[
https://issues.apache.org/jira/browse/AMQ-4889?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13835322#comment-13835322
]
matteo rulli commented on AMQ-4889:
-----------------------------------
Actually the problem in ProxyConnector hid something more than a simple problem
ProxyConnector.java file.
Apparently some clean-up invocations were missing in SSLNIOTransport as well.
This resulted in many leaked file descriptors in linux (RHEL 6.3 64 bit).
Before the attached patches the lsof command gave tons of
{noformat}
java 6530 platone 8u sock 0,6 0t0 2489682586 can't
identify protocol
java 6530 platone 9u sock 0,6 0t0 2489745273 can't
identify protocol
java 6530 platone 15u sock 0,6 0t0 2489727545 can't
identify protocol
java 6530 platone 66u sock 0,6 0t0 2489683982 can't
identify protocol
java 6530 platone 69u sock 0,6 0t0 2489684335 can't
identify protocol
java 6530 platone 72u sock 0,6 0t0 2489684339 can't
identify protocol
java 6530 platone 74u sock 0,6 0t0 2489684688 can't
identify protocol
java 6530 platone 76u sock 0,6 0t0 2489685055 can't
identify protocol
java 6530 platone 77u sock 0,6 0t0 2489716539 can't
identify protocol
java 6530 platone 97u sock 0,6 0t0 2489683431 can't
identify protocol
java 6530 platone 98u sock 0,6 0t0 2489683450 can't
identify protocol
java 6530 platone 99u sock 0,6 0t0 2489684695 can't
identify protocol
java 6530 platone 100u sock 0,6 0t0 2489683990 can't
identify protocol
java 6530 platone 101u sock 0,6 0t0 2489702245 can't
identify protocol
java 6530 platone 102u sock 0,6 0t0 2489685058 can't
identify protocol
...
{noformat}
(see attached lsof.txt) with the following socket stats:
{noformat}
sockets: used 3067
TCP: inuse 21 orphan 0 tw 0 alloc 2808 mem 2648
UDP: inuse 6 mem 5
UDPLITE: inuse 0
RAW: inuse 1
FRAG: inuse 0 memory 0
{noformat}
and the following netstat outcome:
{noformat}
...
tcp 0 0 ::ffff:192.168.24.82:44201 ::ffff:192.168.16.166:61616
ESTABLISHED 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.66:64601
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.67:9033
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.66:56924
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.66:33021
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.67:56879
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.67:51487
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.66:35295
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.67:49529
CLOSE_WAIT 6530/java off (0.00/0/0)
tcp 0 0 ::ffff:192.168.24.82:61619 ::ffff:192.168.24.67:51309
CLOSE_WAIT 6530/java off (0.00/0/0)
... many other CLOSE_WAIT
{noformat}
We applied the attached patches and everything run better: the lsof stopped
reporting the _can't identify protocol_ records and the number of connections
in CLOSE_WAIT dropped to zero.
So what? Does it make sense to you?
> ProxyConnector memory usage skyrockets when several ssl handshakes fails
> ------------------------------------------------------------------------
>
> Key: AMQ-4889
> URL: https://issues.apache.org/jira/browse/AMQ-4889
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.8.0, 5.9.0
> Environment: Seen in Windows 7 64bit, Windows Server 2008 R2 and
> Linux RHEL 6.3 64 bit
> Reporter: matteo rulli
> Attachments: ProxyConnIssue.rar
>
>
> See
> [nabble|http://activemq.2283324.n4.nabble.com/Proxy-Connector-memory-consumption-td4674255.html]
> for further details.
> To reproduce the issue:
> # Start embedded proxy broker and the AMQ broker that are embedded in
> *AMQTestBroker* project (see attachments);
> # Start the *AMQTestConsumer* project; This program repeatedly tries opening
> a connection to the ProxyConnector with wrong certificates.
> # Open jconsole to monitor AMQTestBroker memory usage: you should experience
> an OOM error within one hour with the suggested settings (Xmx = 2048m).
> Launch configurations and test keystores are attached to this issue along
> with the java projects.
> This behavior seems to affect _ProxyConnector_ only, running the test against
> a standard nio-based _TransportConnector_ does not seem to produce anomalous
> memory consumptions.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
