Few question about Tables

[Draghmar]

Hi. I wanted to check some things on how Tables works and how I could use them. So I made this rough code:

	use Swoole\Coroutine;
	use Swoole\Runtime;
	use Swoole\Coroutine\WaitGroup;
	use function Swoole\Coroutine\run;
	Swoole\Runtime::enableCoroutine(SWOOLE_HOOK_ALL);

	$maxTbl = 10;
	$n = 5;
	$split = $maxTbl / $n;
	$tbls = [];
	for(;$n > 0; $n--) {
		$tbl = new Swoole\Table($split);
		$tbl->column('date', Swoole\Table::TYPE_INT);
		$tbl->column('data', Swoole\Table::TYPE_INT);
		$tbl->create();
		for($i = 1; $i <= $split; $i++) {
			@$tbl->set(md5($n * $i), ['date'=>time(), 'data'=>($n * $i)]);
		}
		$tbls[] = $tbl;
	}

run(function() use ($tbls, $n) {
	try {
		$find = 3;
		$check = false;
		$wg = new WaitGroup;
		foreach($tbls as $n => $tbl) {
			$wg->add();
			go(function() use ($wg, $tbl, $n, $find, &$check) {
				foreach($tbl as $i => $el) {
					if($check) {
						$wg->done();
						return;
					}
					if($el['data'] == $find) {
						$check = $i;
						$wg->done();
						return;
					}
					sleep(1);
				}
				$wg->done();
			});
		}
		$wg->wait();

		var_dump($check);
	} catch(Throwable $e) {
		var_dump($e);
	}
});

The goal here was to split Table into smaller ones to find relevant row faster when there are hundred of thousands of rows and you need to loop each time to get to the data field. After that I wanted to use additional Table for mapping between Table#1 keys and some additional ID.

First thing I learned was that there's a lower limit for rows then what's set up in the Table constructor. I guess I get why but I wonder if it's possible to predict it in any way? Or should I simply use isFull()?

Because I wanted to see if there's any benefit to this splitting, I used sleep(1) there but...adding it crashed with dumping core. And I don't know why. Isn't it possible to use sleep this way?

[NathanFreeman]

Swoole\Table is suitable for inter-process communication; for a single process, you can just use an array

[Draghmar]

Yeah, I know that. The code here was just for testing as it's easier this way. I'm using Tables in WebSocket server for getting data between stuff. And I wanted to check what will happen in scenario like the one above but with simpler code, to simulate.

[NathanFreeman]

https://wiki.swoole.com/en/#/other/issue?id=about-segmentation-fault-core-dump
Can you check where the core dump is caused based on the above document?

[Draghmar]

I will try to do that but I need to have a little bit more time, which is at deficiency for me right now. ;)

[NathanFreeman]

I feel that dividing into several small tables and querying them in batches using multiple coroutines offers no benefits because all operations of Swoole\Table are performed in memory, with no I/O switching involved, so launching multiple coroutines is ineffective.

[Draghmar]

Well, that's probable. But I still need to loop through at least whenever new connection comes in. And loop has its cost on its own. So I thought that maybe splitting big loop into smaller ones, with stopping all when the result is found in one of them, will show some benefits. The simulation I presented above was there to actually answer that question. ;)

[NathanFreeman]

I think that dividing into several small tables and querying them in batches using multiple coroutines offers no benefits because all operations of Swoole\Table are performed in memory, with no I/O switching involved, so launching multiple coroutines is ineffective.