[LU-487] procfs stats file incorrectly records brw_read and brw_write values

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.1.0
• Affects Version/s: Lustre 2.1.0, Lustre 1.8.6
• Severity: 3

Epic

papercut

• Assignee: Richard Henwood (Inactive)
• Reporter: Richard Henwood (Inactive)

Description

brw_read in llite client procfs stats file claims to measure in units of pages. For example:

# cat /proc/fs/lustre/llite/lustre-ffff88000449a800/stats
snapshot_time             1308264214.464257 secs.usecs
dirty_pages_hits          1419716 samples [regs]
dirty_pages_misses        833472 samples [regs]
read_bytes                322963 samples [bytes] 1 2684356 154236888794
write_bytes               225001 samples [bytes] 0 125912 3323679002
brw_read                  15285 samples [pages] 4096 4096 62607360

However, in ./lustre/llite/lloop.c the tallying of the LPROC_LL_BRW_WRITE/READ counter increments by the value of:

page_count << PAGE_CACHE_SHIFT

Using this value has the effect of not measure pages, but the total bytes of the pages. A simple fix maybe to remove the PAGE_CACHE_SHIFT. Or, rename '[pages]' to '[bytes]' in the output.

Issue Links

• Trackbacks: Lustre 1.8.x known issues tracker While testing against Lustre b18 branch, we would hit known bugs which were already reported in Lustre Bugzilla https://bugzilla.lustre.org/. In order to move away from relying on Bugzilla, we would create a JIRA