Hey there,

While I don't know specifically why, I have a strong suspicion that the Zend library might get a little quirky around sockets within the context of a forking. That being said, I'll see if I can replicate your issue based on the stack overflow question you mentioned.

Cheers dude,
Mike

Brilliant thanks. For the time being I have just disabled php notice errors for the scripts that fork. Makes my log files a little quieter ;)

Hi. I was woundering if you have had chance to look into this yet?

Bump*

+1 Also seeing this issue with workers that fork in PHPResque https://github.com/chrisboulton/php-resque

+1 also seeing this in php resque workers.

Appears harmless... guessing its "QUIT\r\n"

I know how to fix this bug. Just close the connect to redis before pcntl_fork and re-connect to redis after pcntl_fork.

no updates?

+1

Same here. PHP 5.5.9, phpredis 2.2.7

Simple script to replicate this bug https://gist.github.com/lexaurin/8897f1be48b9e8747b7d/e8a5fcd750623b6fc250f9c529fc2e65cc55893a

Expected result:

bool(true)

Actual result:

bool(true)
PHP Notice:  Unknown: send of 6 bytes failed with errno=32 Broken pipe in Unknown on line 0

Notice: Unknown: send of 6 bytes failed with errno=32 Broken pipe in Unknown on line 0

PHP Notice:  Unknown: send of 6 bytes failed with errno=32 Broken pipe in Unknown on line 0

Notice: Unknown: send of 6 bytes failed with errno=32 Broken pipe in Unknown on line 0

Connections before script (lsof -i :6379):

COMMAND     PID USER   FD   TYPE   DEVICE SIZE/OFF NODE NAME
redis-ser 20319 root    4u  IPv4 10422993      0t0  TCP *:6379 (LISTEN)

Connections after first child forked:

COMMAND     PID USER   FD   TYPE   DEVICE SIZE/OFF NODE NAME
redis-ser 20319 root    4u  IPv4 10422993      0t0  TCP *:6379 (LISTEN)
redis-ser 20319 root    6u  IPv4 10446216      0t0  TCP config0.local:6379->config0.local:52050 (ESTABLISHED)
php       21993  lex    8u  IPv4 10446860      0t0  TCP config0.local:52050->config0.local:6379 (ESTABLISHED)
php       21994  lex    8u  IPv4 10446860      0t0  TCP config0.local:52050->config0.local:6379 (ESTABLISHED)

Connections after second child forked:

COMMAND     PID USER   FD   TYPE   DEVICE SIZE/OFF NODE NAME
redis-ser 20319 root    4u  IPv4 10422993      0t0  TCP *:6379 (LISTEN)
php       21993  lex    8u  IPv4 10446860      0t0  TCP config0.local:52050->config0.local:6379 (CLOSE_WAIT)
php       22020  lex    8u  IPv4 10446860      0t0  TCP config0.local:52050->config0.local:6379 (CLOSE_WAIT)

Connections after third child forked:

COMMAND     PID USER   FD   TYPE   DEVICE SIZE/OFF NODE NAME
redis-ser 20319 root    4u  IPv4 10422993      0t0  TCP *:6379 (LISTEN)

Note: Fact that child closes "both" connections at the end is right (since it shares file descriptor with parent).

Modified my gist a bit to see which process produces notice https://gist.github.com/lexaurin/8897f1be48b9e8747b7d/d46ccd5c2ffabb1c3a09118db80654cb382c6c96
and it's interesting - third child at the end and following children and then parent at the end.

My log:

bool(true)
Parent pid: 22968
Child pid: 22969
Child pid: 22971
Child pid: 22972
Error in PID 22972: Unknown: send of 6 bytes failed with errno=32 Broken pipe
Child pid: 22973
Error in PID 22973: Unknown: send of 6 bytes failed with errno=32 Broken pipe
Child pid: 22974
Error in PID 22974: Unknown: send of 6 bytes failed with errno=32 Broken pipe
Error in PID 22968: Unknown: send of 6 bytes failed with errno=32 Broken pipe 

I have same problem, I use pconnect to connect redis.

After fork, I pconnect again, But nothing help.

Sometime always throw exception like this.

Exception Redis::lLen(): send of 38 bytes failed with errno=32 Broken pipe in

This code just manager jobs, and run it.

