[LU-19147] don't try to cache large objects

• Type: Bug
• Resolution: Unresolved
• Priority: Medium
• Labels: easy
• Severity: 3
• Assignee: Chakshu Kansal
• Reporter: Andreas Dilger

Description

The default osd-ldiskfs.*. readcache_max_filesize parameter is currently "~0ULL" so the OSS will try to cache even very large files that do not fit into the total RAM size. This is inefficient to try and cache these objects.

The default readcache_max_filesize should be tuned at startup to take the actual RAM size into account (e.g. (totalram_pages << PAGE_SHIFT) / 64) so that the OSS RAM can be better utilized to cache objects that actually fit into memory.

Issue Links

• is related to: LU-19267 OSS readcache page allocations can deadlock during local IO
• is related to: LU-19131 DIO read/write can livelock on swapped pages in get_user_pages()
• is related to: LU-14717 osd-ldiskfs: set od_readcache_max_filesize with maximal value when setting with negative value via lproc