[Tarantool-patches] [PATCH v1 luajit 38/41] perf: add aggregator helper for bench statistics

[Sergey Kaplun via Tarantool-patches <tarantool-patches@dev.tarantool.org>]

This patch adds a helper script to aggregate the benchmark results from
JSON files to the format parsable by the InfluxDB line protocol [1].

All JSON files from each suite in the <perf/output> directory are
considered as the benchmark results and aggregated into the
<perf/output/summary.txt> file that can be posted to the InfluxDB. The
results are aggregated via the new target LuaJIT-perf-aggregate.

[1]: https://docs.influxdata.com/influxdb/v2/reference/syntax/line-protocol/
---
 perf/CMakeLists.txt        |  13 ++++
 perf/helpers/aggregate.lua | 124 +++++++++++++++++++++++++++++++++++++
 2 files changed, 137 insertions(+)
 create mode 100644 perf/helpers/aggregate.lua

diff --git a/perf/CMakeLists.txt b/perf/CMakeLists.txt
index cc3c312f..68e561fd 100644
--- a/perf/CMakeLists.txt
+++ b/perf/CMakeLists.txt
@@ -97,3 +97,16 @@ add_custom_target(${PROJECT_NAME}-perf
 add_custom_target(${PROJECT_NAME}-perf-console
   DEPENDS LuaJIT-benches-console
 )
+
+set(PERF_SUMMARY ${PERF_OUTPUT_DIR}/summary.txt)
+add_custom_target(${PROJECT_NAME}-perf-aggregate
+  BYPRODUCTS ${PERF_SUMMARY}
+  COMMENT "Aggregate performance test results into ${PERF_SUMMARY}"
+  COMMAND ${CMAKE_COMMAND} -E env
+    LUA_CPATH="${LUA_CPATH}"
+      ${LUAJIT_BINARY} ${CMAKE_CURRENT_SOURCE_DIR}/helpers/aggregate.lua
+        ${PERF_SUMMARY}
+        ${PERF_OUTPUT_DIR}
+  WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
+  DEPENDS luajit-main
+)
diff --git a/perf/helpers/aggregate.lua b/perf/helpers/aggregate.lua
new file mode 100644
index 00000000..12a8ab89
--- /dev/null
+++ b/perf/helpers/aggregate.lua
@@ -0,0 +1,124 @@
+local json = require('cjson')
+
+-- File to aggregate the benchmark results from JSON files to the
+-- format parsable by the InfluxDB line protocol [1]:
+-- <measurement>,<tag_set> <field_set> <timestamp>
+--
+-- <tag_set> and <field_set> have the following format:
+-- <key1>=<value1>,<key2>=<value2>
+--
+-- The reported tag set is a set of values that can be used for
+-- filtering data (i.e., branch or benchmark name).
+--
+-- luacheck: push no max comment line length
+--
+-- [1]: https://docs.influxdata.com/influxdb/v2/reference/syntax/line-protocol/
+--
+-- luacheck: pop
+
+local output = assert(arg[1], 'Output file is required as the first argument')
+local input_dir = arg[2] or '.'
+
+local out_fh = assert(io.open(output, 'w+'))
+
+local function exec(cmd)
+  return io.popen(cmd):read('*all'):gsub('%s+$', '')
+end
+
+local commit = os.getenv('PERF_COMMIT') or exec('git rev-parse --short HEAD')
+assert(commit, 'can not determine the commit')
+
+local branch = os.getenv('PERF_BRANCH') or
+  exec('git rev-parse --abbrev-ref HEAD')
+assert(branch, 'can not determine the branch')
+
+-- Not very robust, but OK for our needs.
+local function listdir(path)
+  local handle = io.popen('ls -1 ' .. path)
+
+  local files = {}
+  for file in handle:lines() do
+    table.insert(files, file)
+  end
+
+  return files
+end
+
+local tag_set = {branch = branch}
+
+local function table_plain_copy(src)
+  local dst = {}
+  for k, v in pairs(src) do
+    dst[k] = v
+  end
+  return dst
+end
+
+local function read_all(file)
+  local fh = assert(io.open(file, 'rb'))
+  local content = fh:read('*all')
+  fh:close()
+  return content
+end
+
+local REPORTED_FIELDS = {
+  'cpu_time',
+  'items_per_second',
+  'iterations',
+  'real_time',
+}
+
+local function influx_kv(tab)
+  local kv_string = {}
+  for k, v in pairs(tab) do
+    table.insert(kv_string, ('%s=%s'):format(k, v))
+  end
+  return table.concat(kv_string, ',')
+end
+
+local time = os.time()
+local function influx_line(measurement, tags, fields)
+  return ('%s,%s %s %d\n'):format(measurement, influx_kv(tags),
+          influx_kv(fields), time)
+end
+
+for _, suite_name in pairs(listdir(input_dir)) do
+  -- May list the report file, but will be ignored by the
+  -- condition below.
+  local suite_dir = ('%s/%s'):format(input_dir, suite_name)
+  for _, file in pairs(listdir(suite_dir)) do
+    -- Skip files in which we are not interested.
+    if not file:match('%.json$') then goto continue end
+
+    local data = read_all(('%s/%s'):format(suite_dir, file))
+    local bench_name = file:match('([^/]+)%.json')
+    local bench_data = json.decode(data)
+    local benchmarks = bench_data.benchmarks
+    local arch = bench_data.context.arch
+    local gc64 = bench_data.context.gc64
+    local jit = bench_data.context.jit
+
+    for _, bench in ipairs(benchmarks) do
+      local full_tag_set = table_plain_copy(tag_set)
+      full_tag_set.name = bench.name
+      full_tag_set.suite = suite_name
+      full_tag_set.arch = arch
+      full_tag_set.gc64 = gc64
+      full_tag_set.jit = jit
+
+      -- Save the commit as a field, since we don't want to filter
+      -- benchmarks by the commit (one point of data).
+      local field_set = {commit = ('"%s"'):format(commit)}
+
+      for _, field in ipairs(REPORTED_FIELDS) do
+          field_set[field] = bench[field]
+      end
+
+      local line = influx_line(bench_name, full_tag_set, field_set)
+      out_fh:write(line)
+    end
+    ::continue::
+  end
+end
+
+out_fh:close()
-- 
2.51.0