From: Amir Goldstein <amir73il@gmail.com>
To: Jan Kara <jack@suse.cz>
Cc: linux-fsdevel <linux-fsdevel@vger.kernel.org>
Subject: Re: [PATCH 5/7] fanotify: limit number of event merge attempts
Date: Sat, 27 Feb 2021 10:31:52 +0200 [thread overview]
Message-ID: <CAOQ4uxiqnD7Qr=__apodWYfQYQ_JOvVnaZsi4jjGQmJ9S5hMyA@mail.gmail.com> (raw)
In-Reply-To: <20210202162010.305971-6-amir73il@gmail.com>
On Tue, Feb 2, 2021 at 6:20 PM Amir Goldstein <amir73il@gmail.com> wrote:
>
> Event merges are expensive when event queue size is large.
> Limit the linear search to 128 merge tests.
> In combination with 128 hash lists, there is a potential to
> merge with up to 16K events in the hashed queue.
>
> Signed-off-by: Amir Goldstein <amir73il@gmail.com>
> ---
> fs/notify/fanotify/fanotify.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/fs/notify/fanotify/fanotify.c b/fs/notify/fanotify/fanotify.c
> index 12df6957e4d8..6d3807012851 100644
> --- a/fs/notify/fanotify/fanotify.c
> +++ b/fs/notify/fanotify/fanotify.c
> @@ -129,11 +129,15 @@ static bool fanotify_should_merge(struct fsnotify_event *old_fsn,
> return false;
> }
>
> +/* Limit event merges to limit CPU overhead per event */
> +#define FANOTIFY_MAX_MERGE_EVENTS 128
> +
> /* and the list better be locked by something too! */
> static int fanotify_merge(struct list_head *list, struct fsnotify_event *event)
> {
> struct fsnotify_event *test_event;
> struct fanotify_event *new;
> + int i = 0;
>
> pr_debug("%s: list=%p event=%p\n", __func__, list, event);
> new = FANOTIFY_E(event);
> @@ -147,6 +151,8 @@ static int fanotify_merge(struct list_head *list, struct fsnotify_event *event)
> return 0;
>
> list_for_each_entry_reverse(test_event, list, list) {
> + if (++i > FANOTIFY_MAX_MERGE_EVENTS)
> + break;
> if (fanotify_should_merge(test_event, event)) {
> FANOTIFY_E(test_event)->mask |= new->mask;
> return 1;
> --
> 2.25.1
>
Jan,
I was thinking that this patch or a variant thereof should be applied to stable
kernels, but not the entire series.
OTOH, I am concerned about regressing existing workloads that depend on
merging events on more than 128 inodes.
I thought of this compromise between performance and functional regressions:
/*
* Limit event merges to limit CPU overhead per new event.
* For legacy mode, avoid unlimited CPU overhead, but do not regress the event
* merge ratio in heavy concurrent workloads with default queue size.
* For new FAN_REPORT_FID modes, make sure that CPU overhead is low.
*/
#define FANOTIFY_MAX_MERGE_OLD_EVENTS 16384
#define FANOTIFY_MAX_MERGE_FID_EVENTS 128
static inline int fanotify_max_merge_events(struct fsnotify_group *group)
{
if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS))
return FANOTIFY_MAX_MERGE_FID_EVENTS;
else
return FANOTIFY_MAX_MERGE_OLD_EVENTS;
}
I can start the series with this patch and change that to:
#define FANOTIFY_MAX_MERGE_FID_EVENTS 128
static inline int fanotify_max_merge_events(struct fsnotify_group *group)
{
return FANOTIFY_MAX_MERGE_EVENTS;
}
At the end of the series.
What do you think?
Thanks,
Amir.
next prev parent reply other threads:[~2021-02-27 8:33 UTC|newest]
Thread overview: 36+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-02-02 16:20 [PATCH 0/7] Performance improvement for fanotify merge Amir Goldstein
2021-02-02 16:20 ` [PATCH 1/7] fsnotify: allow fsnotify_{peek,remove}_first_event with empty queue Amir Goldstein
2021-02-02 16:20 ` [PATCH 2/7] fsnotify: support hashed notification queue Amir Goldstein
2021-02-16 15:02 ` Jan Kara
2021-02-17 12:33 ` Amir Goldstein
2021-02-17 13:48 ` Jan Kara
2021-02-17 15:42 ` Amir Goldstein
2021-02-17 16:49 ` Jan Kara
2021-02-18 10:52 ` Amir Goldstein
2021-02-02 16:20 ` [PATCH 3/7] fsnotify: read events from hashed notification queue by order of insertion Amir Goldstein
2021-02-16 15:10 ` Jan Kara
2021-02-02 16:20 ` [PATCH 4/7] fanotify: enable hashed notification queue for FAN_CLASS_NOTIF groups Amir Goldstein
2021-02-02 16:20 ` [PATCH 5/7] fanotify: limit number of event merge attempts Amir Goldstein
2021-02-27 8:31 ` Amir Goldstein [this message]
2021-03-01 13:08 ` Jan Kara
2021-03-01 13:58 ` Amir Goldstein
2021-09-15 12:39 ` Amir Goldstein
2021-09-15 16:33 ` Jan Kara
2021-02-02 16:20 ` [PATCH 6/7] fanotify: mix event info into merge key hash Amir Goldstein
2021-02-16 15:39 ` Jan Kara
2021-02-17 10:13 ` Amir Goldstein
2021-02-18 10:46 ` Amir Goldstein
2021-02-18 11:11 ` Jan Kara
2021-02-18 12:17 ` Amir Goldstein
2021-02-02 16:20 ` [PATCH 7/7] fsnotify: print some debug stats on hashed queue overflow Amir Goldstein
2021-02-16 16:02 ` [PATCH 0/7] Performance improvement for fanotify merge Jan Kara
2021-02-17 10:52 ` Amir Goldstein
2021-02-17 11:25 ` Jan Kara
2021-02-18 10:56 ` Amir Goldstein
2021-02-18 11:15 ` Jan Kara
2021-02-18 12:35 ` Amir Goldstein
2021-02-19 10:15 ` Jan Kara
2021-02-19 10:21 ` Jan Kara
2021-02-19 13:38 ` Amir Goldstein
2021-02-21 12:53 ` Amir Goldstein
2021-02-22 9:29 ` Jan Kara
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAOQ4uxiqnD7Qr=__apodWYfQYQ_JOvVnaZsi4jjGQmJ9S5hMyA@mail.gmail.com' \
--to=amir73il@gmail.com \
--cc=jack@suse.cz \
--cc=linux-fsdevel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).