From: Brian Foster <bfoster@redhat.com>
To: Dave Chinner <david@fromorbit.com>
Cc: xfs@oss.sgi.com
Subject: Re: [PATCH 5/9] repair: parallelise uncertin inode processing in phase 3
Date: Mon, 4 Jan 2016 14:12:14 -0500 [thread overview]
Message-ID: <20160104191213.GE19852@bfoster.bfoster> (raw)
In-Reply-To: <1450733829-9319-6-git-send-email-david@fromorbit.com>
On Tue, Dec 22, 2015 at 08:37:05AM +1100, Dave Chinner wrote:
> From: Dave Chinner <dchinner@redhat.com>
>
> This can take a long time when there are millions of uncertain inodes in badly
> broken filesystems. THe processing is per-ag, the data structures are all
> per-ag, and the load is mostly CPU time spent checking CRCs on each
> uncertaini inode. Parallelising reduced the runtime of this phase on a badly
> broken filesytem from ~30 minutes to under 5 miniutes.
>
> Signed-off-by: Dave Chinner <dchinner@redhat.com>
> ---
This one seems a bit more scary simply because the amount of work
involved in phase 3, such as inode processing and whatnot. I don't think
that everything in there is necessarily AG local as the commit log
description above implies. On a skim through, I do see that we have
ag_locks[agno].lock for cases that can cross AG boundaries, such as
checking block allocation state (get_bmap()/set_bmap()) of bmapbt blocks
(iiuc?), for example.
So I can't really spot any actual problems, but there's a lot of code
down in there. I'm fine with it as long as testing bears out and this
gets a reasonable amount of soak time such that we can hopefully catch
any serious issues or areas currently lacking sufficient locking:
Reviewed-by: Brian Foster <bfoster@redhat.com>
> repair/phase3.c | 59 ++++++++++++++++++++++++++++++++++++++++++++---------
> repair/protos.h | 2 +-
> repair/xfs_repair.c | 2 +-
> 3 files changed, 51 insertions(+), 12 deletions(-)
>
> diff --git a/repair/phase3.c b/repair/phase3.c
> index 76c9440..0890a27 100644
> --- a/repair/phase3.c
> +++ b/repair/phase3.c
> @@ -28,6 +28,7 @@
> #include "dinode.h"
> #include "progress.h"
> #include "bmap.h"
> +#include "threads.h"
>
> static void
> process_agi_unlinked(
> @@ -87,10 +88,33 @@ process_ags(
> do_inode_prefetch(mp, ag_stride, process_ag_func, false, false);
> }
>
> +static void
> +do_uncertain_aginodes(
> + work_queue_t *wq,
> + xfs_agnumber_t agno,
> + void *arg)
> +{
> + int *count = arg;
> +
> + *count = process_uncertain_aginodes(wq->mp, agno);
> +
> +#ifdef XR_INODE_TRACE
> + fprintf(stderr,
> + "\t\t phase 3 - ag %d process_uncertain_inodes returns %d\n",
> + *count, j);
> +#endif
> +
> + PROG_RPT_INC(prog_rpt_done[agno], 1);
> +}
> +
> void
> -phase3(xfs_mount_t *mp)
> +phase3(
> + struct xfs_mount *mp,
> + int scan_threads)
> {
> - int i, j;
> + int i, j;
> + int *counts;
> + work_queue_t wq;
>
> do_log(_("Phase 3 - for each AG...\n"));
> if (!no_modify)
> @@ -129,20 +153,35 @@ phase3(xfs_mount_t *mp)
> */
> do_log(_(" - process newly discovered inodes...\n"));
> set_progress_msg(PROG_FMT_NEW_INODES, (__uint64_t) glob_agcount);
> +
> + counts = calloc(sizeof(*counts), mp->m_sb.sb_agcount);
> + if (!counts) {
> + do_abort(_("no memory for uncertain inode counts\n"));
> + return;
> + }
> +
> do {
> /*
> * have to loop until no ag has any uncertain
