GitLab

Reworked the queueing of retrieve jobs in the OStoreDB.
Improved some variable names.

Please note that you will have to install the following CASTOR rpms
in order to compile:

  castor
  castor-devel
  castor-lib
  castor-lib-server
  castor-lib-tape

Renamed CTA/tapeserver/castor to CTA/tapeserver/castor_checkout in preparation for the moving the sub-directory CTA/tapeserver/castor/castor up the directory tree

a new composite transferTime (member function) which now matches better the previous definition of transfeTime.

Fixed unhomogeneous error names.

Propagated references to the watchdog to disk and tape threads and tasks.
Added a maps for storing the count of all errors that occured during the session.
Added propagation of the counts to the initial process.
Added error reporting in disk and tapes threads and tasks by using a string marker
allowing to know which part went wrong in high level exception. Exceptions then
bumps up the count in the watchdog (synchronously), and the watchdog sends the
new count to the initial thread later (in the watchdog's thread).
Plus some missing logs fixed when an exception is thrown in a disk write task.

Added interface to reference the recall watchdog to the recall disk thread pool.
Next: Do same for migration. Actually store the reference. Add new error map storing in watchdog. Add error reporting in tape and disk threads.

WIP: Added missing file ID for disk write thread. Added missing error log when an exception is thrown in a disk write task.

WIP: switching to CASTOR-4839 tapeserverd: task injector should decide on closing the session earlier

WIP. Next: log unmount errors (in the RAII)

Finished error counting in data threads.

Removed leftovers of special case snprintf call for double, now that snprintf is the default for doubles.

CASTOR-4823: tapeserverd's main process should log statistics and status of the closing session in all cases

Added new messages between the session process and the main process: addition and removal of log parameters.
Added storage of the log parameters in the catalogueTransferSession.
Added logging of the tape session at the end in the initial process, allowing statistics gathering even in
case of crash. This standard log message is intended to be used as an input for the tape log system.
Changed the default off the formatting of doubles in the log so that we don't end up with hard to parse
scientific notation in the logs.

The offending message has been changed in a previous case: CASTOR-4749

Reviewed all bandwidth calculations to homogenize the statistics printing.
Fixed a bug where the total time for a session was not properly retrieved before logging, leading to NaN printouts.

I believe the Timer class is generic enough to be used by any
part of CASTOR and not just tape.

CASTOR-4765: The heartbeat messages of the data-transfer session of tapeserverd should include valid drive unit-names

Partial implementation of the ticket:
- The reporting to client and session exit part has been dismantled.
- The template based implementation of the watchdog has been changed to a more classing inheritance schema.
- The reporting to mother process of the moved blocks has been re-enabled.

Fixed more slashes in paramter names (MB/s replaced by MBps) and put all tape server stats in MB (as opposed to Mib).

Also removed slashes from parameter names.

Reviewed and fixed the error handling for migrations.
A deadlock was created by the competing usage of a stop variable an normal information flow
through the threads. The global variable's usage has been reduced and clarified:
it is set by the tape thread to indicated an error reached it. From that point
on, the remaining disk threads can stop what they are doing (and not before as failing
to read a file should not block the completion of previous ones).
The file errors are now transmitted to the client before end of session, and in case of flush.
The unit tests have been updated accordingly.
Report messages have been improved.

The disk write task can already be destroy by the time the push of NULL is attempted.

Added support for cancellation of write tasks. This is used to flush tasks when tape read session is lost after a failed positioning.

Only one applies at a time in read or write context, but the naming will make the code more undrestandable.