View Revisions: Issue #2580

Summary 0002580: Send keepalive during push or pull to prevent connection timeout
Revision 2016-04-29 23:45 by elong
Description When pushing batches with stream to stage enabled, the batches are written to staging and then loaded to the database, while the client waits on an idle HTTP connection. The connection will eventually time out or be forgotten by the firewall. Then the client tries to push again and gets "server too busy" or the http connection will hang until the timeout threshold. Let's send an occasional partial ack response (like a name/value of 1=1&) that keeps the connection alive. Then we can lower the HTTP time out to be able to recover from actual network problems more quickly.

Similar situation with pulls that timeout while waiting for batch extraction to send batch CSV data. Let's send an occasional partial batch response (like nodeid, 0000) that keeps the connection alive.
Revision 2016-04-28 17:22 by elong
Description When pushing batches with stream to stage enabled, the batches are written to staging and then loaded to the database, while the client waits on an idle HTTP connection. The connection will eventually time out or be forgotten by the firewall. Then the client tries to push again and gets "server too busy" or the http connection will hang until the timeout threshold. Let's send an occasional partial ack response (like a name/value of 1=1&) that keeps the connection from being idle. Then we can lower the HTTP time out to be able to recover from actual network problems more quickly.