[ 
https://issues.apache.org/jira/browse/HTTPASYNC-79?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14063668#comment-14063668
 ] 

Przemysław Ołtarzewski commented on HTTPASYNC-79:
-------------------------------------------------

Oleg,

Thank you for a quick response and for providing the patch.

We have build an experimental version of our solution and deployed it on 
several nodes of the production environment. We will need a couple of days to 
observe the results and verify if the patch fixes our problem.

Przemek

> Connection leak due to race condition on keep-alive boundary
> ------------------------------------------------------------
>
>                 Key: HTTPASYNC-79
>                 URL: https://issues.apache.org/jira/browse/HTTPASYNC-79
>             Project: HttpComponents HttpAsyncClient
>          Issue Type: Bug
>    Affects Versions: 4.0.1
>         Environment: SunOS 5.10
>            Reporter: Przemysław Ołtarzewski
>              Labels: connection, cxf, httpcore, leak
>             Fix For: 4.0.2, 4.1-alpha1
>
>
> We are experiencing a race condition between:
> 1. I/O dispatcher closing a connection due to keep-alive expiry
> 2. PoolingNHttpClientConnectionManager leasing the same connection
> Consider valid exchange 1 provided below:
> {noformat}
> 2014.07.07 08:01:59.282 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG 
> [HLAPI-52] PoolingNHttpClientConnectionManager:245 Connection request: 
> [route: {}->http://10.250.32.140:8082][total kept alive: 1; route allocated: 
> 7 of 1000; total allocated: 7 of 5000]
> 2014.07.07 08:01:59.283 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG 
> [HLAPI-52] PoolingNHttpClientConnectionManager$InternalPoolEntryCallback:462 
> Connection leased: [id: http-outgoing-20156][route: 
> {}->http://10.250.32.140:8082][total kept alive: 0; route allocated: 7 of 
> 1000; total allocated: 7 of 5000]
> 2014.07.07 08:01:59.283 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-52] LoggingIOSession:164 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Set attribute 
> http.nio.exchange-handler
> 2014.07.07 08:01:59.285 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:141 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][rw:w]: Set timeout 20000
> 2014.07.07 08:01:59.285 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-52] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][rw:r]: Event set [w]
> 2014.07.07 08:01:59.286 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][rw:w]: Event set [w]
> 2014.07.07 08:01:59.286 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:201 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][rw:w]: 1111 bytes written
> 2014.07.07 08:01:59.287 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:109 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:w]: Event cleared [w]
> 2014.07.07 08:01:59.688 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 8443 bytes read
> 2014.07.07 08:01:59.690 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 8140 bytes read
> 2014.07.07 08:01:59.690 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 8190 bytes read
> 2014.07.07 08:01:59.691 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:109 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][:r]: Event cleared [r]
> 2014.07.07 08:01:59.692 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-52] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Event set [r]
> 2014.07.07 08:01:59.693 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 8124 bytes read
> 2014.07.07 08:01:59.693 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 0 bytes read
> 2014.07.07 08:01:59.694 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:109 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][:r]: Event cleared [r]
> 2014.07.07 08:01:59.695 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-52] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Event set [r]
> 2014.07.07 08:01:59.695 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-52] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Event set [r]
> 2014.07.07 08:01:59.696 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-52] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Event set [r]
> 2014.07.07 08:01:59.697 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 0 bytes read
> 2014.07.07 08:01:59.697 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: 0 bytes read
> 2014.07.07 08:01:59.698 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:171 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Remove attribute 
> http.nio.exchange-handler
> 2014.07.07 08:01:59.699 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG [I/O 
> dispatcher 61] PoolingNHttpClientConnectionManager:279 Releasing connection: 
> [id: http-outgoing-20156][route: {}->http://10.250.32.140:8082][total kept 
> alive: 0; route allocated: 7 of 1000; total allocated: 7 of 5000]
> 2014.07.07 08:01:59.699 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG [I/O 
> dispatcher 61] PoolingNHttpClientConnectionManager:293 Connection [id: 
