[07/23] xfs_scrub: use datadev parallelization estimates for thread count

Commit Message

Darrick J. Wong March 1, 2019, 11:27 p.m. UTC

From: Darrick J. Wong <[e-mail]>

During phases 2-5, xfs_scrub should estimate the level of
parallelization possible on the data device to determine the number of
threads spawned to scrub filesystem metadata, not just blindly using the
number of CPUs.  This avoids flooding non-rotational storage with random
reads, which totally destroys performance and makes scrub runtimes
higher.

Signed-off-by: Darrick J. Wong <[e-mail]>
---
 scrub/phase1.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

Patch

diff --git a/scrub/phase1.c b/scrub/phase1.c
index 2113014b..6b472147 100644
--- a/scrub/phase1.c
+++ b/scrub/phase1.c
@@ -109,13 +109,6 @@  _("Must be root to run scrub."));
 		return false;
 	}
 
-	ctx->nr_io_threads = nproc;
-	if (verbose) {
-		fprintf(stdout, _("%s: using %d threads to scrub.\n"),
-				ctx->mntpoint, scrub_nproc(ctx));
-		fflush(stdout);
-	}
-
 	if (!platform_test_xfs_fd(ctx->mnt_fd)) {
 		str_info(ctx, ctx->mntpoint,
 _("Does not appear to be an XFS filesystem!"));
@@ -193,6 +186,13 @@  _("Unable to find realtime device path."));
 		return false;
 	}
 
+	ctx->nr_io_threads = disk_heads(ctx->datadev);
+	if (verbose) {
+		fprintf(stdout, _("%s: using %d threads to scrub.\n"),
+				ctx->mntpoint, scrub_nproc(ctx));
+		fflush(stdout);
+	}
+
 	if (ctx->fsinfo.fs_log) {
 		ctx->logdev = disk_open(ctx->fsinfo.fs_log);
 		if (error) {