Redis driver reserved id repush to wating list before delete

[sanwv]

What steps will reproduce the problem?

• set ttr=5 , one job sleep 10s
• use supervisor start 10 worker to handle jobs
• push 100 job to queue sometimes will error, when one job moveExpired before delete

What's expected?

What do you get instead?

multiprocess need data in time sequence, i just give $queue$.reserved item a greater score than time()+ttr , a temporary situation

Additional info

Q	A
Yii version
PHP version
Operating system

[sanwv]

when error_reporting(0) the error msg : job message data been deleted , cant resolve ttr from the serialize data

Exception 'TypeError' with message 'Argument 5 passed to Symfony\Component\Process\Process::__construct() must be of the type float or null, string given, called in /var/www/test/vendor/yiisoft/yii2-queue/src/cli/Command.php on line 185'

in /var/www/test/vendor/symfony/process/Process.php:140

[samdark]

$ttr passed at https://github.com/yiisoft/yii2-queue/blob/master/src/cli/Command.php#L185 is a string while it should be int. Any idea why?

[sanwv]

the main point is retry a job and the job timeout be kill is all same time(after ttr seconds), the two event is standalone on multiprocess（retry maybe before job timeout or job timeout before retry）

if a retry before the job be kill, the retry job maybe get empty message from $queue$.message and break,in milliseconds

• a timeout job id from *.reserved repush to *.waiting（in A worker）
• the job just be kill after repush some milliseconds and invoke delete, the job data in *.messages deleted (in B worker)
• the job id pop from *.waiting and continue execution, fetch data from *.messages. damn！where is my data! ^_...
• the $ttr and $meesage is null ,pass to Symfony Process, the process become scapegoat ^.^

[sanwv]

more possible to happen when a great many worker and job execution time more than ttr

in fact, can not ensure a job always finish in ttr seconds

[samdark]

How would you solve it, @sanwv?

[sanwv]

$this->redis->zadd("$this->channel.reserved", time() + $ttr+60, $id);

or

 'queue' => [
            'ttr' => 86400*30
],

a little funny, just a temporary situation

[Roland-Zhu]

Got the same problem, waiting for official solution

[sanwv]

$ttr passed at https://github.com/yiisoft/yii2-queue/blob/master/src/cli/Command.php#L185 is a string while it should be int. Any idea why?

when repush to waiting list (line 137) before delete message (line 182), so will get id (line 145) but lost payload (line 151) , $payload=NULL and $ttr=""

the problem cause fatal error, the worker progress break off

https://github.com/yiisoft/yii2-queue/blob/master/src/drivers/redis/Queue.php#L137 https://github.com/yiisoft/yii2-queue/blob/master/src/drivers/redis/Queue.php#L182 https://github.com/yiisoft/yii2-queue/blob/master/src/drivers/redis/Queue.php#L145 https://github.com/yiisoft/yii2-queue/blob/master/src/drivers/redis/Queue.php#L151

I'm not sure if other reasons also cause this issue, such as redis lru policy or line 145 and line 151 is not atomic

[sanwv]

related #218 #312 when many workes and job run time longer than ttr, job retry not set, handleMessage (line 61 ) get true and delete message. delete message and moveExpired (line 137 ) run simultaneously in two progress, when moveExpired before than delete message will cause the issue.