I have the same problem, though in a long running script on a replica where (apparently) a connection reset sometimes occurs. After this happened I get the same 'NOAUTH Authentication required' error.

I have the same problem. Only if I catch the exception and close the connection, the error 'NOAUTH AUTHENTICATION required' disappear.

I can't replicate this with PhpRedis 5.3.0. Both of the examples work fine (and reauthorize the connection)

class RedisBase extends Redis	
	public function __construct()
    {
        $this->connect('127.0.0.1', 6379);
        $this->auth('password');
    }
}

new RedisBase();

When i use tcpkill to kill the tcp connection, I get the 'NOAUTH Authentication required' throwed by connect function.

Can you provide a bit more context?

From the above sample code, it seems like you'd have a difficult time timing the tcpkill call. Or do you mean you're calling it in a loop?

Well. I use php 7.3.6 And php-redis 5.3.0
Use nginx and php-fpm

I use the script as below, and I can get the right value.

<?php
$redis = new Redis();
$redis->pconnect('127.0.0.1');
$redis->auth('password');
echo $redis->get('a');

After that, I use tcpkill to kill the connection. And then stop tcpkill running.
Then I visit the page, I will get the error below.

Stack trace:
#0 /home/www/missevan-web/public/index_temp.php(3): Redis->pconnect('r-bp1bts3fbiqme...')
#1 {main}
thrown in /home/www/missevan-web/public/index_temp.php on line 3
NOTICE: PHP message: PHP Fatal error: Uncaught RedisException: NOAUTH Authentication required. in /home/www/missevan-web/public/index_temp.php:3

And then I do the samething to the script below. It's OK. No error appear.

<?php
$redis = new Redis();
try{
  $redis->pconnect('127.0.0.1');
  $redis->auth('password');
  echo $redis->get('a');
} catch(Exception $e) {
  $redis->close();
}

Ok I will try that setup tomorrow morning.

I can replicate this. I'll post an update when I figure out exactly what's going on.

OK, this exception is being thrown in our ECHO "liveness check".

There are two workarounds (as long as you're running >= 5.3.0):

1. Change your pconnect call to include the AUTH information:

$redis->pconnect('127.0.0.1', 9999, 1, NULL, 0, 0, ['auth' => ['password']]);

2. Disable the redis.pconnect.echo_check_liveness INI setting. I don't recommend this if you've ever had persistent connections get into a bad state. If this has never happened to you then it's probably OK.

Thanks @michael-grunder

Will this essentially be a necessity to fix the issue for persistent connections, and the best practice moving forward? Or is the underlying issue with the auth() method also fixable in a future version?

Once persistent connections are fixed we won't need the liveness check at all, so the problem will go away.

That said, this specific issue is fixable before that, because in actuality a NOAUTH error is sufficient in and of itself to confirm "liveness". The whole point of the liveness check is to make sure there aren't unconsumed replies on the connection when we pull it from the persistent connection pool. If Redis tells us NOAUTH we know for a fact there aren't any replies since we're not even logged in.

I think it will be safe to add NOAUTH as a "success" for the purposes of the liveness check, which should also be pretty simple to implement.

Thanks @michael-grunder 。I will try。