> * inodes
> */
> j = 0;
> - for (i = 0; i < mp->m_sb.sb_agcount; i++) {
> - j += process_uncertain_aginodes(mp, i);
> -#ifdef XR_INODE_TRACE
> - fprintf(stderr,
> - "\t\t phase 3 - process_uncertain_inodes returns %d\n", j);
> -#endif
> - PROG_RPT_INC(prog_rpt_done[i], 1);
> - }
> + memset(counts, 0, mp->m_sb.sb_agcount * sizeof(*counts));
> +
> + create_work_queue(&wq, mp, scan_threads);
> +
> + for (i = 0; i < mp->m_sb.sb_agcount; i++)
> + queue_work(&wq, do_uncertain_aginodes, i, &counts[i]);
> +
> + destroy_work_queue(&wq);
> +
> + /* tally up the counts */
> + for (i = 0; i < mp->m_sb.sb_agcount; i++)
> + j += counts[i];
> +
> } while (j != 0);
> +
> + free(counts);
> +
> print_final_rpt();
> }
> diff --git a/repair/protos.h b/repair/protos.h
> index b113aca..0290420 100644
> --- a/repair/protos.h
> +++ b/repair/protos.h
> @@ -46,7 +46,7 @@ void thread_init(void);
>
> void phase1(struct xfs_mount *);
> void phase2(struct xfs_mount *, int);
> -void phase3(struct xfs_mount *);
> +void phase3(struct xfs_mount *, int);
> void phase4(struct xfs_mount *);
> void phase5(struct xfs_mount *);
> void phase6(struct xfs_mount *);
> diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> index fcdb212..5d5f3aa 100644
> --- a/repair/xfs_repair.c
> +++ b/repair/xfs_repair.c
> @@ -871,7 +871,7 @@ main(int argc, char **argv)
> if (do_prefetch)
> init_prefetch(mp);
>
> - phase3(mp);
> + phase3(mp, phase2_threads);
> timestamp(PHASE_END, 3, NULL);
>
> phase4(mp);
> --
> 2.5.0
>
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs
next prev parent reply other threads:[~2016-01-04 19:12 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-12-21 21:37 [PATCH 0/9] xfsprogs: big, broken filesystems cause pain Dave Chinner
2015-12-21 21:37 ` [PATCH 1/9] metadump: clean up btree block region zeroing Dave Chinner
2016-01-04 19:11 ` Brian Foster
2015-12-21 21:37 ` [PATCH 2/9] metadump: bounds check btree block regions being zeroed Dave Chinner
2016-01-04 19:11 ` Brian Foster
2015-12-21 21:37 ` [PATCH 3/9] xfs_mdrestore: correctly account bytes read Dave Chinner
2016-01-04 19:12 ` Brian Foster
2015-12-21 21:37 ` [PATCH 4/9] repair: parallelise phase 7 Dave Chinner
2016-01-04 19:12 ` Brian Foster
2015-12-21 21:37 ` [PATCH 5/9] repair: parallelise uncertin inode processing in phase 3 Dave Chinner
2016-01-04 19:12 ` Brian Foster [this message]
2015-12-21 21:37 ` [PATCH 6/9] libxfs: directory node splitting does not have an extra block Dave Chinner
2016-01-05 18:34 ` Brian Foster
2016-01-05 22:07 ` Dave Chinner
2015-12-21 21:37 ` [PATCH 7/9] libxfs: don't discard dirty buffers Dave Chinner
2016-01-05 18:34 ` Brian Foster
2015-12-21 21:37 ` [PATCH 8/9] libxfs: don't repeatedly shake unwritable buffers Dave Chinner
2016-01-05 18:34 ` Brian Foster
2015-12-21 21:37 ` [PATCH 9/9] libxfs: keep unflushable buffers off the cache MRUs Dave Chinner
2016-01-05 18:34 ` Brian Foster
2016-01-05 23:58 ` Dave Chinner
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=20160104191213.GE19852@bfoster.bfoster \
--to=bfoster@redhat.com \
--cc=david@fromorbit.com \
--cc=xfs@oss.sgi.com \
/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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.