Option "expire_items_per_iter" does not work

[zhka]

1. I set expire_items_per_iter=1000 and run

$echo "flush_all" | nc 127.0.0.1 11211
OK

Tarantool removes 50 records per second.
2. I found a typo in init.lua script:

--- a/memcached/init.lua
+++ b/memcached/init.lua
@@ -236,7 +236,7 @@ local stat_table = {
 local conf_table = {
     readahead            = ffi.C.MEMCACHED_OPT_READAHEAD,
     expire_enabled       = ffi.C.MEMCACHED_OPT_EXPIRE_ENABLED,
-    expire_items_per_item= ffi.C.MEMCACHED_OPT_EXPIRE_COUNT,
+    expire_items_per_iter = ffi.C.MEMCACHED_OPT_EXPIRE_COUNT,
     expire_full_scan_time= ffi.C.MEMCACHED_OPT_EXPIRE_TIME,
     verbosity            = ffi.C.MEMCACHED_OPT_VERBOSITY,
     protocol             = ffi.C.MEMCACHED_OPT_PROTOCOL

After fixing tarantool uses 100% CPU and do not remove anything.

[zhka]

Why not use truncate() function for "flush_all"?

[bigbes]

@zhka we can use truncate when flush_all is called without arguments, but it'll be relatively slow, if cache is big. This operation will block Tarantool.
I'll look why it hangs Taratnool tomorrow

[bigbes]

@zhka I've applied your patch, but can't reproduce Tarantool's 100% CPU usage. Could you provide me test case?

[zhka]

Configuration

I use tarantool-1.6.7 from this package: http://tarantool.org/dist/master/centos/6/os/x86_64/Packages/tarantool-common-1.6.7-463.noarch.rpm,
tarantool/memcached-0.1.0,
small-1.0.0
on CentOS 6.6 x86_64
Startup config:

box.cfg{
listen=3301;
background = true;
}
local console = require('console')
console.listen('127.0.0.1:3313')

local memcached = require('memcached')
local instance = memcached.create('test', '0.0.0.0:11211', {expire_enabled = true; expire_items_per_iter=1000; space_name = 'memcached'; verbosity=3})

Test case

• Start tarantool
• Load 10000 keys in "memcached"

$ memcslap --servers=127.0.0.1 --verbose --test=set
    Threads connecting to servers 1
    Took 11.771 seconds to load data

• Check:

localhost:3313> box.space.memcached:len{}
---
- 10000
...

• Run "flush_all":

$ echo "flush_all" | nc 127.0.0.1 11211
OK

• After about 4 minutes box.space.memcached:len{} returns 0 (Removal rate = 50 keys per second)
• Patch /usr/share/tarantool/memcached/init.lua
• Restart tarantool
• Get CPU load:

$ ps aux | grep tarantool
494      25846 100  1.8 1529920 74496 ?       Rsl  17:41   0:14 tarantool memcached.lua <running>

100%!

• Load 10000 keys in "memcached"

$ memcslap --servers=127.0.0.1 --verbose --test=set
    Threads connecting to servers 1
    Took 12.914 seconds to load data

• Run "flush_all":

$ echo "flush_all" | nc 127.0.0.1 11211
OK

• After about 4 minutes box.space.memcached:len{} returns 10000

[bigbes]

@zhka Thank you! I'll look into it.

[zhka]

@bigbes Thank you!
I build a branch "macos-fix". The problem is resolved.

[bigbes]

Thank you, sorry for inconvenience.

[bigbes]

I'll merge this branch into master, when we'll fix the builds.