[RFC] [PATCH] drop_caches: add syslog entry

[RFC] [PATCH] drop_caches: add syslog entry

[Martin Tegtmeier]

Dear Maintainers,

currently dropping the file system cache ("echo 1 >
/proc/sys/vm/drop_caches") doesn't leave any trace. However dropping
the fs cache can severely impact system behaviour and application
response times. Therefore I suggest to write a syslog entry if the
entire inode page cache is scrapped.
Since it is not an easy task to calculate the size of the droppable
filesystem cache I also suggest to add the number of dropped pages to
the syslog entry. This can be accomplished by saving the return value
of invalidate_mapping_pages().

The number of dropped pages is an important measure for capacity
planning. For the deployment of new SAP application instances we
would like to know the amount of memory that was freed from
fs caches.

Thanks,
   -Martin


 drop_caches.c |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)


commit b2219e84647bdf64fd6e7f9c5260c1e6bed24d58
Author: Martin Tegtmeier <martin.tegtmeier@gmail.com>
Date:   Tue May 24 15:24:20 2011 +0200

    drop_caches: add syslog entry

    Dropping the entire file system cache (inode cache) can severely
influence system behaviour
    yet currently dropping the file system cache is NOT traceable.
    This patch adds an entry to /var/log/messages with a time stamp
and the number of dropped pages.


    Signed-off-by: Martin Tegtmeier <martin.tegtmeier@gmail.com>

diff --git a/fs/drop_caches.c b/fs/drop_caches.c
index 98b77c8..f2e4dc4 100644
--- a/fs/drop_caches.c
+++ b/fs/drop_caches.c
@@ -12,6 +12,7 @@

 /* A global variable is a bit ugly, but it keeps the code simple */
 int sysctl_drop_caches;
+unsigned long pages_dropped;

 static void drop_pagecache_sb(struct super_block *sb, void *unused)
 {
@@ -28,7 +29,7 @@ static void drop_pagecache_sb(struct super_block
*sb, void *unused)
 		__iget(inode);
 		spin_unlock(&inode->i_lock);
 		spin_unlock(&inode_sb_list_lock);
-		invalidate_mapping_pages(inode->i_mapping, 0, -1);
+		pages_dropped += invalidate_mapping_pages(inode->i_mapping, 0, -1);
 		iput(toput_inode);
 		toput_inode = inode;
 		spin_lock(&inode_sb_list_lock);
@@ -55,8 +56,12 @@ int drop_caches_sysctl_handler(ctl_table *table, int write,
 	if (ret)
 		return ret;
 	if (write) {
-		if (sysctl_drop_caches & 1)
+		if (sysctl_drop_caches & 1) {
+			pages_dropped = 0;
 			iterate_supers(drop_pagecache_sb, NULL);
+			printk(KERN_INFO "drop_caches: %lu pages dropped from inode cache\n",
+				pages_dropped);
+		}
 		if (sysctl_drop_caches & 2)
 			drop_slab();
 	}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC] [PATCH] drop_caches: add syslog entry

[Andreas Dilger]

On May 24, 2011, at 08:16, Martin Tegtmeier wrote:
> currently dropping the file system cache ("echo 1 >
> /proc/sys/vm/drop_caches") doesn't leave any trace. However dropping
> the fs cache can severely impact system behaviour and application
> response times. Therefore I suggest to write a syslog entry if the
> entire inode page cache is scrapped.
> Since it is not an easy task to calculate the size of the droppable
> filesystem cache I also suggest to add the number of dropped pages to
> the syslog entry. This can be accomplished by saving the return value
> of invalidate_mapping_pages().
> 
> The number of dropped pages is an important measure for capacity
> planning. For the deployment of new SAP application instances we
> would like to know the amount of memory that was freed from
> fs caches.

I'm hugely in favour of this, because we don't need syslog to be an
auditing mechanism for userspace.

However, if something like this goes into the kernel it would probably
make a lot more sense to print out the values on a per-filesystem
basis, if any pages are dropped from a particular filesystem, so one
can see which filesystem was using the most cache.

> drop_caches.c |    9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
> 
> 
> commit b2219e84647bdf64fd6e7f9c5260c1e6bed24d58
> Author: Martin Tegtmeier <martin.tegtmeier@gmail.com>
> Date:   Tue May 24 15:24:20 2011 +0200
> 
>    drop_caches: add syslog entry
> 
>    Dropping the entire file system cache (inode cache) can severely
> influence system behaviour
>    yet currently dropping the file system cache is NOT traceable.
>    This patch adds an entry to /var/log/messages with a time stamp
> and the number of dropped pages.
> 
> 
>    Signed-off-by: Martin Tegtmeier <martin.tegtmeier@gmail.com>
> 
> diff --git a/fs/drop_caches.c b/fs/drop_caches.c
> index 98b77c8..f2e4dc4 100644
> --- a/fs/drop_caches.c
> +++ b/fs/drop_caches.c
> @@ -12,6 +12,7 @@
> 
> /* A global variable is a bit ugly, but it keeps the code simple */
> int sysctl_drop_caches;
> +unsigned long pages_dropped;
> 
> static void drop_pagecache_sb(struct super_block *sb, void *unused)
> {
> @@ -28,7 +29,7 @@ static void drop_pagecache_sb(struct super_block
> *sb, void *unused)
> 		__iget(inode);
> 		spin_unlock(&inode->i_lock);
> 		spin_unlock(&inode_sb_list_lock);
> -		invalidate_mapping_pages(inode->i_mapping, 0, -1);
> +		pages_dropped += invalidate_mapping_pages(inode->i_mapping, 0, -1);
> 		iput(toput_inode);
> 		toput_inode = inode;
> 		spin_lock(&inode_sb_list_lock);
> @@ -55,8 +56,12 @@ int drop_caches_sysctl_handler(ctl_table *table, int write,
> 	if (ret)
> 		return ret;
> 	if (write) {
> -		if (sysctl_drop_caches & 1)
> +		if (sysctl_drop_caches & 1) {
> +			pages_dropped = 0;
> 			iterate_supers(drop_pagecache_sb, NULL);
> +			printk(KERN_INFO "drop_caches: %lu pages dropped from inode cache\n",
> +				pages_dropped);
> +		}
> 		if (sysctl_drop_caches & 2)
> 			drop_slab();
> 	}
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


Cheers, Andreas





--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>