        $files = glob(__DIR__ . '/jobs/*.php');
        foreach ($files as $_file) {
          $this->workers[$name . $i] = pcntl_fork();
          $redis = new Redis();
          $redis->pconnect('127.0.0.1', 0.3);
          if ($this->workers[$name . $i] === 0) {
            posix_kill(posix_getpid(), SIGTERM);
          }
        }
        while (count($this->workers)) {
            $pid = pcntl_wait($status);
            usleep(50000);
        }

Then the unix crontab will run it. Same job will pconnect Redis, after pcntl_fork() pconnect Redis can't fix

+1

Having the same issue when using https://github.com/kriswallsmith/spork

this bug is from php pcntl, just in the worker colse redis, will solve this problem.

Same issue happens when using PHP 5.5.9 + phpredis 2.2.7 + redis engine 2.8.19 + Magento CE 1.9.0.1 + Cm_Cache_Backend_Redis environment.
Error:

Redis::exec(): send of 8192 bytes failed with errno=32 Broken pipe

+1

+1 :)

+1

You guys know there's now a "+1" equivalent "like button" on GitHub right? Without spamming the participants with emails (unless that's your objective!)

+1

php 5.5.9 suddenly started doing same behaviour. I cannot use it anymore in production. This is high priority bug

Could someone provide script which cause this error?

@yatsukhnenko here you go https://gist.github.com/lexaurin/8897f1be48b9e8747b7d/d46ccd5c2ffabb1c3a09118db80654cb382c6c96

@lexaurin, thanks.
There is no error on my gentoo box

php redis-pipe-bug.php 
bool(true)
Parent pid: 27921
Child pid: 27922
Child pid: 27923
Child pid: 27924
Child pid: 27925
Child pid: 27926

Tested on php5.3-php7.1

@lexaurin depends environment?

I can't reproduce it on my desktop too (PHP 7.0.8-0ubuntu0.16.04.3)

Problem was on Debian machine a while ago and I don't have access there anymore. PHP 5.5.9, phpredis 2.2.7 at that time.

Maybe someone with "+1" can provide more infomration and test it out.

phpredis 2.2.7 is to old. problem may be fixed long time ago :)

I know, it was reported in 2014. If no one can confirm it on newer versions of phpredis I suggest to close this bug with resolution "Upgrade to 3.x"

Env: PHP7.1.0+phpredis-3.1.0
use RedisCluster connect redis cluster server.
when the high amount of concurrency will appear when the following error:
[2017-01-04 03:18:24.999684] [Notice] RedisCluster::set(): send of 3182 bytes failed with errno=32 Broken pipe at /home/development/pandy/Swoole_Pool/RedisPool/RedisClient.php[512]
[2017-01-04 03:18:24.999892] [Notice] RedisCluster::set(): send of 3182 bytes failed with errno=32 Broken pipe at /home/development/pandy/Swoole_Pool/RedisPool/RedisClient.php[512]

Same error happen to me under debian8, redis 3.2.7, with php7.0 & Mediawiki 1.28.
PHP Notice: Redis::set(): send of 82 bytes failed with errno=32 Broken pipe in /wiki/includes/libs/redis/RedisConnRef.php on line 96

Same error happens to me under CentOS 6 , Redis 3.2.1 with PHP 7.0 & Magento 2

[2017-03-25 00:00:38] Subscription Logger Publish profile to message queue.CRITICAL: ErrorException: Redis::hGet(): send of 71 bytes failed with errno=32 Broken pipe in /opt/nestle/magento/releases/20170324062419/vendor/colinmollenhour/credis/Client.php:1044
Stack trace:
#0 [internal function]: Composer\Util\ErrorHandler::handle(8, 'Redis::hGet(): ...', '/opt/nestle/mag...', 1044, Array)
#1 /opt/nestle/magento/releases/20170324062419/vendor/colinmollenhour/credis/Client.php(1044): Redis->hGet('zc:k:b22_GLOBAL...', 'd')
#2 /opt/nestle/magento/releases/20170324062419/lib/internal/Cm/Cache/Backend/Redis.php(211): Credis_Client->__call('hget', Array)
#3 /opt/nestle/magento/releases/20170324062419/vendor/magento/zendframework1/library/Zend/Cache/Core.php(306): Cm_Cache_Backend_Redis->load('b22_GLOBAL_PRIM...', false)

@ppanphper, @zoglun, @ngoctruongbc, do you have a script to reproduce this issue?

In my opinion this should be closed as it's a duplicate of #70 -- the broken pipe is the effective result that leads to all sorts of read/write errors.

@yatsukhnenko Script to reproduce is here: #70 (comment)

Duplicate #70

Same error happens in swoole worker...