-
Bug
-
Resolution: Duplicate
-
Major
-
None
-
4.14.0, 4.15.0, 4.16.0
-
None
-
False
-
Description of problem:
When several CDI host-assisted clones from the same PVC are done concurrently, the source volume files (including disk.img) ctime may change during the operation, so tar fails and the pod is restarted: clone-source.go:41] Subprocess did not execute successfully, result is: '\x01' /usr/bin/tar: disk.img: file changed as we read it
The issue happens with the current tar version 1.34, and seems to be solved in 1.35:
* Warn file changed as we read it less often. Formerly, tar warned if the file's size or ctime changed. However, this generated a false positive if tar read a file while another process hard-linked to it, changing its ctime.
We tested it by explicitly using rhel10 stream tar 1.35. It depends on newer glibc (glibc-2.39-15.el10.x86_64.rpm), so we added it as well.
Since we are based on the rhel9 stream, we would like to have tar 1.35 rpm backport build there to solve the issue, as well of the required glibc (unless a static build is provided).
Version-Release number of selected component (if applicable):
OCP 4.16.0
How reproducible:
100%
For steps to Reproduce, actual results, expected results etc:
see original issue https://issues.redhat.com/browse/CNV-43105
Additional info:
- is depended on by
-
CNV-43105 [4.16] Concurrent host-assisted clone may have many restarts, log says "file changed as we read it "
-
- Closed
-
- relates to
-
RHEL-50158 RHEL9 stream tar 1.34 is failing on concurrent file read ctime update; should be updated to 1.35 where it's fixed
-
- Closed
-