Adam Lofts created COUCHDB-1702:
-----------------------------------
Summary: Recv failure: Connection reset by peer in form_submit.js
test
Key: COUCHDB-1702
URL: https://issues.apache.org/jira/browse/COUCHDB-1702
Project: CouchDB
Issue Type: Bug
Reporter: Adam Lofts
On Ubuntu 12.10 the form_submit test case sometimes fails with:
{code}
not ok 1 form_submit
Reason: Failed to execute HTTP request: Recv failure: Connection reset by peer
Trace back (most recent call first):
37: /home/adam/dev/fp3/couchdb/test/javascript/couch_http.js
("{}")
425: /home/adam/dev/fp3/couchdb/share/www/script/couch.js
("POST","/test_suite_db/baz",[object Object])
20: /home/adam/dev/fp3/couchdb/share/www/script/test/form_submit.js
()
53: /home/adam/dev/fp3/couchdb/test/javascript/cli_runner.js
runTestConsole(1,"form_submit",(function (debug) {var db = new CouchDB
72: /home/adam/dev/fp3/couchdb/test/javascript/cli_runner.js
runAllTestsConsole()
85: /home/adam/dev/fp3/couchdb/test/javascript/cli_runner.js
{code}
By debugging the send() calls I see that couchdb is closing the socket before
the response body gets sent. I.e. The data is getting discarded in the kernel
buffer.
There are a variety of references for this issue:
http://stackoverflow.com/questions/8874021/close-socket-directly-after-send-unsafe
http://ia600609.us.archive.org/22/items/TheUltimateSo_lingerPageOrWhyIsMyTcpNotReliable/the-ultimate-so_linger-page-or-why-is-my-tcp-not-reliable.html
In my testing I have found that adding a shutdown() call before close() fixes
this issue for me.
A branch with the shutdown() call is available here:
https://github.com/adamlofts/couchdb/tree/form-submit-test-failure
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
