http output plugin leaks FDs

Description

(similar to LOGSTASH-892)

FD count continously increases e.g.:

  • right after start: 78

  • then a few minutes later 80 then 91 and so on... keeps increasing until the OS allows the process to allocate

lsof -p 29261 | wc -l
93

it would appear these build up over time:
lsof -p 29261 | grep sock
java 29261 root 12u unix 0xffff88005f3360c0 0t0 207711512 socket
java 29261 root 18r sock 0,6 0t0 207711896 can't identify protocol
java 29261 root 24u sock 0,6 0t0 207712548 can't identify protocol
java 29261 root 27u sock 0,6 0t0 207713925 can't identify protocol
java 29261 root 29u sock 0,6 0t0 207714404 can't identify protocol
java 29261 root 32u sock 0,6 0t0 207714764 can't identify protocol
java 29261 root 33u sock 0,6 0t0 207715157 can't identify protocol
java 29261 root 34u sock 0,6 0t0 207715489 can't identify protocol
java 29261 root 36u sock 0,6 0t0 207715857 can't identify protocol
java 29261 root 37u sock 0,6 0t0 207716268 can't identify protocol
java 29261 root 38u sock 0,6 0t0 207716680 can't identify protocol
java 29261 root 39u sock 0,6 0t0 207717141 can't identify protocol
java 29261 root 40u sock 0,6 0t0 207717342 can't identify protocol
java 29261 root 41u sock 0,6 0t0 207717596 can't identify protocol
java 29261 root 42u sock 0,6 0t0 207717911 can't identify protocol
java 29261 root 43u sock 0,6 0t0 207718800 can't identify protocol
java 29261 root 44u sock 0,6 0t0 207719144 can't identify protocol

confirmed with other output plugin e.g. elasticsearch transport that this does not happen and FD count stays stable

the plugin is configured something like this:

http {
http_method => "post"
url => "http://logs-01.loggly.com/inputs/<loggly id>/tag/tag1,tag2/"
}

Environment

None

Status

Assignee

Logstash Developers

Reporter

Bela Pesics

Labels

None

Fix versions

Affects versions

1.4.1 (bugfix only)

Priority