This problem was reported running IDS 14.10.FC6 on linux x86_64.
The customer reported that ontape -r was failing with EPIPE when
the RESTORE_FILTER
was being used.
The problem was reported for any combo of BACKUP_FILTER and
RESTORE_FILTER in the customerâ ™s
environment, for instance, the ontape -r failed for both value
sets below:
BACKUP_FILTER /bin/gzip
RESTORE_FILTER /bin/gzip -d
BACKUP_FILTER /bin/compress
RESTORE_FILTER /bin/uncompress
When the problem occurs the ontape -r restores is able to pull
the chunk and dbspace data from the
backup device, but before asking if the user would like to
â œContinue restore? (y/n)â , the process fails
with an EPIPE error like:⠨⠨
Write failed on parent's output pipe. errno = 32.
Physical restore failed - function close archive tape failed
code 255 errno 0
This was observed on RHEL 8.
Interestingly enough, the problem does not persist across all
platforms running RHEL 8 as our initial
attempts to reproduce the failure were unsuccessful and the
restore process was success using
the RESTORE_FILTER.
Also, adding a little complexity to the defect, on a linux
machine runningn RHEL 7.4, it was observed
in my testing that the ontape process could hang indefinitely
after printing out dbspace and chunk
info but before asking if the user would like to â œContinue
restore (y/n)â
In the hang case, you can see the TAPEIO process that is forked
from the ontape -r process waits
indefinitely in the following stack, gathered from pstack.⠨⠨
#0 0x00007feef429a690 in __write_nocancel () at
../sysdeps/unix/syscall-template.S:81
#1 0x000000000068fa74 in write_to_filter ()
#2 0x000000000049cdb5 in read_filter_from_dev ()
#3 0x000000000068e258 in create_filter ()
#4 0x000000000049d099 in start_restore_filter ()
#5 0x00000000004ad1d9 in first_read_phys_tape ()
#6 0x00000000004a79ad in do_cold_phys_restore ()
#7 0x000000000049c47e in main ()
Not using BACKUP_FILTER and RESTORE_FILTER could be a
workaround, but it you are in a situation
where you need to restore from an archive that used a
BACKUP_FILTER, the workaround is useless
as you have to use RESTORE_FILTER. |
