Merge pull request #15 from clue-labs/benchmark

Add decoding benchmark plus benchmark for GZIP-compressed CSV files

Author: clue

composer.json

@@ -18,7 +18,8 @@
         "react/stream": "^1.0 || ^0.7 || ^0.6"
     },
     "require-dev": {
-        "react/event-loop": "^1.0 || ^0.5 || ^0.4 || ^0.3",
-        "phpunit/phpunit": "^7.0 || ^6.0 || ^5.7 || ^4.8.35"
+        "phpunit/phpunit": "^7.0 || ^6.0 || ^5.7 || ^4.8.35",
+        "react/child-process": "^0.6",
+        "react/event-loop": "^1.0 || ^0.5 || ^0.4 || ^0.3"
     }
 }

examples/91-benchmark-count.php

@@ -0,0 +1,43 @@
+<?php
+
+// simple usage:
+// $ php examples/91-benchmark-count.php < examples/users.csv
+//
+// getting reasonable results requires a large data set:
+// 1) download a large CSV data set, for example from https://github.com/fivethirtyeight/russian-troll-tweets
+// $ curl -OL https://github.com/fivethirtyeight/russian-troll-tweets/raw/master/IRAhandle_tweets_1.csv
+//
+// 2) pipe CSV into benchmark script:
+// $ php examples/91-benchmark-count.php < IRAhandle_tweets_1.csv
+
+use Clue\React\Csv\AssocDecoder;
+use React\EventLoop\Factory;
+use React\Stream\ReadableResourceStream;
+
+require __DIR__ . '/../vendor/autoload.php';
+
+if (extension_loaded('xdebug')) {
+    echo 'NOTICE: The "xdebug" extension is loaded, this has a major impact on performance.' . PHP_EOL;
+}
+
+$loop = Factory::create();
+$decoder = new AssocDecoder(new ReadableResourceStream(STDIN, $loop));
+
+$count = 0;
+$decoder->on('data', function () use (&$count) {
+    ++$count;
+});
+
+$start = microtime(true);
+$report = $loop->addPeriodicTimer(0.05, function () use (&$count, $start) {
+    printf("\r%d records in %0.3fs...", $count, microtime(true) - $start);
+});
+
+$decoder->on('close', function () use (&$count, $report, $loop, $start) {
+    $now = microtime(true);
+    $loop->cancelTimer($report);
+
+    printf("\r%d records in %0.3fs => %d records/s\n", $count, $now - $start, $count / ($now - $start));
+});
+
+$loop->run();

examples/92-benchmark-count-gzip.php

@@ -0,0 +1,55 @@
+<?php
+
+// getting reasonable results requires a large data set:
+// 1) download a large CSV data set, for example from https://github.com/fivethirtyeight/russian-troll-tweets
+// $ curl -OL https://github.com/fivethirtyeight/russian-troll-tweets/raw/master/IRAhandle_tweets_1.csv
+//
+// 2) If your data set it not already in gzip format, compress it:
+// $ gzip < IRAhandle_tweets_1.csv > IRAhandle_tweets_1.csv.gz
+//
+// 3) pipe compressed CSV into benchmark script:
+// $ php examples/92-benchmark-count-gzip.php < IRAhandle_tweets_1.csv.gz
+
+use Clue\React\Csv\AssocDecoder;
+use React\ChildProcess\Process;
+use React\EventLoop\Factory;
+
+require __DIR__ . '/../vendor/autoload.php';
+
+if (extension_loaded('xdebug')) {
+    echo 'NOTICE: The "xdebug" extension is loaded, this has a major impact on performance.' . PHP_EOL;
+}
+
+$loop = Factory::create();
+
+// This benchmark example spawns the decompressor in a child `gunzip` process
+// because parsing CSV files is already mostly CPU-bound and multi-processing
+// is preferred here. If the input source is slower (such as an HTTP download)
+// or if `gunzip` is not available (Windows), using a built-in decompressor
+// such as https://github.com/clue/reactphp-zlib would be preferable.
+$process = new Process('exec gunzip', null, null, array(
+    0 => STDIN,
+    1 => array('pipe', 'w'),
+    STDERR
+));
+$process->start($loop);
+$decoder = new AssocDecoder($process->stdout);
+
+$count = 0;
+$decoder->on('data', function () use (&$count) {
+    ++$count;
+});
+
+$start = microtime(true);
+$report = $loop->addPeriodicTimer(0.05, function () use (&$count, $start) {
+    printf("\r%d records in %0.3fs...", $count, microtime(true) - $start);
+});
+
+$decoder->on('close', function () use (&$count, $report, $loop, $start) {
+    $now = microtime(true);
+    $loop->cancelTimer($report);
+
+    printf("\r%d records in %0.3fs => %d records/s\n", $count, $now - $start, $count / ($now - $start));
+});
+
+$loop->run();