errors.tsv')) + +printSimpleTable('Tests with most unstable queries', + ['Test', 'Unstable', 'Changed perf', 'Total not OK'], + tsvRows('bad-tests.tsv')) + +def print_test_times(): + global long_tests + rows = tsvRows('test-times.tsv') + if not rows: + return + + columns = [ 'Test', #0 'Wall clock time, s', #1 'Total client time, s', #2 @@ -181,10 +196,20 @@ def print_test_times(): 'Longest query
(sum for all runs), s', #5 'Avg wall clock time
(sum for all runs), s', #6 'Shortest query
(sum for all runs), s', #7 - ])) + ] + print(tableStart('Test times')) + print(tableHeader(columns)) + + attrs = ['' for c in columns] for r in rows: - print(tableRow(r)) + if float(r[6]) > 10: + long_tests += 1 + attrs[6] = 'style="background: #ffb0a0"' + else: + attrs[6] = '' + + print(tableRow(r, attrs)) print(tableEnd()) @@ -207,7 +232,7 @@ print(""" """) if report_errors: - status = 'error' + status = 'failure' message = 'Errors while building the report.' print("""