dCache Golden Release 1.9.5-8
recommended for first LHC Production period.
Thanks to the great support we got from our early adopters, the
dCache team feels confident that with dCache 1.9.5-8, the Golden
Release is ready to be used for the upcoming LHC production period. The
upgrade procedure has been reported to be easy and painless. Please
find the most prominent fixes below. For details consult the
corresponding release notes at dCache.ORG
Fixed in 1.9.5-7/-8
Tape Protection broken in 1.9.4 and 1.9.5
Tape Protection is a mechanism within dCache, restricting tape
access to eligible users. Only DN/FQANs, listed in a white list
file, are allowed to restore files back from tape to disk.
DN/FQAN's not specified, will receive an error when trying to access
files not being on disk but only on tape. This feature was introduced
with 1.9.4 but turned out not to work correctly in all cases. With
1.9.5 (prior -8), the bug can be overcome by a configuration change.
However with 1.9.4-4 and 1.9.5-8 all those issues are fixed.
SRM-LS and DCAP-LS report incomplete directory listing
In 1.9.5(prior -7), the srm-ls and the dcap-ls commands/api eventually
return an incomplete directory listing. This happens silently. No error
is reported and the command return code is zero. The issue is fixed
with 1.9.5-7.
Delivery of swapped files using passive dcap.
There is a non zero probability that by using passive dCap,
dCap clients receive the wrong file. The problem always happens in
pairs. Two file requests scheduled to the same pool within the same
millisecond may result in returning the files being swapped. The dCap
client doesn't indicate that the wrong file was received. Only checking
the file size, checksum or content would indicate the error. The
probability of this error is extremely low but increases with the
number of requests sent to a single pool at the same time. The flaw is
present in the dCache pool code since the introduction of passive dCap
(in all releases). It is fixed with 1.9.4-4 and 1.9.5-8.
Remaining (known) issue in 1.9.5-8
Verbose but harmless output in pool node log files.
When using dCap the pool node log files will report messages
similar to those :
"[DCap-t2-srm-02-Unknown-6369 PoolMoverKill]
java.util.NoSuchElementException: Job not found : Job-10290"
Although annoying, those message are harmless. This is fixed with
1.9.5-9 (expected next week)
Chimera deployment
An impressive number of sites has been upgrading to chimera for the
last couple of months and by reporting issues, the migration procedure
could even been improved. Nevertheless you need to test the entire
procedure successfully before scheduling the final production
migration. The reason is that over time, your PNFS instance may have
collected hidden inconsistencies which could cause the migration to
halt.
X-mas gift
As we expect that there is a time after the LHC start-up, dCache.org
continues to develop the dCache storage element beyond the Golden
Release. The next release (1.9.6) is scheduled for the first week of
December. It will provide NFS 4.1 and the first version of the dCache
WebDAV implementation. More information will be provided with the
announcement of 1.9.6.
|
