Memory leak in Rediska

[rauanmayemir]

I'm getting errors in log:
PHP Fatal error: Allowed memory size of 67108864 bytes exhausted (tried to allocate 32 bytes) in Rediska/Connection.php on line 358.

Earlier I had memory_limit 128Mb and had the same issues.
Any recommendations?

[rauanmayemir]

Redis is mostly used for php session storage.
Here's info:

redis> info
redis_version:2.0.4
redis_git_sha1:00000000
redis_git_dirty:0
arch_bits:64
multiplexing_api:epoll
process_id:21529
uptime_in_seconds:170728
uptime_in_days:1
connected_clients:54
connected_slaves:0
blocked_clients:0
used_memory:1342753216
used_memory_human:1.25G
changes_since_last_save:12084
bgsave_in_progress:1
last_save_time:1289659666
bgrewriteaof_in_progress:0
total_connections_received:11565156
total_commands_processed:65579009
expired_keys:540808
hash_max_zipmap_entries:64
hash_max_zipmap_value:512
pubsub_channels:0
pubsub_patterns:0
vm_enabled:0
role:master
db0:keys=1255559,expires=898138
db12:keys=530010,expires=0
db13:keys=10,expires=0

[rauanmayemir]

Rediska uses default db0.

[rauanmayemir]

This caused that project works too long, apparently rediska is the reason.
P.S. sorry for non-stop spamming.

[rauanmayemir]

I've just realized that Rediska_Zend_Session_SaveHandler_Redis->gc tries to get all that bunch of thousands sessions. Is there better way to do gc?

[ybogdanov]

We have a fix on this. I will publish it at monday.

[rauanmayemir]

Happy to hear that. For now we're using dirty fix that just returns true in gc.

[rauanmayemir]

Any news on this?

[ybogdanov]

Sorry for the late answer.

We made a separate class for sessions handling (temporary), which is much more simpler than in Rediska. It uses native redis expire functionality, so garbage collection handling on redis side.

http://pastebin.com/sbTsbwmy

Ivan, maybe we need to enhance Rediska_Zend_Session_SaveHandler_Redis:

1. We don't need to use sets for storing session keys
2. Let Redis do the garbage collection job (expire)

[rauanmayemir]

0ctave, thanks.
I really cannot understand why should we use set for all the sessions.

[shumkov]

Set used for getting all online users.

[ybogdanov]

There's another ways to get online users count. At least, why do we need to grab ALL of this users to each scenario while collecting garbage? :) If I have millions online?

I think, the solution is to put garbage collection process on redis itself - it is native functionality.

[till]

+1 for letting redis handle gc. I think it's much more efficient.

PHP-based GC is a bottleneck in my different session handlers. Due to PHP's general script nature it's blocking too. ;-)

[till]

Btw, I checked in the session save handler:

in __construct it uses session.gc_maxlifetime -- then in write() it uses Rediska::expire() to set the objects TTL.

It looks like we don't need to fix much here:

• we disable gc()
• we tell people to explicitly set session.gc_maxlifetime instead

We could do something like that:

if (ini_get('session.gc_maxlifetime') === null || ini_get('session.gc_maxlifetime') == 0) {
    trigger_error('Please set session.gc_maxlifetime to enable garbage collection.', E_USER_WARNING);
}

[till]

c637134