> http-outgoing-20156][route: {}->http://10.250.32.140:8082] can be kept alive 
> for 5.0 seconds
> 2014.07.07 08:01:59.700 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG [I/O 
> dispatcher 61] PoolingNHttpClientConnectionManager:299 Connection released: 
> [id: http-outgoing-20156][route: {}->http://10.250.32.140:8082][total kept 
> alive: 1; route allocated: 7 of 1000; total allocated: 7 of 5000]
> {noformat}
> Connection id=20156 has been used for communication, returned to the 
> PoolingNHttpClientConnectionManager (and underlying 
> AbstractNIOConnectionPool). The connection may be kept alive for 5 seconds.
> Now analyze the second provided exchange using the same connection:
> {noformat}
> 2014.07.07 08:02:04.689 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG 
> [HLAPI-60] PoolingNHttpClientConnectionManager:245 Connection request: 
> [route: {}->http://10.250.32.140:8082][total kept alive: 1; route allocated: 
> 7 of 1000; total allocated: 7 of 5000]
> 2014.07.07 08:02:04.690 
> org.apache.http.impl.nio.conn.PoolingNHttpClientConnectionManager DEBUG 
> [HLAPI-60] PoolingNHttpClientConnectionManager$InternalPoolEntryCallback:462 
> Connection leased: [id: http-outgoing-20156][route: 
> {}->http://10.250.32.140:8082][total kept alive: 0; route allocated: 7 of 
> 1000; total allocated: 7 of 5000]
> 2014.07.07 08:02:04.691 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession$LoggingByteChannel:186 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: -1 bytes read
> 2014.07.07 08:02:04.691 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-60] LoggingIOSession:164 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Set attribute 
> http.nio.exchange-handler
> 2014.07.07 08:02:04.692 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG [I/O 
> dispatcher 61] LoggingIOSession:115 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][r:r]: Close
> 2014.07.07 08:02:04.692 
> org.apache.http.impl.nio.conn.ManagedNHttpClientConnectionImpl DEBUG 
> [HLAPI-60] LoggingIOSession:102 http-outgoing-20156 
> 10.252.152.177:61023<->10.250.32.140:8082[ACTIVE][rw:r]: Event set [w]
> {noformat}
> Notice the time difference between exchanges being about 5 seconds.
> 1. At the keep-alive boundary, the connection is requested and leased from 
> the PoolingNHttpClientConnectionManager.
> 2. Then, the I/O Dispatcher with id=61, which performed previous exchange, 
> attempts to read from the connection, receiving no data (-1 bytes).
> 3. In the meantime, HLAPI-60 thread that requested the connection, sets the 
> http.nio.exchange-handler attribute.
> 4. However the connection is already recognized as 'dead' by the I/O 
> dispatcher and is then closed.
> 5. In the end, the PoolingNHttpClientConnectionManager prepares to write to 
> the connection. This is the last activity for connection id=20156. It is 
> never released to the connection pool.
> We use Http Components in conjunction with CXF and in the end get a 
> SocketTimeoutException from Async Transport:
> {noformat}
> Caused by: java.net.SocketTimeoutException: Read Timeout
>         at 
> org.apache.cxf.transport.http.asyncclient.AsyncHTTPConduit$AsyncWrappedOutputStream.getHttpResponse(AsyncHTTPConduit.java:614)
>         at 
> org.apache.cxf.transport.http.asyncclient.AsyncHTTPConduit$AsyncWrappedOutputStream.getResponseCode(AsyncHTTPConduit.java:699)
>         at 
> org.apache.cxf.transport.http.HTTPConduit$WrappedOutputStream.handleResponseInternal(HTTPConduit.java:1555)
>         at 
> org.apache.cxf.transport.http.HTTPConduit$WrappedOutputStream.handleResponse(HTTPConduit.java:1525)
>         at 
> org.apache.cxf.transport.http.HTTPConduit$WrappedOutputStream.close(HTTPConduit.java:1330)
>         ... 25 common frames omitted
> {noformat}
> This scenario is non-deterministic. There are about 10-15 connections leaked 
> this way from the connection pool of 1000 each day. The connection pool 
> slowly degrades and in the end requires a restart of the whole component. Our 
> solution is suppose to be HA, so this is a major problem for us and for our 
> business client.
> We have also noticed, that:
> 1. If the connection is reused earlier (i.e. after 3-4 seconds), there is no 
> read attempt from the I/O dispatcher that causes the '-1' result in the 
> negative scenario. Connection is used to perform a valid exchange and may be 
> further kept alive. Maybe it is a matter of setting the 
> 'http.nio.exchange-handler' attribute before the I/O dispatcher attempts to 
> read from connection due to keep-alive timeout?
> 2. Also, if around the keep-alive time boundary the connection is not reused, 
> it does not leak from the connection pool (it is never leased after being 
> closed by the I/O dispatcher).
> The negative scenario occurs only if there is an interleave between the I/O 
> dispatcher closing the connection and the connection being leased from 
> PoolingNHttpClientConnectionManager at the keep-alive boundary - about 5 
> seconds in our case.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to