Task #6845
Bug #6789: Read API calls to the CN are hanging
Look into why the CNs are hanging connections
100%
Description
Because timeouts seem to solve the problem, there might be something in how the CNs are handling connections - possibly not sending the client the signal that no more information is coming.
It would be good to rule out the CLOSE_WAIT situation, something that troubled us in the past, but doesn't seem to have the same smell.
see http://hc.apache.org/httpcomponents-client-4.3.x/tutorial/html/connmgmt.html
and http://blogs.msdn.com/b/spike/archive/2008/10/09/tcp-connections-hanging-in-the-close-wait-and-fin-wait-2-state.aspx
CLOSE_WAIT situations are the client's responsibility to handle, while other sources of the problem may fall into a CN issue.
History
#1 Updated by Rob Nahf about 9 years ago
did a bit of reading: the new HttpClient minor version bump (4.4-GA) includes better stale connection management:
- More efficient stale connection checking: indiscriminate connection checking which results in approximately 20 to 50 ms overhead per request has been deprecated in favor of conditional connection state validation (persistent connections are to be re-validated only if a specified period inactivity has elapsed)
If the CLOSE_WAIT issue is at issue, this might be a cleaner way to handle it rather than coding in a monitoring thread
#2 Updated by Rob Nahf about 9 years ago
- Subject changed from Look into why the CNs are not handing connections to Look into why the CNs are hanging connections
#3 Updated by Rob Nahf about 9 years ago
- Project changed from Java Client to CN REST
#4 Updated by Rob Nahf almost 9 years ago
- Status changed from New to In Progress
- % Done changed from 0 to 30
#5 Updated by Rob Nahf almost 9 years ago
would like to keep this one open at least until v2 testing is more complete. There is not enough information to be proactive on this one, other than upgrading HttpClient version, which may be premature.
#6 Updated by Rob Nahf almost 8 years ago
- Project changed from CN REST to Java Client
- translation missing: en.field_remaining_hours set to 0.0
- % Done changed from 30 to 100
- Status changed from In Progress to Closed
- Category set to d1_libclient_java
extensive reworking of connection management since this bug reported that has seemed to address the issue.
