Console script. Child processes become zombies when cUrl is used - php

Question

Console script perform data import from external API. For boosting import loading performed in concurrent processes that is created by pcntl_fork command.
For communication with API cUrl is used. Communication performed by https protocol.
By some undefined reason periodically some children becomes zombie. There is no errors/warnings/notices in console and also there is not logs is written. Errors level is configured appropriately.
After investigation I suppose that there problem in curl extension since without it, with fake connection, there is no problems.
Also if run import in single process mode - there is no problems at all.
PHP: 7.2.4,
OS: Debian 9,
Curl: 7.59.0 (x86_64-pc-linux-gnu) libcurl/7.47.0 OpenSSL/1.0.2g zlib/1.2.8 libidn/1.32 librtmp/2.3
Maybe someone encountered similar problem or know possible reasons of this strange behavior?
Pseudo code sample of child logic (main part of child showed):
while (true) {
$socket->writeRawString(Signal::MESSAGE_REQUEST_DATA);
$response = $socket->readRawString();
if (Signal::MESSAGE_TERMINATE_PROCESS === $response) {
break;
}
$response = json_decode($response, true);
if (empty($response) || empty($response['deltaId'])) {
continue;
}
$delta = $this->providerConnection->getChanges($response['deltaId']);
if(empty($delta)) {
continue;
}
$xmlReader = new \XMLReader();
$xmlReader->XML($delta);
$xmlReader->read();
$xmlReader->read();
$hasNext = true;
while ($hasNext && 'updated' !== $xmlReader->name) {
$hasNext = $xmlReader->next();
}
if ('updated' !== $xmlReader->name) {
throw new \RuntimeException('Deltas file do not contain updated date.');
}
if (strtotime($xmlReader->readString()) < $endDateTimestamp) {
$socket->writeRawString(self::SIGNAL_END_DATE_REACHED);
continue;
}
}
posix_kill(\posix_getpid(), SIGTERM);
In providerConnection->getChanges($response['deltaId']); request performed via cUrl. For work with cUrl used Php cUrl class extension

Answer 1

As mentioned in my comments, your problem probably is, that childprocesses that died/finished need to be collected by the parent process, or they remain as zombies.
First solution:
Install a signal handler in the parent. Something like this:
pcntl_signal(SIGCHLD, [$this, 'handleSignals']);
With a signal handler that could look like this:
/**
* #param integer $signal
*/
public function handleSignals($signal) {
switch($signal) {
case SIGCHLD:
do {
$pid = pcntl_wait($status, WNOHANG);
} while($pid > 0);
break;
default:
//Nothing to do
}
}
I normally store the pids of forked children and check them all individually with pcntl_waitpid, but this could get you going.
Second Solution:
Use a double-fork to spawn the child-processes, if the parent does not need to wait for all sub-tasks to finish. A double fork looks like this:
$pid = pcntl_fork();
if ($pid == -1) handleError();
elseif ($pid == 0) { // child
$pid = pcntl_fork();
if ($pid == -1) handleChildError();
elseif($pid == 0) { // second level child
exit(startWork()); // init will take over this process
}
// exit first level child
exit(0);
} else {
// parent, wait for first level child
pcntl_wait($pid, $status); // forked child returns almost immediatly, so blocking wait is in order
}

Answer 2

I give up using cUrl for my task. Today I switched to Guzzle with StreamHandler instead of cUrl and it solved all my problems.
I suppose, that due to some internal errors in cUrl, system was killing my child processes.
This is not answer to my question. It just workaround of my problem for those who may also encounter similar problem.
Topic is still open for possible suggestions/explanations.