Added detection of double scheduling, that is cases where a diskserver wants to restart a transfer that the head node believed was already running. This can happen in case of timeout in the connection while the server has already answered

Improved listtransfers so that it dose not hide diskservers that could not be contacted by warns the user or displays '-' for its numbers in the summaries

Fixed finalization of diskmanagerd. The ActivityControl thread was stuck and the threads were not joined properly

in STAGEOUT after preset checksum mismatch.
Now cases whereby the user drops (nsrm) the file during the
transfer, resulting in Cns_closex returning ENOENT,
are also taken into account by calling putFailed.

Fixed: #81247: Communication errors result in file descriptor leaks in the rmmaster collector thread

Fixed syntax error in SQL (previous fix was the right one in the wrong place). Reverted the wrong one and add the right fix.

The downgrading of the log messages was too drastic, hence the information
about the requester, which was logged separately, got lost. To fix this problem
we now pass down this information to the handlers, which will log it systematically.
In the mean time, the handler map mechanism has been removed to replace it with
a simpler switch statement, which allows easier debugging.



Fixed a typo in a comment.

Trying eclipse & teamcity's autocommit on successful build.

Implemented a thread pool server for rpyc. The default threaded server is creating one thread per request (no reuse) which slows down the application

Optimized handling of pending d2d destinations by replacing a list by a dictionnary. This makes the deletion of completed transfers run in log(n) time rather than n

Fixed handling of exceptions when informing the destinations of disk to disk copies that it can start

Fixed atomicity of transferStarting for source end of a disk to disk copy in case we timeout in the connection to the disk server

Fixed default (commented) values of the SynchronizationInterval and ManagementInterval options of the TransferManager and JobManager. They had been inverted

Make sure that the d2dend command is atomic, that is that it does not leave the d2dsrc on the diskserver while removing it from the headnode

Limit the length of the internal queue between the DB extraction and the actual dispatching to 2x the number of workers. Also take care that the queue is emptied when we exit gracefully

Do not use intermediate status BEINGSCHED for subrequests. It was only present for the cases where LSF was losing jobs. This is no more hapenning, and would it happen, the synchronization would take care of it