[HTTPCORE-567] Connection leak when aborting httpRequest after connection leasing

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 4.4.11
• Fix Version/s: 4.4.12
• Assignee: Unassigned
• Reporter: Zoe Wang
• Created: 22/Jan/19 22:45
• Updated: 01/Feb/19 08:22
• Resolved: 01/Feb/19 08:22

Description

This is a follow up on https://issues.apache.org/jira/browse/HTTPCLIENT-1958. 

The leaking thread interrupt flag is fixed but we noticed that there's an edge case when a request gets aborted right after a new connection is leased, that connection will never get re-used and closed. 

 

From the source code, looks like the edge case that needs to be handled is "if (entry != null && future.isCancelled())"

Code to reproduce:

ScheduledExecutorService scheduledExecutorService =           Executors.newScheduledThreadPool(1);
PoolingHttpClientConnectionManager poolingHttpClientConnectionManager = new PoolingHttpClientConnectionManager();

CloseableHttpClient httpclient = HttpClients.custom().setConnectionManager(poolingHttpClientConnectionManager).build();

try {
HttpGet httpGet = new HttpGet("http://somehost");
scheduledExecutorService.schedule(httpGet::abort, 40, TimeUnit.MILLISECONDS);
CloseableHttpResponse response = httpclient.execute(httpGet);

HttpEntity entity = response.getEntity();
EntityUtils.consume(entity);
} catch (Exception e) {
    e.printStackTrace();
} finally {
scheduledExecutorService.shutdown();
}

System.out.println(poolingHttpClientConnectionManager.getTotalStats());

The printed out log was: [leased: 1; pending: 0; available: 0; max: 20]

Note that you might need to change the abort time to reproduce the issue.