later on.

options to DiskManager/*PendingTimeout to improve usability.

affecting draining operations.

Completed the fix to take into account disk-to-disk copies failed
at the start.
To have a full tracking, a src diskcopy ID field has been added
to Disk2DiskCopyJob, and populated at start time when available.

Fixed tight loop on errors when filesystems of the gcd are badly configured. This could go up to 10KHz !

However, this is to be evolved to a more robust schema, where
DISABLED hardware for too long is deemed not to have any transfer any
longer and won't stop the synchronization from being executed on the
rest of the cluster.

affecting draining operations.

Fixed all items mentioned in the ticket, plus added a timestamp column
on DrainingErrors that is shown by draindiskserver -q -f.
Furthermore, as DrainingErrors now contains a reference to DiskCopy,
draindiskserver -q -f prints the full path of the failed draining DC.

- to missing req or subreq UUIDs
- to exceptions (e.g. from cx_Oracle) that would throw AttributeErrors

One more improvement after a campaign of stress test: avoid to
signal() when someone else already did it. This helps reducing
contention on SYS.DBMS_ALERT_INFO - and ought to have been provided
by the Oracle DBMS_ALERT API...

are too off when portion is small.

cancelled

resilient to hardware unavailability.

broken when dealing with double copy recalls.

This fix is not optimal: we're still using fileTransactionId being NULL
as a flag while we should really create an extra status
RECALLJOB_SELECTED2ND or SELECTEDSHADOW. Not on this hot fix...

This commit also includes a simpler way to select new recall candidates,
which makes it more immune to bugs like the above.
Plus a fix related to the precision used to check if a file was
overwritten while being recalled.

This was supposed to be part of commit 1f82df87 but got lost.

Plus one more fix for bug #103242 Incorrect evaluation of
lastModificationTime leads to double recalls + dark data.
Hotfix script to be produced yet.

categories.

simply not called and is dead wood:

    CuuidToString()
    dumpFileReq()
    dumpTapeReq()
    rtcp_NewTapeList()
    rtcpc_BuildDumpTapeReq()
    rtcpc_InitDumpTapeReq()

not called and is dead wood:

rtcpd_WaitForPosition()

double recalls + dark data.
The dark data generation is prevented by creating a DiskCopy entry in
all cases at the end of a recall.
The false positives have been fixed in both NS 'native' and
'compatibility' mode by always using the openTime when available
and defaulting to the mtime as we used to do in 2.1.13. Clearly
only in native mode can we exploit the microsecond precision.