Joe McDonnell created IMPALA-13629:
--------------------------------------
Summary: Data cache's hole punching check fails with XFS allocsize
> 64k
Key: IMPALA-13629
URL: https://issues.apache.org/jira/browse/IMPALA-13629
Project: IMPALA
Issue Type: Bug
Components: Backend
Affects Versions: Impala 4.5.0
Reporter: Joe McDonnell
The data cache relies on hole punching, so it does a sanity check at startup to
make sure hole punching is working. It does these steps:
# Create a file
# Write 64kb to the file in four 16kb chunks
# Check the size of the file on disk and verify it is 64kb
# Punch a hole that removes the middle 32kb
# Check that the size of the file on disk is now 32kb
XFS's allocsize option can break this check. If XFS's allocsize option is set
to >64kb (e.g. 256kb), then writing 64kb doesn't result in a 64kb file.
Instead, it would be the allocsize (e.g. 256kb). This cause the hole punching
check to fail with output like:
{noformat}
Unexpected pre-punch file size for
/mnt/cache/impala-datacache-0/714a91193b8ed96d:8bb3d659c0be24a8: expected 65536
but got 262144{noformat}
The data cache should be able to operate on a system where allocsize is set
larger than 64kb, but this check will need to modified to handle that.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
