Memory Leak issue

[gmoskovicz]

I am using a simple configuration:

input {

    file {

        type => "one_type"

        start_position => "end"

        path => [

        "/Users/Gabriel/Documents/ElasticSearch/Test/logstash-2.1.2/logs/*"

        ]

        sincedb_path => "/Users/Gabriel/Documents/ElasticSearch/Test/.sincedb"

        codec => multiline {

        auto_flush_interval => 120

        max_lines => 100

        max_bytes => "10 MiB"

        pattern => "NVRAZBYCX"

        what => "next"

        negate => true

        }

    }

}

filter {

    drop{}

}

And generating random lines of logs using a simple command line that writes new lines to a log inside logs folder.

The memory consumption is going up and up with no stop, and looks like the GC is not being able to collect much garbage:

I generated a heap dump from the logstash instance and got the following:

HTML to be opened on a browser (remove the .txt since Github prevents people from uploading html):

Objects-by-class.html.txt

Image:

[gmoskovicz]

Same test without the auto_flush_interval is showing better results:

[guyboertje]

I can confirm that a library we use concurrent_ruby is not fully removing cancelled timer tasks.

I am working on two solutions for two different scenarios.

• when ML is used without identity mapping, e.g. stdin input. I have a solution for this that uses one timer.
• when ML is used with identity mapping, .e.g. file input. Here we can't have a timer per codec as each file path (identity) seen creates a copy of the codec and therefore a timer and a timer thread - 20K files = 20K codecs = 20K timer threads. I have a solution and am coding it now.

[ph]

I can confirm that #29 PR fixes this issue, I used @gmoskovicz configuration and a script that generate log line. I ran the scripts for 2 hours and monitored the memory usager. GC was correctly behaving with no leak.

[gmoskovicz]

Thanks for taking care of this @ph @guyboertje