Redis sessions, when set with the persistent flag and echo_check_liveness, generate an enormous number of echo calls

[bitactive]

Expected behaviour

When phpredis used as session handler it should ECHO similar amount of times as GET/SETEX on each server.

Actual behaviour

One of 12 redis instances got 10x more ECHO commands than others. Disabling echo_check_liveness fixes the problem.

Steps to reproduce, backtrace or example script

I have redis session handler to multiple redis instances using following string:

session.save_path = "tcp://10.0.1.11:7010?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.11:7011?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.11:7012?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.11:7013?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.11:7014?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.11:7015?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.12:7010?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.12:7011?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.12:7012?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.12:7013?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.12:7014?weight=1&timeout=0.2&persistent=1&read_timeout=0.5,tcp://10.0.1.12:7015?weight=1&timeout=0.2&persistent=1&read_timeout=0.5"

But one instance (10.0.1.11:7013) is getting ~5x higher load then other instances. I have run redis monitor on each instance for 5 seconds and check amount of GET/SETEX/ECHO commands.

Each instance has similar amount of GET/SETEX/ECHO:

# redis-cli -h 10.0.1.11 -p 7012 monitor > redismon
# cat redismon | grep GET | wc
  19371   98177 3416606
# cat redismon | grep SETEX | wc
  19340  203899 83526191
# cat redismon | grep ECHO | wc
  19327   96635 1526833

Except one: 10.0.1.11:7013

# redis-cli -h 10.0.1.11 -p 7013 monitor > redismon
# cat redismon | grep SETEX | wc
  16345  175717 65224193
# cat redismon | grep GET | wc
  16356   81820 1646265
# cat redismon | grep ECHO | wc
  187111  935555 14781769

for some unknown reason, this instance is getting 10x more ECHO commands than SETEX/GET resulting in 5x higher total load.
Setting
redis.pconnect.echo_check_liveness = 0
fixes the problem, so the issue is for sure in persistent liveness check. But why only one of 12 instances got 10x more ECHO commands?

I'm seeing this behaviour on

• OS: RHEL9
• Redis: 6.2.7-1.el9
• PHP: 8.2.12-1.el9.remi
• phpredis: php-pecl-redis5-5.3.7-2.el9.remi

Steps to reproduce, backtrace or example script

I've checked

• There is no similar issue from other users
• Issue isn't fixed in develop branch