On 10/13/2016 01:20 AM, Josef Bacik wrote:
On 10/12/2016 05:03 AM, Wang Xiaoguang wrote:
Since commit b02441999efcc6152b87cd58e7970bb7843f76cf, we don't wait all
ordered extents, but I run into some enospc errors when doing large file
create and delete tests, it's because shrink_delalloc() does not write
enough delalloc bytes and wait them finished:
    From: Miao Xie <mi...@cn.fujitsu.com>
    Date: Mon, 4 Nov 2013 23:13:25 +0800
Subject: [PATCH] Btrfs: don't wait for the completion of all the ordered extents

It is very likely that there are lots of ordered extents in the filesytem, if we wait for the completion of all of them when we want to reclaim some space for the metadata space reservation, we would be blocked for a long
    time. The performance would drop down suddenly for a long time.

Here we introduce a simple reclaim_priority variable, the higher the
value, the higher the priority, 0 is the minimum priority. The core
idea is:
        if (reclaim_priority >= 3)
to_reclaim = percpu_counter_sum_positive(&root->fs_info->delalloc_bytes);
                to_reclaim = orig * (2 << (reclaim_priority + 1));
As the priority increases, we try wo write more delalloc bytes, here orig
is the number of metadata we want to reserve. Meanwhile if
"reclaim_priority >= 3" returns true, we'll also wait all current ordered
extents to finish.

Ok this is closer to what I want but not quite there. I want to get rid of the magic numbers, so we start with


and once priority == 0 we know it's time to flush and wait for all the things. Then we use the priority as a multiplier, with DEFAULT being no multiplier, so it would look something like this

if (reclaim_priority)
    to_reclaim = orig * (2 << BTRFS_DEFAULT_FLUSH_PRIORITY - priority);
    to_reclaim = delalloc_bytes;
if (reclaim_priority)
    items_to_wait = calc_reclaim_items_nr(root, to_reclaim);
    items_to_wait = -1;

and then instead of

if (reclaim_priority > 3) {

we can change the while loop to be

while ((flush_state <= COMMIT_TRANS) || (--reclaim_priority >= 0));

and move the wake_all_tickets() to outside the loop, and add a

if (flush_state > COMMIT_TRANS)
    flush_state = FLUSH_DELAYED_ITEMS_NR;

to the start of the do loop.  Does that make sense?  Thanks,
Yes, thanks for your kindly help.
For "wake_all_tickets()", I think we still need to put it in the while loop,
If we move it outside, we need to relock btrfs_space_info's lock, but
here there maybe some race window, some new tickets maybe added in,
we should not wake up them directly and should also do some flush work
for them.

Xiaoguang Wang


To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to